Geforce gts 250 tests in games. Video cards

It's similar in Crysis, but as resolution and graphics quality go up, the extra 512MB starts to play a major role in performance, which can be as much as 15% faster than older solutions. It remains only to compare with the GeForce GTS 250 with different amounts of video memory, which we will do in the near future.

conclusions

It would seem that the epic with the renaming of video cards based on GPU G92 was supposed to end last year. But no, the chip turned out so successful that it can be used for at least another year or two. And we will soon be able to please accelerators such as GeForce GT 350, GT 420 or even GT 510. And why not?

AMD, a direct competitor in the graphics market, is still bringing down the price, and in this case it is possible to release new generations of video cards based on old designs, reducing their cost and positioning products in lower and lower market segments. The defunct 3dfx behaved in a similar way due to a miscalculation in the company's management until it was acquired by NVIDIA. Will this be the fate of the Californians? It's hard to say.

The company has a huge potential of engineers and programmers and will be able to fight back without any problems either by lowering prices for high-performance solutions (read, hi-end-class cards), or by developing and releasing a new product on an urgent basis. But the competitor is still sleeping, moreover, in a lethargic dream, otherwise how can one explain the appearance of the G92 chip in the third generation of the GeForce series video card?

As for the so-called novelty, the GeForce GTS 250 video cards in the reference version have become smaller in size and now more closely match GeForce cards 8800 GTS 512MB; officially acquired 1 GB of video memory and can even be equipped with double the amount. Due to the use of the 55nm G92b chip, the power supply subsystem has been somewhat simplified compared to the GeForce 9800 GTX(+). Also, manufacturers can safely produce cards on a redesigned PCB design with original system cooling. But the main advantage of the GeForce GTS 250 is the recommended price for video cards - about $ 130 for a 512-MB version, which will seriously compete with both NVIDIA's own G92-based solutions and AMD products.

Regarding the Zotac GeForce GTS250 AMP! note the increased operating frequencies, one gigabyte of video memory, which will allow you to play more comfortably with high resolutions, and, of course, the package, which includes all the necessary accessories and additional bonuses in the form of the XIII Century game and the 3DMark Vantage test package.

The Palit GF9800GTX+ 512M DDR3 card is a solid mid-range card with a Spartan package, which, nevertheless, is adapted to connect to a modern cinema or monitor with an HDMI interface, and you can connect as many as two devices. For owners of inexpensive TFT-monitors with an analog connector, D-Sub is already provided on the accelerator, which will eliminate the need for extra adapters. But the lack of a conventional TV-out is somewhat disappointing, given the prevalence of standard TVs in our area.

>> <<

At the very beginning of spring, Nvidia suddenly decided to please us with its next announcement. Naturally, the name of the presented model was brought into line with the new principle, and the announced video card was named Geforce GTS 250. GTS in this case means a lower level of performance compared to GTX, and 250 is the model number in the line.

According to the characteristics of the video card, it becomes clear that the entire article could fit in this one picture, or even the line: "Geforce GTS 250 = Geforce 9800 GTX +". Well, or “Geforce GTS 250 = Geforce 9800 GTX + + 1GB”, although this clarification does not change anything much. But, being exemplary journalists who need to tell readers about the "novelty", we have made an almost full-fledged article. “Almost” is because without the usual theoretical data that have not changed for a long time, and without synthetic tests, which simply do not make sense in this case.

So, GeForce GTS 250 is an "upgrade" (as Nvidia calls it) of the popular Geforce 9800 GTX+ model, which competes with the RADEON HD 4850. Let's trace the history of all such "updates". The first model based on the full-fledged G92 chip, with all stream processors included, was the Geforce 8800 GTS 512MB, released in late 2007 for a suggested price of $349-399. Even then, we wondered why it was not given its own name, and confused buyers with the multifaceted 8800 series.

Then the GeForce 9800 GTX appeared - a bit faster, but on the whole almost the same card, but with a new PCB design. It is the turn to be surprised at the fact why, due to a very small increase in frequencies, they changed quite a decent design? The 9800 GTX was announced in February 2008, with a suggested price of $299-$349. Further, in the summer of the same last year, it was again "updated" with a small overclock, and became known as Geforce 9800 GTX+. It was already in June 2008 when the price of the solution dropped to $229.

Now, the same card based on the G92 chip (albeit already manufactured using 55 nm technology), but with a simplified PCB design, more like the good old design of the 8800 GTS 512, has been "updated" once again, and has received the name Geforce GTS 250. Although it is not so important for potential buyers what it is called, in contrast to the fact that for a 512-MB model they will now ask for only $129, and a board with a gigabyte of local video memory will cost $149. However, the indicated prices are recommended for the North American market, our retailers habitually skim the cream for a couple of months.

In general, despite all our taunts, the release of the Geforce GTS 250 model can be called quite timely for several reasons. The first has already pretty much set the teeth on edge in all the media and in the mouths of friends and relatives - the world economic situation. At such a time, people try not to buy what is not necessary, or simply moderate their appetites. Which is what the GTS 250 will be able to do, offering a very good level of performance, sufficient for all multi-platform projects and most PC-exclusive games, and for only $130-150.

The second reason is the rather early spring release of AMD's competing solutions designed for the lower price range (less than $100), as well as the release of the next "updated" product from Nvidia for the same sector, and the corresponding price cuts for more expensive models. By the way, the competitor has already supported the price reduction proposed by Nvidia with the release of the GTS 250, also reducing the prices of the competing RADEON HD 4850 and HD 4870.

How is Nvidia's solution different, what can they offer customers, except for the third easy upgrade of the same product (8800 GTS 512 - 9800 GTX - 9800 GTX + - GTS 250)? Representatives of the company highlight the following (in comparison with a competitor, and not with "old" cards on the G92, of course): support for Nvidia CUDA and Nvidia PhysX technologies, as well as Nvidia 3D Vision, and, well, some "readiness" for the release of the new Microsoft Windows operating system 7. Next, we will dwell on these features in a little more detail, since there really is nothing to write about the new video card.

As you understand, the theoretical part of the Geforce GTS 250 review will be short, we have already considered the G92 architecture, which is based on the G80 architecture, and nothing has changed in the G92b, manufactured according to 55 nm technological standards. Compared to the GeForce 9800 GTX(+), the only significant differences for the user can be considered only lower power requirements and slightly reduced heat dissipation. Well, let's not forget about the price.

If for some strange reason you are not yet familiar with modern Nvidia architectures, then you can read all the details about the G92 in the basic review on our website. This is a further development of the G8x architecture with some improvements. Also, before reading this material, we recommend that you carefully read the basic theoretical materials DX Current, DX Next and Longhorn, which describe various aspects of modern hardware graphics accelerators and the architectural features of previous Nvidia and AMD products.

These materials fairly accurately predicted the current situation with video chip architectures, and many assumptions about future solutions came true. Additional information about the unified architectures of Nvidia G8x/G9x using the example of previous solutions can be found in the following articles:

Let's assume that most readers have been familiar with the G9x architecture for a long time, and consider the detailed characteristics of the first model of the Geforce GTS 200 series video card based on the G92 brand GPU, made using the 55 nm process technology.

Graphic accelerator Geforce GTS 250

Chip codename G92
55 nm production technology
754 million transistors
Unified architecture with an array of common processors for vertex and pixel streaming, and other kinds of data
Hardware support for DirectX 10, including shader model - Shader Model 4.0, geometry generation and recording intermediate data from shaders (stream output)
256-bit memory bus, four independent 64-bit wide controllers
Core clock 738 MHz
More than double the 1836 MHz ALU frequency
128 scalar floating point ALUs (integer and float formats, FP32 support for IEEE 754 precision, two MAD+MUL operations per clock)
64 texture addressing and filtering units with support for FP16 and FP32 components in textures
Possibility of dynamic branching in pixel and vertex shaders
4 wide ROPs (16 pixels) with support for anti-aliasing modes up to 16 samples per pixel, including with FP16 or FP32 framebuffer format. Each block consists of an array of flexibly configurable ALUs and is responsible for generating and comparing Z, MSAA, blending. Peak performance of the entire subsystem up to 64 MSAA samples (+ 64 Z) per clock, in colorless mode (Z only) - 128 samples per clock
Write results to 8 frame buffers simultaneously (MRT)
All interfaces (two RAMDAC, two Dual Link DVI, HDMI, DisplayPort, HDTV) are integrated on the chip

Geforce GTS 250 reference graphics card specifications

Core clock 738 MHz
Frequency of universal processors 1836 MHz
Number of universal processors 128
Number of texture units - 64, blending units - 16
Effective memory frequency 2200 (2*1100) MHz
Memory type GDDR3
Memory size 512/1024/2048 megabytes
Memory bandwidth 70.4 GB/s
Theoretical maximum fill rate is 11.8 gigapixels per second.
Theoretical texture sampling rate up to 47.2 gigatexels per second.
Two DVI-I Dual Link connectors, supports output at resolutions up to 2560x1600
Dual SLI connector
PCI Express 2.0 bus
TV-Out, HDTV-Out, HDCP support, HDMI, DisplayPort
Power consumption up to 150 W (one 6-pin connector)
Dual slot design
MSRP $129/$149/$169

In general, nothing interesting can be seen, the "new" video card based on the 55 nm G92 chip does not differ from the GeForce 9800 GTX+ in any way. However, the release of the new model can be partially justified by installing not 512 MB of video memory on it, like in the 9800 GTX+, but GB, which greatly affects performance in heavy modes with maximum quality settings, high resolutions with full-screen anti-aliasing enabled. And there are also two-gigabyte options, but this is already more of a marketing advantage than a real one.

Under such conditions, the older versions of the Geforce GTS 250 should really be noticeably faster than the Geforce 9800 GTX+ due to the increased memory. And some of the more recent games will benefit even at lower resolutions. Everything would be fine, but some card manufacturers released Geforce 9800 GTX+ with a gigabyte of memory even earlier...

The production of G92b video chips according to 55 nm technological standards and a noticeable simplification of the PCB design allowed Nvidia to make a solution similar to the Geforce 9800 GTX in terms of characteristics, but with a lower price and reduced power consumption and heat dissipation. And now, in order to provide the Geforce GTS 250 with power, only one 6-pin PCI-E power connector is installed on the board.

We already wrote about the name of the announced model at the very beginning. In this case, Nvidia simply changed the naming of its G92-based cards to the accepted modern order. After the general name Geforce comes the series index (GTX, GTS, GT, G), and everything ends with the model number, where the first digit indicates the generation (it doesn’t mean much, however, because the 200 series includes cards based on both GT200 and G92), and the other two are the position of the model inside the ruler.

Architecture and features of the solution

We can't tell readers anything new about the architecture of the chip on which the Geforce GTS 250 is based, because the G92 chip is the same GPU we've known since autumn 2007, just a smaller area and requiring less power to operate. The use of an improved architecture based on the 55 nm process technology made the Nvidia Geforce GTS 250 cheaper, quieter and more power efficient than the Geforce 9800 GTX+.

In general, it is very interesting what the chip design departments are doing at Nvidia. In fact, for almost three years we have not seen anything really new, only small improvements (well, there were quite a lot of changes in the GT200, but it's still just an improved G8x architecture), and even more often updates in the form of changed frequencies, design and name.

Recall that the GT200 architecture, announced last summer, is a modified G8x/G9x architecture, which was initiated by the G80 chip released back in 2006. The main differences between the G92 chip and the G80 were the 65 nm manufacturing technology, the hardware video decoding unit, and some improvements in texture modules, while the innovations of the GT200 are mostly quantitative. Therefore, we continue to wait for a really new architecture (with DirectX 11 API support) from Nvidia, and we really hope that no financial difficulties will affect these plans.

NVIDIA CUDA

Like all other modern solutions of the company, the Geforce GTS 250 model supports the technology of computing on video chips - CUDA. Two of our materials are devoted to this technology, so we will not dwell on its description in detail, everything can be read there.

It is important that programs using CUDA, intended for ordinary users, began to appear little by little. One of the first such applications to use CUDA is Elemental's video transcoding software, oddly named Badaboom. But not a single badabum ... Now we can already mention other useful ones (meaning the visible benefit here and now, therefore [email protected] and [email protected] skip) applications with CUDA support: Pegasys TMPGEnc 4 - GPU-accelerated video post-processing, ArcSoft TotalMedia Theater - GPU-accelerated video stream scaling to HD resolutions.

Pegasys TMPGEnc 4 (formerly known as Tsunami MPEG Encoder) is a multi-format video encoding utility that also offers some stream editing and post-processing filters. Recently, TMPGEnc has been using CUDA technology to speed up de video encoding (GPU acceleration of video encoding is not yet supported), as well as to increase the performance of filters such as color correction, noise removal and sharpening (smart sharpen).

The listed tasks are very demanding on computing power, and the processing of a video stream when using them takes much more time. When using Nvidia graphics cards that support CUDA technology, filters can be accelerated up to 4-5 times compared to similar processing on the CPU. Maybe for the time being this is not so impressive, but it is still able to make the work easier for some users.

ArcSoft TotalMedia Theater is an application for viewing videos of various formats on a computer that supports Nvidia CUDA technology for very high-quality scaling of video stream from DVD and other low-resolution formats to HD resolutions (1280x720, 1920x1080). The upscaling technology of ArcSoft SimHD in its resolution upscaling algorithm shifts part of the calculations to the GPU using CUDA, which provides the possibility of such processing in real time, at a speed of 30 frames per second. For comparison, without CUDA acceleration, conventional CPUs provide performance in the same task only in units of FPS.

Do not forget that this is all just a theory, and we have plans for a practical study of all such capabilities of both Nvidia and AMD chips. If only there were more applications by then, especially those where you can compare competing solutions head-to-head...

Nvidia PhysX

And yet, for a gaming video card, all of the above is not so important yet, but improvements in games are a completely different matter. We have repeatedly written about the technology of hardware acceleration of physical effects using the GPU. PhysX allows you to use such computationally complex effects as the simulation of fluids and tissues, as well as calculate physics for many thousands of objects at the same time.

In this case, the support of the gaming industry is very important, without which any good undertaking will end as soon as it starts. And PhysX has such support. Last year, three big game publishers: Electronic Arts, Take-Two Interactive and THQ announced the use of PhysX technology in the development of games by their game studios. Given the appearance of games with active use of physical PhysX effects, hardware-accelerated on Nvidia video cards, already last year, we can say with confidence that in 2009 the number of such games will only increase.

A good example is the games Anabiosis and Mirror's Edge. Although PhysX technology does not affect the gameplay of these games so much that its support cannot be abandoned, it brings players a new experience, increasing the interactivity of the game world, adding high-quality simulation effects fabrics, glass, smoke and liquids Games with enabled PhysX effects are more interactive and physically correct behavior of objects.

Naturally, so far all gaming applications are created taking into account systems that do not accelerate PhysX in hardware. There are a lot of such systems, and in their case, the entire burden of calculations falls on the CPU, which cannot provide high performance in this task. For example, at some levels of the game Mirror's Edge, systems without hardware support for PhysX technology cannot even provide 30 frames per second, while the Geforce GTS 250 announced today provides comfortable gameplay in this game even at maximum settings in resolutions up to 2560 × 1600 inclusive .

Geforce 3D Vision

Geforce 3D Vision has become one of the company's latest technologies, supported by Geforce GTS 250 as well. This is a turnkey stereo viewing solution using wireless shutter glasses. The shutter principle provides better image quality than passive glasses based on polarizing filters. This applies to both increased resolution and b O greater viewing angles.

Another important advantage of Geforce 3D Vision technology against the competition is excellent compatibility with 3D games. The Nvidia stereo driver provides native stereo rendering support for over 300 games. In addition to games, 3D video players like 3dtv Stereoscopic Player are supported, as well as viewing stereoscopic photos, for which a special utility is included in the package.

It is important that Nvidia stereo drivers are based on the same code as regular drivers and do not use performance-degrading wrappers. Support for stereo in games is made similar to SLI profiles, with ready-made settings for each specific game. Nvidia employees research gaming applications and find the optimal settings for each game, so the user is not required to select them himself.

However, we see the main problem hindering the implementation of Geforce 3D Vision not even an additional $200 for glasses, and not doubled requirements for the performance of the video subsystem (after all, frames need to be calculated for each eye), but the almost complete absence of compatible display devices on our market. with this technology. After all, this requires not just monitors with 120 Hz panels, but devices that receive a DVI signal at a frequency of 120 Hz.

Of all the monitors, only two models so far have such capabilities: SAMSUNG SyncMaster 2233RZ and ViewSonic VX2265wm, which are not even sold here yet. Although many modern LCD TVs support 100 or 120 Hz, this only applies to the screen output frequency, and they cannot perceive a 120 Hz signal via HDMI, and therefore they cannot be used to output a stereo image. Yes, and CRT monitors with a frequency of 100 Hz or more can be used, but it is imperative that they have a two-channel DVI input, which is also a rarity in reality. It is also possible to work on Mitsubishi 1080p DLP devices and DepthQ HD 3D projectors, but this is also unlikely to seriously increase the circle of potential Geforce 3D Vision buyers.

And yet, despite some difficulties, in general, stereo imaging technology is very interesting, and in the near future we will try to tell you more about it if possible. So far, the technology has an unpleasant feature that can be considered the main drawback - strict (if not the most severe) requirements for the output device. As related monitors and TVs enter the market, the situation should improve significantly. We wait.

In the conclusions of the theoretical part of the article, we can note that the Nvidia Geforce GTS 250, although it is an almost complete copy of the Geforce 9800 GTX+ in terms of technical characteristics, is at the same time an excellent market offer with higher performance compared to the previous 9800 GTX+ model. It seems that the GTS 250 should become successful in the video card market, making a very good competition for the RADEON HD 4850 from AMD.

However, all performance issues in modern 3D games are discussed in detail in the next part of our article. It is she who shows who is who among the current solutions, and also dots the i's in matters of relative rendering speed, and hence the justification for the release of an updated solution to the market.

GTS 250 is the latest video chip model based on the G92 platform. This architecture originates back in 2007, and only after the release of the monitored video card did it end its life.

The first solution on the G92 was the 8800 series video card with 512MB of memory. After that, Nvidia released the 9800 GTX and 9800 GTX+. This was the end point in the development of an aging platform.

The developers decided to squeeze all the juice out of it and release a modified G92b graphics chip, you can call it overclocking. At that time, a competitor appeared on the market in the face of the Radeon 4850. The improved processor is quite enough to compete with the new motherboard from AMD, so it was decided to release a video card with the GTS250 index. This model was the last to be developed on the G92b.

Specifications

In fact, the GTS 250, although its performance has improved significantly, remains the same as the 8800. The only thing that has changed is higher frequencies and an increase in video memory. The model was now offered with two options - 512 MB and 1 GB of memory.

So, the video card has a core frequency of 738 MHz and a shader processing frequency of 1800 MHz. buses - 70 Gb / s. The interface for connecting a video card to the motherboard is PCI Express 2.0.

All the innovations that distinguish the GTS250 video card from its predecessor 9800 are implemented at the software level. Support for PhysX, CUDA - all these changes are at the driver level. Therefore, we can conclude that work in the architecture of the board was not carried out. So how much better does this model look like? We will learn about this later from testing.

At the time of the release of the video card, its cost ranged from 6 to 8 thousand rubles, which made it possible to easily compete with AMD in the middle price segment. For the option with a memory capacity of 1 GB, I had to pay several thousand more. Initially, the company's plans were to release the third version of the Nvidia GTS 250, the characteristics of which would differ in the amount of video memory of 2 GB. But due to low popularity and small sales, this idea had to be frozen. From this you can immediately understand that such solutions on older platforms did not bring success to Nvidia.

ASUS and Palit

In order not to be unfounded, let's take two specific video cards from well-known companies. The first one is ASUS GTS 250 and the second one is Palit GTS 250. Both are 512 MB. First, let's take a closer look at their characteristics. Let's take a look at the kits as well.

Asus GeForce GTS 250 Specifications

The video card from ASUS exactly repeats the original. With only one difference: in this board, the memory frequency was increased to 1900 MHz. Inside the box of the device, we are waiting for one adapter for PCI-E, an audio cable, a disk with all the necessary drivers that were relevant at the time of release, and instructions. The video card itself has small dimensions and will easily fit into the compact case of the system unit. The cooling system consists of one cooler and a radiator. Three connectors are traditionally located on the back side - HDMI, DVI and D-SUB. The bottom side is completely empty - all microcircuits are located directly under the cooler and radiator, which provides them with good airflow and low temperatures even at high loads.

The cooling design of the GeForce GTS 250 512mb, which has default characteristics, is rather complicated. The cooler is connected to a radiator, and there are also four heat pipes. Even at the highest load in demanding games at maximum graphics settings, the temperature of the device did not rise above 60 degrees Celsius. relatively low, given the budget price category of this model.

Palit GeForce 250 GTS Specifications

Let's move on to the video card from Palit, whose technical characteristics are somewhat lower, but the variability of use is greater. The box and equipment are standard - all the same adapters, drivers and instructions.

Externally, the board looks more solid. The cooler and cooling is covered by a translucent plastic case. In addition to standard connectors and interfaces, there is an output for connecting a video card in SLI mode (three identical video cards work simultaneously).

The cooling system is much simpler - one simple radiator and a rather noisy cooler on top. It is understandable, video cards from Palit have always been more budget options. Because of this, much stronger. The maximum recorded temperature in Celsius during testing is 80 degrees. But in terms of noise level, both video cards are approximately on the same level. The power consumption of the Palit video card is somewhat lower than that of the ASUS GTS 250 512mb. The characteristics of the video card are otherwise the same as the previous model.

test stand

For testing in games and programs, we chose a balanced PC configuration. The processor from the company Intel - Core 2 Quad quad-core with a frequency of 2.8 GHz for each core. 4 GB RAM and 320 GB hard drive. The system also has an 850W power supply. The computer is controlled by the operating system Windows 7 64 Bit. All tests were carried out in the same resolution - 1680 by 1050 pixels.

Competitors

For video cards GTS 250, whose characteristics are almost the same, a competitor was chosen in the person of 4850. The memory capacity of the competitor is the same - 512 MB. In general, all video cards are in the same class. After all, the GTS 250 was even created as a direct competitor to the model from AMD.

Test results

The first game was Call Of Duty World At War. Both GTS 250 video cards produced 36-38 frames per second. The video card from AMD also did not lag behind - about 34 frames per second.

Further testing was carried out in the game Mafia 2, which is also quite demanding on hardware. And again, all video cards showed exactly the same results. So all three models show themselves in absolutely any games. The difference is only in the heating temperature, noise level, energy consumption and cost.

nVidia GeForce GTS 250 | Introduction

In 2008, we saw a lot of ups and downs in the graphics card market. First, nVidia introduced its GT200 GPU and a couple of graphics cards on it. It swept everything else out of the lead - not surprising given AMD's mid-range Radeon HD 3800 lineup, which has already given up.

Then AMD surprised everyone by introducing the RV770 GPU and two graphics cards based on this processor. The fastest Radeon HD 4870 was still not fast enough to beat nVidia's fastest GPU, but fast enough that the dual-socket Radeon HD 4870 X2, which AMD announced when the RV770 was released, would later take the lead.

Then AMD was busy filling its line of mid- and entry-level graphics cards with GPUs of the same architecture. Radeon HD 4830 has become the most inexpensive high-performance graphics card. The Radeon HD 4670 and 4650 models have become AMD's midrange strike force. And the Radeon HD 4500/4300 video cards formed the entry level.

Nvidia has responded to AMD's actions with several moves. In the high-end market, it introduced its own dual-GPU GeForce GTX 295 graphics card. The center of the performance line was strengthened by the revised GeForce GTX 260 with 216 stream processors, which became a worthy competitor to the AMD Radeon HD 4850 (and 4870 with 512 MB of memory, as you will see in our tests). And the 55nm GT200 replacement allowed the company to introduce the latest GeForce GTX 285.

Of course, nVidia has also placed emphasis on add-ons such as CUDA, PhysX and 3D Vision, all of which have become supported through the company's drivers. Yes, this tech trio is still in the early stages of mass market adaptation, but technically we still have an advantage. AMD is working on its own Stream video encoder, does not offer any kind of physics acceleration, and is not very active in talking about the results of a collaboration with 3D monitor manufacturer iZ3D, although it was announced at this CES and is intended to bring 3D games to AMD or nVidia video cards.

nVidia GeForce GTS 250 | Need an answer for the mass market

Of course, nVidia managed to close all the "holes", but still the huge 55nm GT200 die is too large (and therefore expensive) to work in video cards cheaper than the GeForce GTX 260, which leaves nVidia without a worthy successor to the aging G92, which is almost a year and a half old .

Luckily for Nvidia, this relatively old architecture was designed and released very well, and it has already survived the transition from 65nm to 55nm. Even today, it is quite capable of competing with the RV770 line from AMD - this fact is proved by the current release of the GeForce GTS 250.

If the name of the board sounds like something new, located between the GTX 260/285 and the old GeForce 9800 line, in fact we are facing the rebirth of the G92. To be more precise, we have a GeForce GTX 9800+, which switched to a smaller technical process GeForce GTX 9800, which, in turn, was a slightly overclocked version of the GeForce 8800 GTS.

nVidia GeForce GTS 250 | In detail

As we mentioned above, the GTS 250 is a GeForce 9800+ under a new name. Therefore, the specifications of the card can be easily remembered if you were already familiar with the nVidia line.

Click on the picture to enlarge.

Manufactured using TSMC's 55nm process, the G92 GPU uses 754 million transistors and covers a die area of 230 square millimeters - less than half of the GT200. It contains 128 unified stream processors, 64 texture units and 16 raster operations units (ROPs).

While most GeForce GTX 9800+ cards shipped with 512MB of GDDR3 memory, the reference GeForce GTS 250 comes with 1GB of GDDR3, according to BFG and Zotac. However, both old and new video cards based on the G92 will use a 256-bit memory bus.

The reference GTS 250 is based on a GPU with a frequency of 738 MHz, a frequency of shader units (stream processors) of 1836 MHz. The memory frequency is 1.1 GHz (2.2 GHz effective). But BFG sent a GeForce GTS 250 OC Edition to our lab with a higher core clock of 750 MHz and a memory clock of 1.12 GHz, while the shader clock remained unchanged.

	GeForce GTX 260 Core 216	GeForce GTS 250	GeForce GTX 9800+	Radeon HD 4870	Radeon HD 4850	Radeon HD 4830
Process technology	55 nm TSMC	55 nm TSMC	55 nm TSMC	55 nm TSMC	55 nm TSMC	55 nm TSMC
Number of stream processors	216	128	128	800	800	640
Core frequency	576 MHz	738 MHz	738 MHz	750 MHz	625 MHz	575 MHz
Shader unit frequency (stream processors)	1242 MHz	1836 MHz	1836 MHz	750 MHz	625 MHz	575 MHz
Memory frequency	999MHz GDDR3	1100MHz GDDR3	1100MHz GDDR3	900MHz GDDR5	993MHz GDDR3	900MHz GDDR3
Memory	896 MB	1 GB	512 MB	512 MB	512 MB	512 MB
Memory bus width	448 bit	256 bit	256 bit	256 bit	256 bit	256 bit
Number of ROPs	28	16	16	16	16	16
International market price	$229	$149	$144	$164	$139	$89

The BFG video card has two dual-link DVI outputs. HDMI is also supported, but this feature requires an adapter and audio cable sold separately. In addition, the card supports 2- and 3-way SLI; the required bridges must be included with the SLI-compatible motherboard.

nVidia GeForce GTS 250 | Is it worth taking?

Click on the picture to enlarge.

At this point, it is clear that the old graphics architecture has been reused. But is it really that bad? Are there any problems associated with this?

Technologically, there are no problems. None of the modern nVidia video cards support DirectX 10.1, so the use of the old design does not give any technological lags between the GeForce GTS 250 and the "older" GT200 models. In fact, as long as productivity increases and prices fall, we're not in the least against reusing proven technologies. Moreover, with software support for CUDA, PhysX and 3D Vision on all graphics cards up to the GeForce 8 line (and PureVideo HD is supported up to the 9 line), we see no reason why the GTS 250 should not be taken just because it is GeForce GTX 9800+ update.

IntroductionAs you know, the first launch of the Nvidia GeForce 200 architecture is unlikely to be called successful. Having bet on maximizing the complexity and performance of a monolithic core, Nvidia seemed to have lost when ATI responded with its "secret weapon" to the release of the G200 - a relatively simple and significantly less expensive RV770 processor. Solutions based on the Nvidia G200, already expensive due to the monstrous complexity, instantly became unprofitable when, in an attempt to make them competitive, the company was forced to drastically reduce prices. And even such a step had little effect on their attractiveness - the fact is that the chip, which consisted of almost one and a half billion transistors and was produced using 65-nm technological standards, by definition could not work at high frequencies. In the older model of the new family, Nvidia barely managed to reach the GPU frequencies up to 600/1300 MHz, and the younger one was even slower, and, as a result, could not compete with the Radeon HD 4870, which, moreover, was much cheaper.

But time passed, and as time went on, Nvidia gradually managed to, as they say, “pick up the tails”. First, the company raised the performance of the younger model of the family, the GeForce GTX 260, by unlocking some of the GPU functional blocks, as a result of which the new product, which received the “Core 216” additive to the name, was able to successfully compete with the ATI Radeon HD 4870. Then, later, it also became a testing ground, on which a new version of the G200 core was tested for the first time, transferred to the 55-nm process technology, as well as a new, significantly simplified, and therefore lower-cost PCB design was tested, which, under conditions of severe pressure from ATI, was necessary for solutions based on G200 is like air. The design of the GeForce GTX 285 was also radically simplified, but thanks to a more advanced technical process, the frequency potential of the G200b was significantly increased, as a result of which the heir to the GeForce GTX 280 in some cases was able to compete on equal terms with the ATI Radeon HD 4850 X2. In addition, the availability of the 55nm G200 variant in the arsenal allowed Nvidia to finally strike back at ATI in the sector of the most productive dual-processor gaming solutions, and the GeForce GTX 295, announced on January 8, 2009, really easily removed the ATI Radeon HD 4870 from the throne of the king of three-dimensional graphics x2.

However, in the sector of less expensive solutions, things are not so rosy for Nvidia: the monstrous G200, even in its new, 55-nm incarnation, is completely unsuitable for creating such cards, due to complexity and high cost, and the company does not yet have a new budget core. All that it has at the moment is the 55-nm version of the G92, which was chosen as the basis for creating a new mass graphics adapter, designed to replenish the GeForce 200 line from below. But the company didn't just rename the GeForce 9800 GTX+, and we need to explain why.

To begin with, it's worth making a digression and reminding readers that even the G200 is not a revolutionary chip, in fact, it differs from the G92 only in the number of functional units and ALU organization, which was described in the corresponding review. The G92 itself, in turn, traces its pedigree from the G80, announced in November 2006, in fact, differing from its ancestor only in the architecture of texture processors.

Thus, by the standards of the computer industry, the current Nvidia architecture beats all longevity records. On the one hand, this can be seen as retrograde, but on the other hand, as evidence of the maturity of the architecture itself. In fact, solutions based on the G92 may well be included in the GeForce 200 family, and there will be nothing reprehensible or illogical in this, since there are no cardinal differences between this core and the G200. Of course, this thesis does not cancel the natural question of what the Nvidia development department has been doing all this time, but it is beyond the scope of this review, although we still know something about the future generation of Nvidia graphics cores.

So, the GeForce 9800 GTX+ based on the 55nm G92b core turned out to be a good competitor for the "popular video adapter" in the face of the Radeon HD 4850, but here, in addition to creating confusion in the names of Nvidia product lines, more significant pitfalls were discovered. The fact is that this card inherited the PCB design from the GeForce 9800 GTX, which was originally created as the flagship of the GeForce 9 line, designed to replace the GeForce 8800 GTX/Ultra. For solutions with a recommended cost of less than $ 150, such a complex and, therefore, expensive design was, to put it mildly, not the best fit. Something had to be done.

Nvidia GeForce GTS 250: birth and specifications

I must say that Nvidia had a suitable board design at its disposal for quite a long time, starting from the end of 2007 - we are talking about the GeForce 8800 GT / GTS 512 family, which also used the G92 chip, however, for an unknown reason, the company decided not to use it, designing a new PCB for the GeForce GTS 250 from scratch. The motives for such a decision are completely unclear to us, since the development of a new board always means additional time and financial costs, which, in this case, could well have been avoided.

It is possible that Nvidia took such a step to avoid unnecessary criticism, but even the use of the new board does not make the GeForce GTS 250 a unique product - the new product, in fact, remains the same GeForce 8800 GTS 512 with increased clock speeds and doubled in one of the options the amount of local video memory! One way or another, it must be admitted that Nvidia did not present anything new to the public, once again offering customers an old candy in a new wrapper. We will refrain from moral assessments of this act, because, despite its considerable age, the GeForce 8/9 architecture copes well with the tasks assigned to it, however, we note that the use of the “GTS” suffix still creates some confusion. In our opinion, in order to finally unify the names of their products, Nvidia should have followed in the footsteps of ATI, refusing to use letter suffixes completely, but, perhaps, they were kept for a clearer separation of solutions based on the G200 and G92 - GTX for the former and GTS for the latter. . It is possible that soon we will see the GeForce 9800 GT under the name GeForce GTS 240 or 230, although, according to the latest information, video card manufacturers consider such a renaming unnecessary.

So, GeForce GTS 250. How does Nvidia's "new" solution look against the background of rivals from the "red" camp?

As already mentioned, nothing new, except for the increased local video memory in one of the GeForce GTS 250 variants - from 512 to 1024 MB. In fact, we still have the same GeForce 9800 GTX+, and, accordingly, there can be no talk of any noticeable changes in gaming performance, although in some cases a more expensive version can benefit somewhat due to the presence of 1 GB of memory on board.

All the innovations that Nvidia announced, such as support for CUDA and PhysX, are purely software and, in fact, "implemented" by unlocking at the driver level.

Nevertheless, precisely because the GeForce GTS 250 does not differ in its technical characteristics from the GeForce 9800 GTX +, it does not look inferior against the background of the Radeon HD 4850, except for the features that are not directly related to the field of gaming 3D, namely - a less advanced video processor that is unable to fully decode the VC-1 format in hardware, as well as the lack of a built-in audio core with stream output to the HDMI interface.

The recommended cost of new items at the time of the announcement was $149 for the version with 1 GB of video memory and $129 for the version with 512 MB of video memory; in addition, there are mentions of a $169 model equipped with 2 GB of local memory, but this option is unlikely to become any popular, since such a volume is not used by modern games, and even more so, it makes no sense in the case of an inexpensive mass solution. Theoretically, such a card could find its place in low-cost GPGPU systems, but the current Nvidia architecture has performance issues in double precision (FP64) mode, and single precision (FP32) in many cases is not enough for serious applications.

In practice, at the time of preparation of the article, cards based on the GTS 250 in Moscow retail cost 6-7 thousand rubles, depending on the amount of memory.

As for specific products, in this review we will talk about an unusual version of the GeForce GTS 250, presented by Palit Microsystems.

Palit GeForce GTS 250 1GB: packaging and equipment

The described product is packaged in an ordinary cardboard box of standard sizes, and the packaging design can hardly be called successful, at least in our opinion:

The color solution used by Palit designers, which uses black and yellow colors, does not look very aesthetically pleasing, especially in combination with the picture depicting a frog in an armored suit, which we have already seen on the packages of Palit GeForce 9800 GTX+ cards.

The described packaging also has a positive feature - its information content. The sticker in the lower right corner, although it does not report the clock speeds, but it correctly indicates the type and amount of video memory, and there is also a mention that this product has GDDR3 chips with an access time of 0.8 nanoseconds. Together with the inscription "Play It, Tweak It, Get More Out of It" and a low price, this indicates that this product is aimed at overclockers who are on a budget.

The bundle is quite modest, but includes everything you need for the full operation of this video adapter:

HDMI → DVI adapter
Adapter 2×PATA → PCIe
S/PDIF connecting cable
Quick User Guide
CD with drivers

It's not rich, but a buyer who is guided by this price category is unlikely to chase after a mass of additional accessories that only increase the price of the product. Note that Palit has taken care of both the owners of two monitors with a DVI interface and those who plan to use the card to output audio via HDMI to an external receiver by supplying their product with the appropriate adapters and cables. It is a bit unusual to see an HDMI → DVI adapter in the package, but in this case its presence is due to the fact that the described version of the GeForce GTS 250 has only one “native” DVI-I port, and the HDMI connector acts as the second.

In general, the packaging and bundle of Palit GeForce GTS 250 1GB are fully consistent with the price category to which this product belongs. However, we believe that the packaging design should have been given more attention, as it does not look very attractive in its current form.

Palit GeForce GTS 250 1GB PCB design and specifications

This version of the GeForce GTS 250 uses a non-standard PCB design that has nothing to do with the reference design developed by Nvidia specifically for the new mainstream video adapter. The Palit GeForce GTS 250 1GB printed circuit board is largely identical to the one described in the review " Brothers in Arms: Two Palit GeForce 9800GTX+ Video Cards”, although there are a number of differences, especially on the left side of the board, which is due to the two-sided arrangement of memory chips, as well as the configuration of the interface connectors that this product is equipped with.

Compared to the GeForce 9800 GTX+ reference variant (board length 267 mm), the Palit design has one advantage - the shorter board length, however, if we compare it with the Nvidia reference design for the GeForce GTS 250, also shortened to 244 mm, then instead of an advantage, we immediately find a rather significant drawback: the need to connect two power connectors instead of one. Considering the G92b's power consumption is about 80W, the presence of two six-pin PCIe 1.0 power connectors looks like overkill, which has no practical justification.

The GPU power system itself is identical to that used in one of the Palit GeForce 9800 GTX+ variants and is a four-phase regulator with three power MOSFETs in each phase, controlled by ON Semiconductor NCP3488 drivers. The NCP5388 chip is used as a controller.

The Anpec APW7068 chip, which is a PWM controller operating at a frequency of 300 kHz, is responsible for controlling the memory power regulator.

The board has 16 Samsung K4J52324QH-HJ08 GDDR3 chips with a capacity of 512 Mbps (16Mx32), designed for a supply voltage of 2.05 V. Half of the chips are located on the reverse side of the board and are covered with an aluminum heat-distributing plate. The total amount of local video memory is 1024 MB, and the width of the access bus to it is 256 bits.

The suffix in the memory marking really indicates an access time of 0.83 ns, which means the ability of these microcircuits to operate at a frequency of 1200 (2400) MHz. According to Nvidia's specifications, the GeForce GTS 250 memory should run at 1100 (2200) MHz, but the launch of GPU-Z revealed that in the Palit variant its frequency is only 1000 (2000) MHz. Thus, in some tests you should expect it to lag slightly behind the reference card, even if it is equipped with half the amount of video memory.

The graphics core has a slightly different marking than that installed on the Palit GeForce 9800 GTX + - "G92-428-B1" versus "G92-420-B1", although the revision number matches and indicates that we have a G92b produced using 55- nm technological standards. This is also indicated by the area of the crystal. This copy was produced in the second week of this year, that is, in early January. It is curious that the marking itself is made in a different, wider font than in older copies of the G92 / G92b.

Like the memory frequency, the GPU clock speeds differ from the official specifications: if the reference values for the GeForce GTS 250 are 738/1826 MHz for the main and computing domains, respectively, then in the Palit version these blocks operate at frequencies of 745/1848 MHz, which -something less can compensate for the lower memory frequency. The graphics core configuration is standard and includes 128 universal shader processors, 64 texture processors and 16 RBE units.

Among other features typical of the Palit GeForce GTS 250 1GB, we should note the non-standard configuration of interface connectors: instead of the usual two DVI-I ports, supplemented by a universal analog video output port, the card has one DVI-I port, one analog D-Sub and one HDMI. A somewhat strange decision for 2009: at present, there is no serious need for a separate D-Sub connector, since the corresponding connection can be easily arranged using a penny DVI-I → D-Sub adapter. But connecting a second monitor with a DVI interface in this configuration will have to be done through an HDMI → DVI adapter, the reliability of which inspires some concern - theoretically, it is possible that under the weight of the connected cable, this adapter may fall out of the socket, because the HDMI port does not have an equally reliable system screw fixation, like a DVI connector. In our opinion, the optimal connector configuration would be "2xDVI-I + 1xHDMI", since the presence of a dedicated HDMI port in our time is much more relevant than the presence of an analog video output connector.

Other connectors present on the board include two standard "combs" of the MIO interface, which is used to combine two or three cards into an SLI or Triple SLI system, respectively, as well as a two-pin S / PDIF connector, designed to organize the transmission of an audio stream to HDMI .

In general, the use of Palit's non-standard motherboard design for the GeForce GTS 250 seems to us a dubious enterprise: perhaps its cost does not exceed the cost of Nvidia's reference design, but the described design is clearly less convenient in operation due to the need to connect two power connectors instead of one.

Palit GeForce GTS 250 1GB: cooling system design

The reference cooling system GeForce GTS 250 is another variation on the theme of the GeForce 8800 GTX cooler, not being a champion, but combining acceptable cooling efficiency with a low noise level. Nevertheless, Palit decided to go its own way in this matter, equipping its version of the GeForce GTS 250 with a non-standard cooling system - most likely, just in order to stand out from a number of cards of the same type in appearance.

Nothing extraordinary, however, it is not; moreover, its design is such that it immediately makes it possible to call into question the thermal efficiency of the applied design. By itself, it is very simple, and, in fact, it is a low heatsink made of thin aluminum plates, connected to a copper base with two heat pipes and blown from above by a fan mounted on the casing.

You can immediately see two main drawbacks of the system: firstly, the area of the radiator itself is relatively small, and secondly, the use of an axial fan at such a height of the radiator does not allow one to hope for the efficiency of blowing the latter, since we are talking about a 90-degree turn of the air flow at a distance of about one and a half centimeters. In addition, a large part of the airflow directed away from the fan is completely wasted, because, unlike the GeForce 7900 GTX cooling system, there are no additional sections of the radiator that use these flows. Slots in the mounting plate of the card with this arrangement of the cooling system play mainly a decorative role - a small amount of air is blown out of the system unit through them

The system uses an Everflow R128015BH fan with an impeller diameter of 80 mm, a height of 15 mm, and a power consumption of 3.84 watts. At maximum speed, the fan is quite noisy, but more on that below.

As in the Gainward HD 4850 1024MB GS, a layer of light gray thermal interface is applied to the copper sole of the heat exchanger, which has a very dry consistency and literally glues the sole to the GPU chip. The heatsink is attached to the board with four screws, which is not a problem due to its low weight and the presence of a protective metal frame on the GPU package, which prevents the heat exchanger from being skewed. The memory chips located on the front side of the board are covered by an L-shaped aluminum plate, and the chips installed on the reverse side are covered by another plate with a large area. The plates are tightened with six screws; elastic thermal pads are used as a thermal interface.

This design is made with claims for high cooling efficiency coupled with low noise level, however, it has a number of shortcomings visible to the naked eye, which cast doubt on the legitimacy of these claims; however, the last word will remain with the test results, which we will talk about in the next chapter of our review.

Palit GeForce GTS 250 1GB: power consumption, thermal conditions, noise and overclocking

Despite the fact that the Palit GeForce GTS 250 1GB uses a non-standard PCB design and power supply circuitry, we still measured its power consumption level using a test bench designed for such purposes with the following configuration:

Processor Intel Pentium 4 560 (3.6 GHz, LGA775)
Motherboard DFI LANParty UT ICFX3200-T2R/G (ATI CrossFire Xpress 3200)
Memory PC2-5300 (2x512 MB, 667 MHz)
Western Digital Raptor WD360ADFD Hard Drive (36 GB)
Chieftec ATX-410-212 Power Supply (410W Rated)
Microsoft Windows Vista Ultimate SP1 32-bit
Futuremark PCMark05 Build 1.2.0
Futuremark 3DMark06 Build 1.1.0

According to the established methodology, the first SM3.0/HDR test of the 3DMark06 package was used to create a load in 3D mode, running in a loop at a resolution of 1600x1200 with forced FSAA 4x and AF 16x. Emulation of the "peak 2D" mode was carried out using the "2D Transparent Windows" test, which is part of PCMark05. The last test is relevant in light of the fact that the Aero window interface of the main gaming operating system - Windows Vista - uses the capabilities of the graphics processor.

As already mentioned, the presence of two power connectors is an overkill for this video adapter, since even in 3D mode its power consumption level does not exceed 81 W, and each of these connectors, capable of withstanding up to 75 W, has a load of only about 28-29 W . Such a reinsurance seems to be absolutely unnecessary, and all that Palit managed to achieve with its help was to complicate the life of owners of inexpensive power supplies equipped with only one six-pin graphics card power connector.

Despite the controversial design of the cooling system, she managed to demonstrate very good results:

Under load, the temperature of the graphics core did not exceed 65 degrees Celsius, which is 5 degrees lower than that of the reference cooling system GeForce 9800 GTX+. Idle mode also pleased, despite the fact that the core temperature in it was still higher than that of the more powerful GeForce GTX 260 Core 216. Thus, our skepticism about the effectiveness of the low-profile heatsink turned out to be dispelled, although we should not forget that the results obtained during largely due to the cost-effectiveness of the 55-nm version of G92. It remains only to look at the noise characteristics:

Contrary to expectations, the Palit cooling system did not blunder here either - the noise level at a distance of a meter from the working test system turned out to be comparable to the performance of the reference GeForce 9800 GTX+ cooling system and only slightly exceeded the background value of 43 dBA. Unfortunately, the described cooler also has a drawback: before the operating system and the driver are loaded, the fan runs at high speeds and is quite noisy, but after loading the noise level returns to normal, and the card becomes almost silent. However, do not forget that the test platform we use is quite noisy in itself due to the Enermax Galaxy DXX EGX1000EWL power supply used in it.

An attempt to overclock the Palit GeForce GTS 250 1GB did not bring any special surprises, but it did not fail either:

Without the use of any extreme means, we managed to raise the core clock frequencies to 800/1984 MHz, and the memory - to 1150 (2300) MHz, while maintaining full stability. In this mode, the card went through the full cycle of testing without any problems, as a result of which we considered it necessary to publish the results obtained along with the results shown by Palit GeForce GTS 250 1GB in normal mode, when operating at factory frequencies.

Zotac GeForce GTS 250 AMP! Edition: reference design card

Above, we noted that Palit took the path of least resistance, modifying the already existing board from the previously released 9800 GTX + video card. This design is different from NVIDIA's reference design, which will no doubt be the basis for most video cards with the GTS 250 chip.

Fortunately, we also had a Zotac GeForce GTS 250 AMP video card in our hands! Edition, equipped with 1024 MB of memory and made according to the reference design, with the exception of slightly increased - up to 750/2300 MHz - operating frequencies. Although we only had the card in our hands for a couple of days, so it was not possible to run it through a full range of tests, we will say a few words about the appearance and design features - well, the performance in practice will only slightly differ from the Palit card discussed in detail in this review .

Externally, the Zotac GeForce GTS 250 resembles the good old GeForce 8800GTS/512, the second card based on the G92 chip - the first was the less powerful 8800GT with a single-slot cooling system. Compared to cards based on the 9800GTX+, the GTS 250 has been reduced in length to 244 mm, which allows it to be easily installed in the vast majority of cases.

The card uses the already classic NVIDIA cooling system, which is very efficient and at the same time one of the quietest among all COs installed on modern video cards. It consists of a large aluminum platform that contacts the memory chips and transistors of the power system through white soft heat-conducting pads, a copper pad pressed against the GPU chip through a layer of thermal paste, and heat pipes that ensure even heat distribution over the radiator fins. The system is blown by one impeller with a 4-wire connection.

Although the Zotac GeForce GTS 250 AMP! Edition is also equipped with 1024 MB of memory, there are no memory chips on the bottom side of the board.

On the top side of the board are eight Hynix H5RS1H23MFR GDDR3 chips, each with a capacity of 1 Gb. However, from the point of view of the buyer, something else is more interesting: the card has only one 6-pin power connector.

Zotac GeForce GTS 250 AMP! Edition uses the G92-428-B1 chip, just like the Palit card.

In fact, video cards based on the reference design on the GTS 250 chip, which include the Zotac GeForce GTS 250 AMP! Edition have two advantages over the Palit GeForce GTS 250: firstly, they have a single 6-pin power connector, and secondly, the NVIDIA-designed cooling system throws all hot air out of the case, unlike the Palit-designed cooling system. The latter is especially important for owners of compact microATX cases, in which new video cards can be easily installed due to the reduced length of the printed circuit board.

Test platform configuration and testing methodology

Palit GeForce GTS 250 1GB performance comparison study was conducted on a test platform with the following configuration:

Processor Intel Core i7-965 Extreme Edition (3.2 GHz, 6.4 GT/s QPI)
Motherboard Asus P6T Deluxe (Intel X58)
Memory Corsair XMS3-12800C9 (3x2 GB, 1333 MHz, 9-9-9-24, 2T)
Maxtor MaXLine III 7B250S0 hard drive (250 GB, SATA-150, 16 MB buffer)
Power supply Enermax Galaxy DXX EGX1000EWL (power 1 kW)
Dell 3007WFP Monitor (30”, maximum resolution [email protected] Hz)
Microsoft Windows Vista Ultimate SP1 64-bit
Nvidia GeForce 182.08 WHQL for Nvidia GeForce
ATI Catalyst 9.2 for ATI Radeon HD

Graphics card drivers have been tuned to provide the highest possible quality of texture filtering with minimal impact from default software optimizations. Transparent texture anti-aliasing has been enabled. As a result, the list of ATI Catalyst and Nvidia GeForce driver settings looks like this:
NVIDIA GeForce:

Texture filtering - Quality: High quality
Texture filtering - Trilinear optimization: Off
Texture filtering - Anisotropic sample optimization: Off
Vertical sync: Force off
Antialiasing - Gamma correction: On
Antialiasing - Transparency: Multisampling
Multi-display mixed-GPU acceleration: Multiple display performance mode
Set PhysX GPU acceleration: Enabled

ATI Catalyst:

Smoothvision HD: Anti-Aliasing: Use application settings/Box Filter
Catalyst A.I. Standard
Mipmap Detail Level: High Quality
Wait for vertical refresh: Always Off
Enable Adaptive Anti-Aliasing: On/Quality
Other settings: default

The package of test applications included the following games and synthetic tests:

3D First Person Shooters:

[**]Call of Duty: World at War
Crysis Warhead
Enemy Territory: Quake Wars
Far Cry 2
F.E.A.R. 2: Project Origin
Left 4 Dead
S.T.A.L.K.E.R.: Clear Sky

Three-dimensional shooters with a third-person view:

Devil May Cry 4
Prince of Persia

RPG:

Fallout 3
mass effect

Simulators:

Race Driver: GRID
X³: Terran Conflict

Strategies:

Red Alert 3
world in conflict

Synthetic tests:

Futuremark 3DMark06
Futuremark 3D Mark Vantage

Each of the games included in the set of test software was configured to provide the highest possible level of detail, and only the tools available in the game itself to any uninitiated user were used. This means a fundamental rejection of manual modification of configuration files, since the player is not required to be able to do this. For some games, exceptions were made, each of which is mentioned separately in the corresponding section of the review.

In addition to the Palit GeForce GTS 250 1GB, the test participants included the following graphics cards:

Nvidia GeForce GTX 260 Core 216 (G200b, 576/1242/2000 MHz, 216 SP, 72 TMU, 28 RBE, 448-bit memory bus, 896 MB GDDR3)
Nvidia GeForce GTS 250 512MB/GeForce 9800 GTX+ (G92b, 738/1836/2200 MHz, 128 SP, 64 TMU, 16 RBE, 256-bit memory bus, 512 MB GDDR3)
ATI Radeon HD 4870 1GB (RV770, 750/750/3600 MHz, 800 SP, 40 TMU, 16 RBE, 256-bit memory bus, 1024 MB GDDR5)
ATI Radeon HD 4850 (RV770, 625/625/2000 MHz, 800 SP, 40 TMU, 16 RBE, 256-bit memory bus, 512 MB GDDR3)

Testing was carried out at resolutions of 1280x1024, 1680x1050 and 1920x1200. Wherever possible, standard 16x anisotropic filtering was supplemented with 4x MSAA anti-aliasing. Activation of anti-aliasing was carried out either by the means of the game itself, or, in their absence, was forced using the appropriate settings of the ATI Catalyst and Nvidia GeForce drivers.

To obtain performance data, we used the tools built into the game with the obligatory recording of original test videos, if possible. If the built-in testing tools allowed, data was recorded not only on the average, but also on the minimum performance. In all other cases, the Fraps 2.9.8 utility was used in manual mode with a three-time test pass, fixing the minimum values and then averaging the final result.

Playtests: Call of Duty: World at War

As expected, the new GeForce GTS 250 did not show any noticeable advantage over the GeForce 9800 GTX+, since it is, in fact, what it is, and 1 GB of video memory in World at War is an obvious overkill. The meager lag at 1920x1200 is due to a slightly lower memory frequency. Compared to the Radeon HD 4850, the new product looks good, demonstrating approximately the same level of average and minimum performance.

Playtests: Crysis Warhead

Games based on the Crysis engine, on the contrary, are quite sensitive to the amount of video memory, especially at high resolutions. However, even at 1280x1024, the Palit GeForce GTS 250 1GB noticeably outperforms the model with 512 MB of video memory on board and successfully competes with the Radeon HD 4850, and even outperforms it at 1920x1200. But don't forget that the figures obtained are largely theoretical, since the overall level of average, and even more so, minimum performance is too low for the practical use of cards of this class in such modes.

Playtests: Enemy Territory: Quake Wars

In order to obtain more complete data on the performance of graphics cards in Quake Wars, the built-in average limiter has been disabled via the game console. Since the testing uses the internal capabilities of the game, there is no information on the minimum performance.

Despite the fact that the game uses ultra-high resolution textures ("MegaTexture" technology), the Palit GeForce GTS 250 1GB does not get any advantage over the GeForce 9800 GTX+ and, accordingly, the GeForce GTS 250 512MB. Moreover, at 1920x1200 the Palit version is noticeably inferior to the reference one due to the lower memory frequency. In general, the new product is not a revelation, continuing to demonstrate performance comparable to that of its direct competitor, ATI Radeon HD 4850.

Playtests: Far Cry 2

At maximum graphics quality settings, Far Cry 2 is less demanding than Crysis Warhead, and the presence of the corresponding amount of video memory on board Palit GeForce GTS 250 1GB gives not only theoretical results - at resolutions of 1680x1050 and 1920x1200, its influence allows you to overcome the barrier of minimum comfort and provide acceptable levels of average and minimum performance. Oddly enough, we didn't see a similar effect in the case of the Radeon HD 4850 equipped with the same amount of local memory. Perhaps the release of ATI Catalyst 9.3 will change the current alignment of forces.

Game Tests: F.E.A.R. 2: Project Origin

The effect of doubling the amount of video memory is noticeable in F.E.A.R. 2, but in this game it is most pronounced at low resolutions; for example, at 1280x1024 the increase in average performance is about 18%, while at 1920x1200 it drops to 4-5%. There is also a solid increase in the minimum performance, and although it does not open up new opportunities for the player, the results of the new mainstream Nvidia card look impressive, especially against the background of the frankly modest results of the Radeon HD 4850.

Playtests: Left 4 Dead

The game is based on the Source engine and has built-in testing tools, which, unfortunately, do not provide information about the minimum performance.

Due to the relatively old engine, the game does not have high requirements for video memory, and the Palit GeForce GTS 250 1GB is on par with the GeForce 9800 GTX +, the new name of which is the GeForce GTS 250 512MB. Thus, it makes no sense for fans of games on the Source engine to pay extra for additional half a gigabyte of video memory. In fact, in this case, the choice between the Radeon HD 4850 and both GeForce GTS 250 models is completely equivalent.

Game Tests: S.T.A.L.K.E.R.: Clear Sky

To ensure an acceptable level of performance in this game, it was decided to abandon the use of FSAA, as well as such resource-intensive options as "Sun rays", "Wet surfaces" and "Volumetric smoke". During testing, the "Enhanced full dynamic lighting (DX10)" mode was used, for ATI cards, the DirectX 10.1 mode was additionally used

Despite the exacting nature of this game, the effect of having 1 GB of video memory on board the graphic card under study is not very significant and manifests itself only at a resolution of 1920x1200, where the average performance increase is only 10%. Not bad, but not enough to provide an acceptable level of performance; however, even at 1280x1024, the Palit GeForce GTS 250 1GB is a bit lacking in minimum speed, unlike the Radeon HD 4850, which can take advantage of Clear Sky's DirectX 10.1 support.

Playtests: Devil May Cry 4

It seems that the frequency of the video memory, and therefore its bandwidth, has a greater impact on the performance of the GeForce GTS 250 than its volume or GPU frequency - in two resolutions out of three, the Palit GeForce GTS 250 1GB is noticeably inferior to the reference card equipped with 512 MB of local memory. This is not a problem, as the overall performance level remains very high, however, the negative effect of the "underclocking" of memory undertaken by Palit is obvious.

Playtests: Prince of Persia

As in the case of the two Radeon HD 4850 models, which differ in the amount of video memory, there is no effect from doubling it, and the Palit GeForce GTS 250 1GB performance is even slightly lower than that of the reference card equipped with 512 MB of memory due to its lower frequency. Also, at resolutions up to 1680x1050 inclusive, there is a very significant advantage in minimum performance over the Radeon HD 4850.

Playtests: Fallout 3

The model GeForce GTS 250, equipped with 1 GB of memory, is ahead of the model with 512 MB on board, but very slightly - the difference does not exceed 2.5-6%. Only overclocking the core to 800 MHz, and the memory to 1150 (2300) MHz allows the new product to reach the level of the Radeon HD 4850.

Playtests: Mass Effect

The Palit product behaves in approximately the same way in Mass Effect, but even the maximum overclocking does not allow achieving a stable minimum performance at a level above 24 frames per second - of the cards presented in this review, only the GeForce GTX 260 Core 216 can do this.

Game Tests: Race Driver: GRID

Both versions of the GeForce GTS 250 demonstrate almost the same results, successfully competing on equal terms with the Radeon HD 4850, but overclocking the Palit GeForce GTS 250 1GB allows you to achieve a solid, about 16-18%, increase in average performance, which brings this card to the same level with more expensive GeForce GTX 260 Core 216.

Playtests: X³: Terran Conflict

Another case where the difference in performance between two almost identical cards, differing only in the amount of video memory, is minimal. However, do not forget about the reduced memory frequency of the Palit GeForce GTS 250 1GB, without which the gain could have been a little more. However, we doubt that this would allow us to catch up with the Radeon HD 4850 - the game clearly prefers the ATI Radeon HD architecture.

Playtests: Red Alert 3

The game contains a non-disableable average performance limiter, fixed at around 30 frames per second.

Nvidia solutions are still experiencing serious performance problems in this game with FSAA 4x enabled, which even the maximum overclocking of the Palit GeForce GTS 250 1GB does not help solve. Note that the presence of a large amount of video memory provides a noticeable increase in the average speed, but a lower memory frequency compared to the reference value leads to a slight drop in the minimum performance.

Game Tests: World in Conflict

The amount of video memory has no visible effect on the performance of different GeForce GTS 250 variants in World in Conflict. It is also obvious that at nominal frequencies their minimum performance is below an acceptable mark, but overclocking allows you to compensate for this and, thus, provide comfortable performance at a resolution of 1280x1024.

Semi-synthetic benchmarks: Futuremark 3DMark06

Despite using a resolution of 1280x1024, the Palit GeForce GTS 250 1GB outperformed the GeForce GTS 250 512MB by a value that is clearly beyond the measurement error. The results of individual test groups revealed that the largest performance gain was observed in the SM2.0 tests, while in the SM3.0 / HDR tests there was almost no increase, and even some lag was observed due to a slightly lower memory frequency.

Semi-synthetic tests: Futuremark 3DMark Vantage

To minimize CPU impact, 3DMark Vantage uses the "Extreme" profile for testing, using 1920x1200 resolution, FSAA 4x, and anisotropic filtering. To complete the picture of performance, the results of individual tests from now on are taken at the entire range of resolutions.

The conditions created by the 3DMark Vantage tests for the graphics subsystem are more in line with current realities, therefore, there is a benefit both from the presence of 1 GB of video memory on board the video adapter under study, and from additional overclocking. In the latter case, the Palit GeForce GTS 250 1GB manages to noticeably outperform the Radeon HD 4850, at least the variant equipped with 512 MB of memory.

Although in the overall standings one could observe an almost linear increase in performance, according to the results of individual tests, it can be seen that the complex overclocking of the GPU and memory gives a significantly greater effect than increasing the amount of video memory from 512 MB to 1 GB, at least at resolutions up to 1920x1200.

Conclusion

As expected, the GeForce GTS 250 did not make a splash - and what could be expected from a graphics adapter, which, in fact, is a renamed GeForce 9800 GTX +, made on a shortened board? Of course, one can talk for a long time about the moral and ethical aspects of Nvidia's act, once again offering customers the old architecture under a new name, however, if you approach the issue pragmatically, it is obvious that it is too early to write off this architecture, since with its direct responsibilities, and Precisely by providing acceptable performance in modern games, it copes quite well. The 55-nm process technology helps her in this, which made it possible to significantly raise the frequency potential of the G92 core.

Alas, in light of the recent price cuts undertaken by ATI, the situation does not look so rosy - at the same price, the GeForce GTS 250 will be hard to compete on equal terms with the Radeon HD 4870. However, let's look at the situation with the performance of new products in detail:

At a resolution of 1280x1024, there are no real advantages from having one of the GeForce GTS 250 gigabytes of video memory on board - the average performance increase is only 4%, and in some cases there is even a slight lag. The latter, however, is a shortcoming of the specific Palit model, which has a memory frequency lower than the nominal one.

If we compare the Palit GeForce GTS 250 1GB with the Radeon HD 4850, then, in general, we can talk about parity, although such a statement is comparable to the joke concept of "average temperature in a hospital." It all depends on the specific game: if in F.E.A.R. 2: Project Origin or Prince of Persia advantage is clearly on the side of Nvidia's decision, then, for example, in X3: Terran Conflict, ATI's brainchild leads by a solid margin.

But in the case of the Radeon HD 4870, the picture looks rather bleak - Palit GeForce GTS 250 1GB lost to it in almost all tests, so the average lag was about 22%. We are talking about the Radeon HD 4870 1GB, but we do not think that in the case of the Radeon HD 4870 512MB, whose official price was recently reduced to $149 (real retail is about 8 thousand rubles, but this, as you understand, is only a matter of time), the situation would improve significantly. Nvidia will clearly have to cut prices for the new family in order to keep it competitive in the market.

As for the effect of overclocking, it varies between 4-29%, averaging about 12%. Quite a good increase, achieved, moreover, with "little bloodshed".

At 1680x1050, the picture does not change: the performance increase relative to the GeForce GTS 250 512MB/GeForce 9800 GTX+ is about 4-5%, the situation with the Radeon HD 4850 should be considered separately for each game, and the average lag behind the Radeon HD 4870 1GB increases to 24%. The effect of overclocking the Palit GeForce GTS 250 1GB also slightly increased, averaging 13.5%.

Using the 1920x1200 mode together with FSAA 4x is already seriously enough loading the memory subsystem for its volume to begin to have a noticeable impact on performance. On average, the gain of Palit GeForce GTS 250 1GB over GeForce GTS 250 512MB increased to 6.5%, and in some cases, such as Crysis Warhead and Far Cry 2, it significantly exceeded 20%. However, in some other games, in particular, Enemy Territory: Quake Wars and Devil May Cry 4, the reduced memory frequency affected and the Palit product lost out to the Nvidia reference card with half the amount of local memory on board.

The situation in the confrontation between Palit GeForce GTS 250 1GB and Radeon HD 4850 has not changed: in a number of games (Crysis Warhead, FEAR 2: Project Origin) the former showed the best performance, while in others (for example, Enemy Territory: Quake Wars or Fallout 3) for the buyer looks preferable to the second.

It is useless to fight the Radeon HD 4870 at this resolution, since it is equipped with ultra-fast GDDR5 memory, which provides it with a bandwidth of 115.2 GB / s. vs. 70.4 GB/sec. for the GeForce GTS 250. In this case, the average loss of the Nvidia novelty was almost 30%.

conclusions

At first glance, the Nvidia GeForce GTS 250 family looks good, and, despite the use of an architecture that is not new by modern standards, it can successfully compete with the extremely popular "people's video adapter" in the face of the ATI Radeon HD 4850. However, the GeForce GTS 250 already costs 500 -1000 rubles more than the Radeon HD 4850, and in the near future the cost of the latter may fall even more - recently the official price of the ATI Radeon HD 4870, a card noticeably more powerful than the GeForce GTS 250, was reduced to $149. Thus, it all depends on Nvidia's strategy, and if the GeForce GTS 250 prices drop at the same rate as the Radeon HD 4850 prices, these cards will have a chance to find some popularity among gamers on a budget. Otherwise, ATI will have a clear advantage.

As for the specific model of the new graphics adapter, described in detail in this review, the Palit GeForce GTS 250 1GB, at first glance, looks quite attractive, combining good performance, low noise level and rather high cooling efficiency. It also has disadvantages, in particular, the need to connect two power connectors, despite the fact that only one connector remained in the reference design.

Palit GeForce GTS 250 1GB: advantages and disadvantages

Advantages:

Relatively high level of performance
Outperforms ATI Radeon HD 4850 in some tests
Availability of 1 GB of local video memory
Wide choice of FSAA modes
Minimal performance impact of FSAA
Excellent quality anisotropic filtering
Support HD video hardware decoding
Support for S/PDIF audio output via HDMI
Low power consumption
Low noise
Dedicated HDMI port

Flaws:

Video memory frequency below reference value
Increased noise level before driver loading
Lack of support for DirectX 10.1 and Shader Model 4.1
Incomplete support for VC-1 decoding
Lack of built-in sound core
Requires two power outlets

Check the availability and cost of video cards based on the GeForce GTS 250