by Alexey Stepin , Yaroslav Lyssenko
11/26/2007 | 12:01 PM
There is a big gap between technical characteristics of top-end and mainstream solutions currently offered by the consumer 3D graphics industry. Of course, there has always been that kind of gap, but it has never been as wide as between the ATI Radeon HD 2900 and ATI Radeon HD 2600. The same is true for Nvidia’s GeForce 8800 and 8600 families, but Nvidia offers a couple of in-between products: the GeForce 8800 GTS 320MB with its aggressive price of $299 and the GeForce 8800 GT can provide quite a high performance in modern games.
The former ATI Technologies, currently the graphics division of Advanced Micro Devices, used to be in a worse position because its Radeon HD 2900 series with advanced technical characteristics began with a $399 product while the Radeon HD 2600 XT GDDR4, the flagship of ATI’s mainstream series, was inferior to the Nvidia GeForce 8600 GTS. According to AMD’s original plans, the price range of $189-249 was to be filled in with a dual-chip solution based on two Radeon HD 2600 processors. Such solutions are doomed from the start, however, as we have had a lot of opportunities to see.
In the best case they can become an exotic rarity unable to provide maximum performance in all games, but in the worst case they are just completely forgotten. That was the fate of such products as the 6800-based dual-chip solutions from ASUS and Gigabyte, the unique ASUS Extreme N7800GT Dual, and the Nvidia GeForce 7950 GX2. Besides the main reason mentioned above, the high complexity and, accordingly, the high manufacturing cost of such products is a strong disadvantage.
There is no reason why the dual-chip Radeon HD 2600 should be different and the graphics card makers’ attitude towards it confirms our point. GeCube, PowerColor and Sapphire Technology were the only manufacturers to announce such products, but Sapphire supplies its solution only to the OEM market and system integrators while the versions provided by PowerColor and GeCube come with relatively slow GDDR3 memory. Thus, the Radeon HD 2600 X2 cannot be a good connecting link between the Radeon HD 2600 and Radeon HD 2900 families.
Poised for a forthcoming release of the ATI Radeon HD 3800, ATI decided not to focus on promoting the Radeon HD 2600 X2, but resorted to a good old trick of dealing with such situations. They have introduced a product that’s going to live only a few weeks on the market.
Historically, ATI and Nvidia both released cheaper graphics cards based on cut-down versions of top-performance GPUs. You may remember the Radeon 9500 and 9500 Pro, GeForce 8600 GS, Radeon X800 GT and GTO, and Radeon X1800 GTO with Radeon X1900 GT. Although not optimal in terms of manufacturing cost, this approach is still better than the production of dual-chip solutions with two less advanced GPUs just because it doesn’t require developing a unique PCB. Moreover, it helps dispose of chips that didn’t pass the quality check and cannot work at the frequencies of the senior model or have defective subunits. The share of such chips may vary greatly depending on tech process, but they are always present in the total chip yield and can be used together with full-featured cores.
That’s the way the ATI Radeon HD 2900 Pro was born. Contrary to the expectations of many gamers and mass media, the configuration of the R600 chip the new card is based upon has not been cut down. The clock rates of the core and memory are the only parameters that suffered a reduction. Thus, the new Radeon has the following specs:
In other words, the ATI Radeon HD 2900 Pro is in fact an ordinary ATI Radeon HD 2900 XT with 512MB or 1GB of GDDR3 memory accessed across a 512-bit memory bus and a graphics core frequency of 600MHz. The reduction of the memory frequency is negligible. According to the manufacturer, ATI R600 chips are clocked at 24 different frequencies, but the difference between the frequencies of its subunits must be small as opposed to Nvidia’s chips. As a result, the reduction of the frequencies is going to have a serious effect on the performance of the Radeon HD 2900 Pro in games.
The ATI Radeon HD 2900 Pro in both its versions may be a very appealing product for users who prefer ATI’s solutions but it cannot become a real bestseller. Why? As far as we know, these graphics cards are indeed manufactured out of defective samples of the R600 that are not capable of working at 742MHz. The total amount of such cards is going to be about 10 thousand at best, which is just a drop in the ocean of consumer 3D graphics. Obviously, the introduction of the ATI Radeon HD 2900 Pro is a temporary measure AMD has resorted to in order to fill in the gap between the ATI Radeon HD 2900 and ATI Radeon HD 2600 families until the announcement and release of mainstream graphics cards based on the new RV670 core.
Being nothing else but a Radeon HD 2900 XT with a modified BIOS, the ATI Radeon HD 2900 Pro is absolutely identical to its progenitor externally.
There is nothing in this card that you can tell a Radeon HD 2900 Pro by. It uses the same PCB, GPU and memory chips as the Radeon HD 2900 XT.
Our sample of ATI Radeon HD 2900 Pro has 512 megabytes of GDDR3 memory in sixteen 256Mb chips from Hynix (HY5RS573225AFP-1, 1000 (2000) MHz rated frequency, 2.2V voltage). Such chips are also installed on the reference sample of ATI Radeon HD 2900 XT 512MB GDDR3. It is possible that some users will get a Radeon HD 2900 Pro, especially in the 1024MB version, with Samsung memory. The memory frequency is lower in comparison with the Radeon HD 2900 XT: 800 (1600) MHz as opposed to 825 (1650) MHz. The reduction of the memory bandwidth from 105.6GB/s to 102.4GB/s can hardly be noticeable in comparison with the considerable GPU frequency reduction. This parameter looks just splendid against the GeForce 8800 GTS with its 64MB/s bandwidth.
The GPU has 320 ALUs grouped into 64 superscalar computing modules with 5 ALUs in each and capable of processing any type of shaders within the Shader Model 4.0 framework. The core also incorporates 4 texture processors (equivalent to 16 TMUs), four raster back-ends (equivalent to 16 ROPs), and a hardware Compositing Engine for building multi-processor CrossFire configurations. Thus, all the subunits of the R600 chip are active and the single difference between the cores of the Radeon HD 2900 Pro and Radeon HD 2900 XT is the clock rate, reduced from 742MHz to 600MHz on the former card. The gaming tests will show us how this reduction affects the card’s performance.
Like its forebear, the ATI Radeon HD 2900 Pro is equipped with two power connectors: a standard 6-pin PCI Express and an 8-pin PCI Express 2.0. You can insert a 6-pin plug into the latter connector, though. The official overclocking tool available in the Catalyst driver is disabled for the card, but you can overclock it using third-party utilities. The GPU working at a reduced frequency, we can expect the card to have a lower power draw – we’ll check it out in the next section.
The Radeon HD 2900 Pro has two dual-link DVI-I ports with HDMI support (including audio-over-HDMI), one universal VIVO port (supported by the Theater 200 processor) and two standard CrossFire connectors.
The card is cooled by a modified version of the reference cooler that exhausts the hot air out of the system case. We described its design in our first review of the ATI Radeon HD 2000 architecture. The modified version has acquired an additional heat pipe for a total of three. This should improve the performance of the heatsink. This modified cooler is installed not only on ATI Radeon HD 2900 Pro but also on all new Radeon HD 2900 XT cards.
As we know from practice, the cooler’s fan has to work at a high speed to keep the temperature of the Radeon HD 2900 XT within reasonable limits in 3D mode and the card gets quite noisy then. The same is true for the ATI Radeon HD 2900 Pro. The card is almost silent in 2D applications, but becomes distinctly audible in 3D mode – it is no better than the ATI Radeon HD 2900 XT in this respect.
We measured the power consumption of the ATI Radeon HD 2900 Pro (we emulated it by lowering the frequencies of a Radeon HD 2900 XT) using a special testbed with a modified Intel Desktop Board D925XCV equipped with measuring shunts. In 3D mode the cards were loaded by the first SM3.0/HDR test from the 3DMark06 suite running in a loop at 1600x1200 with 4x FSAA and 16x AF. The Peak 2D mode was emulated by means of the 2D Transparent Windows test from PCMark05. We got the following results:
Click to enlarge
So, the ATI Radeon HD 2900 Pro consumes about the same amount of power as the senior model. The 8-pin connector supplies most power in 3D mode while the other channels have moderate load. The new card requires about 30W less in 3D mode than the ATI Radeon HD 2900 XT, but its total power consumption is still high. It is as high as that of the Nvidia GeForce 8800 GTX which belongs to the premium product class.
To test the performance of ATI Radeon HD 2900 Pro in popular contemporary applications we assembled the following standard test platform:
Since we believe that the use of texture filtering optimizations is not justified in this case, the ATI and Nvidia graphics card drivers were set up to provide the highest possible quality of tri-linear and anisotropic filtering. Also, to ensure maximum image quality, we enabled transparent texture filtering - Adaptive (super-sampling) antialiasing for ATI Catalyst and Antialiasing – Transparency: multisampling for Nvidia ForceWare. As a result, our ATI and Nvidia driver settings looked as follows:
ATI Catalyst:
Nvidia ForceWare:
For our tests we used the following games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
RPG
Simulators
Strategies
Synthetic Benchmarks
We selected the highest possible graphics quality level in each game using standard tools provided by the game itself from the gaming menu. The games configuration files and Windows registry weren’t modified in any way. Performance was measured with the games’ own tools or, if not available, manually with Fraps utility version 2.9.1. We also measured the minimum speed of the cards where possible.
ATI Radeon HD 2900 Pro will be competing against the following graphics accelerators participating in our test session:
All of them belong to sub-$300 solutions except ATI Radeon HD 2900 XT that was taken as a reference for this test session. It will help us estimate how the performance of R600 solutions drops depending on the GPU frequency reduction.
The ATI Radeon HD 2900 Pro seems to have modest results in this test, but you should keep it in mind that it shares the same price category with the Nvidia GeForce 8600 GTS. Having a comparable price, the AMD solution is much faster and allows playing at a resolution of 1600x1200 with a speed higher than 36fps. So, the only drawback of the Radeon HD 2900 Pro is its low availability in shops. Otherwise, it is a very good choice for gamers who play this popular multiplayer shooter.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
With all its visual splendor BioShock is far from being a killer for today’s graphics cards, at least those of them that are based on the ATI R600 and Nvidia G80 processors. Notwithstanding the highest graphics quality settings all of them gave out a comfortable frame rate for 1280x1024 and an acceptable frame rate for 1600x1200. The GeForce 8800 GTS 320MB can keep the speed above 25fps even at 1920x1200. The Radeon HD 2900 Pro is 20-25% behind the Nvidia card depending on the display resolution.
The hero of this review performs well in this one of the most demanding games of today. It outperforms the Nvidia GeForce 8800 GTS 320MB thanks to the architecture that shows its best in computations-heavy tasks. Few graphics cards priced at less than $300 can boast a comfortable speed in Call of Juarez but the ATI Radeon HD 2900 Pro can do that even at a resolution of 1600x1200, keeping the minimum of speed within acceptable limits. That’s an excellent result.
Unfortunately, we cannot say the same about the Call of Juarez demo that demonstrates new effects created using DirectX 10 capabilities. Well, as we already know, this test is too heavy even for the Nvidia GeForce 8800 GTX. As you could learn from our reviews, the transition to the new API doesn’t go smooth, and even most expensive graphics cards often fail to provide an acceptable speed in available DirectX 10 applications.
The good old Far Cry is not much of a test for the ATI Radeon HD 2900 Pro in any of the standard resolutions. That’s just what you could expect from its full-featured R600 core backed up by a 512-bit memory bus. The GeForce 8800 GTS 320MB offers a higher average frame rate but it is redundant here.
The gap between the ATI Radeon HD 2900 Pro and Nvidia GeForce 8800 GTS 320MB is smaller in the HDR lighting mode, especially on the Research map, and the ATI card delivers a comfortable speed irrespective of the resolution.
The frame rate is fixed at 30fps in this game as this is the rate at which the physical model is being updated at the server. Thus, this 30fps speed is the required minimum for playing the game.
The ATI Radeon HD 2900 Pro seems to cope with the job at 1600x1200, but there is no reserve of speed then. That’s why we recommend you to limit the resolution to 1280x1024, at least if you play with enabled FSAA. Overall, the card is not as good as the Nvidia GeForce 8800 GTS 320MB, but don’t forget that it comes at a lower recommended price.
Having a high average frame rate, the ATI Radeon HD 2900 Pro is inferior to the GeForce 8800 GTS 320MB in terms of minimum speed, which is no less important. The Nvidia card provides a comfortable minimum speed at a resolution of 1600x1200 while the ATI one only offers a speed higher than 25fps at 1280x1024. It is far better than the Nvidia GeForce 8600 GTS at 1280x1024, though.
The ATI Radeon HD 2900 Pro is getting closer to the Nvidia GeForce 8800 GTS 320MB as the resolution grows. The gap shrinks from 25-26% at 1280x1024 to 18% at 1920x1200. You have a comfortable speed in every resolution, so the ATI card isn’t any worse for playing this game. Mainstream graphics cards with a 128-bit memory bus are only good for the resolution of 1280x1024 here.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. This is the reason why we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
The ATI Radeon HD 2000 series performs faster with the new Catalyst but not fast enough to deliver comfortable performance even at 1280x1024. It’s odd that the minimum speeds of the Radeon HD 2900 XT and HD 2900 Pro differ so greatly while they only differ by 25% in terms of average performance. Anyway, the Radeon HD 2900 Pro is not worse than the GeForce 8600 GTS in this test.
This game is no revelation in terms of quality and complexity of its visuals, but none of the mainstream graphics card can even get close to delivering a playable frame rate at the highest settings. As we learned in our review of DirectX 10 games, you can drop some settings from High to Medium to improve performance and lose but little in visual quality.
The Radeon HD 2900 Pro is slightly slower than the GeForce 8800 GTS 320MB and even provides a higher minimum of speed.
We try to get the highest possible image quality from each game, and choosing between FSAA and HDR we preferred the latter. The game was tested with anisotropic filtering only.
The Radeon HD 2900 Pro wins this test just too easily. Delivering the performance of a GeForce 8800 GTS 320MB, the ATI card offers a more advanced architecture at a lower recommended price. The only problem is that it may just be missing at your local shop.
The current version of the game doesn’t support FSAA, so we performed the test with anisotropic filtering only.
The game engine preferring Nvidia’s GeForce 8 architecture, the ATI Radeon HD 2900 Pro cannot rival the GeForce 8800 GTS 320MB here. On the other hand, it provides a comfortable speed at 1280x1024 as opposed to the G84- and RV630-based products.
The game loses much of its visual appeal without HDR. Although some gamers argue that point, we think TES IV looks best with enabled FP HDR and test it in this mode.
The new card from AMD feels at ease in this test, providing a higher minimum of speed than the GeForce 8800 GTS 320MB. This refers to closed environments such as houses, caves, etc. Open scenes of TES IV are far more demanding.
Well, the ATI Radeon HD 2900 Pro has a comfortable speed in open scenes too, even at 1920x1200. Having fewer TMUs, it is behind the Nvidia GeForce 8800 GTS 320MB, though. As you can see, the new Radeon is an excellent and inexpensive choice for playing Oblivion.
The ATI Radeon HD 2900 Pro has a high average speed at every resolution, but it is somewhat slower than the Nvidia GeForce 8800 GTS 320MB in terms of minimum speed, except for 1920x1200. Of course, the Nvidia GeForce 8600 GTS is no match to the new card from AMD in this test.
The game having a frame rate limiter, you should compare the minimum frame rates in the first place because it is the minimum speed that determines your playing comfort in Command & Conquer 3.
As we had expected, the Radeon HD 2900 Pro reached the frame rate limit, ensuring a comfortable minimum speed at every resolution. It looks better in terms of minimum speed than the GeForce 8800 GTS 320MB, not to mention the cards from the ATI Radeon HD 2600 and Nvidia GeForce 8600 series.
The new add-on to Company of Heroes is tested in DirectX 10 mode only since it provides the highest quality of the visuals.
The minimum of speed is low with each of the cards, so neither of them provides really comfortable conditions for play. As for average performance, the ATI Radeon HD 2900 Pro is even slightly faster than the Nvidia GeForce 8800 GTS 320MB at a resolution of 1280x1024, which is the only playable resolution in DirectX 10 mode. The difference of 1.7fps can hardly be spotted with a naked eye, though.
The ATI Radeon HD 2900 Pro scores a mere 787 points less than the senior model, which is a smaller difference than in most of the gaming tests if expressed in percent. That was to be expected because 3DMark defaults to 1024x768 resolution without full-screen antialiasing, while these two cards only differ in the graphics core frequency.
The gap between the Radeon HD 2900 Pro and Radeon HD 2900 XT increases back to normal in the gaming tests. Compared with the GeForce 8800 GTS 320MB, the ATI Radeon HD 2900 Pro is 8-10% slower in the second test, which is not sensitive to the texture processors’ performance, but 30-40% slower in the first and third test.
Considering the results we got at the default settings, it is not the shader processors’ fault. The memory subsystem can’t be a problem either since it is superior to the memory subsystem of the GeForce 8800 GTS 320MB. So, the new card is slower in the individual tests due to its low core frequency as well as to having only 16 TMUs and ROPs. The peculiarities of the VLIW architecture may have affected its results as well.
The overall scores are closer to reality in 3DMark06 which defaults to 1280x1024 and includes more complex tests than those of 3DMark05. Using Shader Model 3.0, its tests resemble modern 3D games.
The ATI Radeon HD 2900 Pro is about as fast as the Nvidia GeForce 8800 GTS 320MB in the SM2.0 tests. The SM3.0/HDR tests are computations-heavy, allowing the ATI card to show its best thanks to 320 ALUs and a more advanced architecture overall.
The Radeon HD 2900 Pro is worse than the GeForce 8800 GTS 320MB in both of the SM2.0 tests: by 15% in the first and by 22% in the second test.
The Radeon HD 2900 Pro almost overtakes the GeForce 8800 GTS 320MB in the first of the SM3.0/HDT tests, reducing the gap to 5-6%. And then the two cards are equals in the second test that doesn’t produce a high texture-mapping load. These results differ from the results of the groups of tests probably due to the use of FSAA which means an increased load on the rasterization subunits.
The ATI Radeon HD 2900 Pro is undoubtedly a success. Priced at only $249 it provides an acceptable speed in most of our tests as opposed to graphics cards with the mainstream GPUs RV630 and G84. We really doubt that even a Radeon HD 2600 X2 would match the performance of the new card because it would be inferior to the Radeon HD 2900 Pro in the amount of ALUs and raster back-ends (240 and 8 against 320 and 16, respectively) as well as in memory subsystem performance. We don’t even count in the possible compatibility problems the dual-chip solution is sure to have.
Considering the imminent arrival of graphics cards with ATI’s new RV670 core, there is no sense in producing the Radeon HD 2600 X2 in mass quantities. The Radeon HD 2900 Pro, even in small quantities, can successfully fill in the gap between the Radeon HD 2900 and HD 2600 families. It fulfils this purpose even better than the sophisticated Radeon HD 2600 X2 with all its potential problems. Well, we’ll obtain one of those unique dual-chip graphics cards in near future and perform a comparative test.
As for drawbacks, the ATI Radeon HD 2900 Pro has them just like any other product, but the most important one is the low availability. As we wrote above, this card is going to come to market in hardly more than 10 thousand copies, and many people just won’t have a chance to buy one. This may also affect the retail price of the card. Besides, the card has all the drawbacks of the ATI Radeon HD 2900 XT such as high power consumption, a high level of noise in 3D mode, a bottleneck of 16 TMUs (which are also clocked at a reduced frequency now), and architectural peculiarities that make the card’s performance dependent on driver optimizations.
And still, with all these minor problems, the ATI Radeon HD 2900 Pro is perhaps the best graphics card in the $249 price category – if you are lucky to find it selling for this price!
Highs:
Lows: