by Alexey Stepin , Yaroslav Lyssenko
09/22/2009 | 09:03 PM
Time is passing by and former ATI Technologies that is currently a division of Advanced Micro Devices doesn’t seem to be slowing down the tempo that they set back in late June of 2008 when they launched RV770 graphics processor. Back then this GPU almost completely transformed the general idea of the pricing and potential of gaming graphics accelerators. ATI’s eternal rival, Nvidia, was momentarily dethroned and started losing their market positions immediately. They managed to really strike back only once since then: by launching a refreshed modification of the dual-processor GeForce GTX 295 graphics adapter, which we have already told you about in one of our previous reviews.
This is a perfect example how successfully chosen strategy may provide a long-term advantage, and the arguable one cause numerous disappointments. As is known, AMD/ATI bet on pretty high-performance, but not very complex and expensive mainstream graphics processor, as the highest demanded solution in the market, and Nvidia, on the contrary, decided to play its cards with the most powerful monolithic chip. We all know what the outcome of these strategic decisions was. ATI Radeon HD 4800 family rapidly became extremely popular due to the highest performance at a very acceptable price point, and the absence of a super-powerful single-processor solution in AMD’s lineup was easily compensated by combining two RV770 on a single card.
At the same time, Nvidia had to rush the GeForce GTX 200 prices down, simplify the PCB layouts and urgently transfer G200 to a finer manufacturing process in order to make up for the possible losses resulting from the arguable strategic decision. Finally, ATI’s main competitor managed to stop the leakage of the market share, but at the point there was a new wonder-weapon being prepared deep inside AMD labs. Right from the start it was intended to strike a really crushing blow, namely, to launch the world’s first graphics processor with DirectX 11 support.
Moreover, they were not going to veer away from the general strategy: the main striking force was supposed to be the relatively inexpensive graphics processor codenamed Cypress. Just like with RV770, the tandem of two graphic cores like that codenamed Hemlock will be responsible for the breakthrough in the high-performance segment. And in the performance, mainstream and entry-level segments we will see other cores called Juniper, Redwood and Cedar, respectively.
Today is a very important day for the Advanced Micro Devices graphics division: for the first time they are going to introduce to the public their main weapon in the upcoming battle for the DirectX 11 compatible graphic card market.
Will Cypress prove up to the expectations? We will try to answer this question in reference to the performance in actual currently existing games. Unfortunately, there are no applications that would take advantage of the DirectX 11 functionality that is why it is hard to tell if the new ATI graphics processor will be an excellent solution for the today’s infrastructure or will most likely become the basis for AMD’s upcoming graphics products.
The new ATI Radeon HD series has valid claims to the number 5 in the generation number but of course, like every modern GPU, the RV870 has inherited a lot of traits of its predecessors RV770 and RV790. Thus, the new generation of ATI top-performance graphics cards has got the name of Radeon HD 5800. At the time of the announcement there are two products in it: the flagship Radeon HD 5870 that enables all of the RV870’s resources and the less advanced Radeon HD 5850 that seems to be based on those graphics processors that did not pass the frequency check and/or have a few defective subunits. The recommended price of the new products is $399 and $299, respectively. Let’s compare their specs with those of the best solution from the previous generation.
Click to enlarge
First off, the new GPU developed by ATI is mind-bogglingly complex incorporating over 2 billion transistors. At this scale, even Nvidia’s G200 looks like an example of simplicity. What is even more incredible, the senior model’s core is clocked at 850MHz. These miracles have been made possible by the 40nm tech process without making the core too large. The RV870 is even smaller than the G200b!
The memory subsystem specs are also impressive due to chip memory subsystem optimization and improving of the GDDR5 frequency specifications. ATI Radeon HD 5850/5870 has lower peak memory bus bandwidth than the GeForce GTX 285 because of the relatively narrow 256 bit bus. Perhaps the single Radeon HD 5870 will be inferior to dual-GPU solutions of the previous generation in highly demanding games and at extremely high display resolutions but this supposition is yet to be confirmed in practical tests. As for the previous-generation single-chip flagship solution, the improved internal structure of the newcomer has every chance to leave it far behind.
Then, the Radeon HD 5870 is the same as the Radeon HD 4870 X2 in terms of potential computing power as well as texture and raster processors, but its GPU works at a higher frequency. Theoretically, the new flagship’s performance is going to match that of the Radeon HD 4870 X2 or be slightly higher because of internal chip optimizations.
Besides the obvious architectural improvements, the new ATI Cypress graphics chip has some principal differences from the predecessors: this is the world’s first graphic core to support DirectX 11. Considering the upcoming Windows 7 release, this API is likely to become a de facto industry standard, so AMD has ensured an advantage over Nvidia that does not yet offer any DirectX 11 compatibles. Besides a few obvious innovations in the 3D part, such as tessellation and more advanced pixel or vertex shaders, DX11 has a number of other new features:
Although Nvidia suggests that DirectX 11 support will not be a decisive factor in favor of a new graphics card (see this story for details), but it really looks like a poor excuse to us.
Summing it up, the new ATI Radeon HD 5000 graphics card family boasts huge potential indeed. But before we proceed to measure its extent, we will scrutinize the RV870 architecture for any changes over the RV770/790.
The general principle of the computing section has not changed much in the RV870. It is still based on shader processors with superscalar design, each processor incorporating five ALUs four of which are general-purpose ALUs and the fifth is a special-purpose ALU capable of executing complex instructions like SIN, COS, LOG, EXP, etc. Besides the ALUs, each shader processor also contains a branch control unit and an array of general-purpose registers.
When we are talking about 1600 stream processors in the RV870, we must keep it in mind that there are actually 320 rather complex 5-stage computing subunits. Provided sufficient code optimization, this design of the GPU’s computing section helps achieve a much higher level of performance than with Nvidia’s scalar architecture. The design of the shader processors and task scheduler has been improved in the new GPU to support the new DirectX 11/DirectCompute 11 capabilities.
As before, the RV870’s shader processors are grouped into SIMD cores with 16 processors in each core, but there are now 20 rather than 10 such cores in the GPU.
Each core is serviced by dedicated logic and has four texture processors and a L1 cache at its disposal. Thus, the amount of texture processors in the RV870 is doubled (from 40 to 80 TMUs). The peak texture sampling performance has doubled, too. The overall architecture of texture processors seems to have been left largely intact. Each of them still consists of 16 FP32 texture fetch units, four address units and four filters. However, with the introduction of DirectX 11 support these processors have acquired support for 16Kx16K-pixel textures, new HDR texture compression modes, Gather technology for nearest textures sampling acceleration, etc. There is also a new anisotropic filtering algorithm that delivers the same high filtering quality irrespective of the angle of inclination of the filtered surface.
The computing cores can communicate on both local and global levels. ATI claims a considerable increase in cache bandwidth. Particularly, the speed of fetching data from the L1 cache is now as high as 1 terabyte per second while the bandwidth of the link between the L1 and L2 caches is increased to 435GBps. The L2 caches have become larger from 64 to 128KB. The ratio of computing to texture-mapping resources has not changed and is still 4 to 1.
According to the developer, the peak computing power of the RV870 is as high as 2.7 teraflops in single-precision mode (FP32) and 544 gigaflops in double precision mode (FP64) which is used for most serious computing tasks. A special mention must be made of the ability to execute threads in protected memory sections which makes it easier to transfer code originally developed for the classic CPU to the GPGPU platform. All these innovations in the RV870’s computing section make it a perfect choice for GPGPU, especially in comparison with Nvidia’s solutions whose double-precision performance is far from ideal.
As for the memory controller, AMD’s graphics department has finally given up the ring-bus topology in the RV870.
Instead, they tried to achieve higher memory frequencies by introducing Error Detection Codes (EDC) and temperature compensation of the frequency of GDDR5. As a result, the memory is stable at frequencies up to 1250 (5000) MHz. This approach has helped increase the memory subsystem performance while keeping the PCB design simple, which has a positive effect on the manufacturing cost of the Radeon HD 5800.
ATI claims that the crossbar memory controller they use again will work just as efficient as the ring-bus one that was employed in the last three generations and was first introduced in ATI R520/Radeon X1800. Keeping in mind that the developer has practically doubled (or maybe increases even more) the size of all on-chip caches and used higher memory clock speed, we shouldn’t expect the RV870 efficiency to be any lower because of the memory controller topology.
Nevertheless, despite the doubling of the peak computational power together with doubling of the rasterization and texturing speeds, the memory bandwidth got only 30% higher. Therefore, we are a little concerned that the graphics processor won’t show its maximum performance in “extreme” conditions.
The amount of rasterization processors (AMD/ATI calls them render back-ends) has also been increased twofold, from 16 to 32. The performance of the rasterization subsystem has grown up accordingly:
The RBE architecture has been improved, too, but we don’t have any reliable info on this point. We only know that the GPU’s performance at full-screen antialiasing has been increased and it is now more efficient working with the Z-buffer/stencil buffer. The RV870’s RBEs support both classic antialiasing modes and programmable CFAA modes among which the Edge Detect mode is the most interesting. It delivers an ideal quality of antialiasing without any fuzziness typical of the CFAA Wide/Narrow Tent modes.
Before the arrival of the eighth generation of ATI Radeon GPUs, the hardware tessellation unit was perhaps the most questionable spot in the entire ATI Radeon HD series. Notwithstanding its support for both vertex and geometric shaders, it was never popular among game developers. The RV870 has got wider tessellation capabilities within the DirectX 11 framework. Two new shader types are support: hull and domain shaders. Tessellation itself can be performed in different ways including N patches, Bezier patches, displacement maps, etc.
There are some other interesting DirectX 11 features, like true multithreading, Order Independent Transparency which improves the quality of overlapping transparent objects by using DirectCompute 11, and the improved post-effects such as depth of field, etc. Besides, Nvidia is no more the exclusive provider of Ambient Occlusion technology. The new-generation Radeon HD supports High Definition Ambient Occlusion with improved shadowing quality.
As opposed to DirectX 10.1, DirectX 11 is going to have a warm welcome from game developers. A few projects will get support for the new API in Q4 of this year, with more such projects to come in 2010. This should give some thinking matter to Nvidia, the main and only opponent to AMD on the 3D graphics market.
The display controller has not been revised for the RV870. Instead, it has been developed from scratch.
The new core offers very flexible display connection opportunities. Theoretically, it can output up to six video streams with 10-bit-per-component precision. Besides delivering them as independent content, the GPU can output a single picture at a huge resolution up to 5760x2160. It is supposed that you can actually play a game at that resolution, but we doubt the Radeon HD 5870, for all its impressive hardware resources, will be able to deliver a playable frame rate in such an extreme display mode. And we guess even building a CrossFireX subsystem out of four Radeon HD 5870 cards won’t help achieve a playable speed in Crysis Warhead at 5760x2160. Anyway, you will have to wait for a special version of the card, called Radeon HD 5870 Eyefinity6 Edition, that is equipped with six DisplayPort connectors in order to check this out in practice. The ordinary Radeon HD 5870 supports up to three monitors simultaneously which might be useful in some games like flight or racing simulators.
AMD’s PowerPlay technology has been improved, too. The peak power draw is declared to be only 188 watts. The new cards’ power consumption in idle mode is a mere 27W. The numbers look fantastic considering the tremendous complexity of the RV870 chip and need to be checked out like any other marketing claim. So, let’s get to practical tests now. First, let’s take a closer look at the graphics card itself.
The Radeon HD 5870 is nothing extraordinary on the outside even though it does look cool and stylish due to the new black casing of the cooling system. The narrow red strip crossing it is a nice touch in the product appearance and a point for AMD’s designers.
The back part of the card is covered by a metallic plate which does not perform any function because there are no elements requiring much cooling there. It is simple to dismantle the cooling system: you only have to undo the screws on the reverse side of the card and the couple of screws on the mounting bracket. And then you will see the following:
The Radeon HD 5870 has a surprisingly simple PCB design for a graphics card with such advanced specs as it has. The back part where we usually find GPU and memory power circuitry looks somewhat empty. At least, there is a lot of free room around the Volterra VT1165MF controller. The GPU voltage regulator has four phases, and there are seats for elements of a fifth phase on the PCB.
The two top phases with single inductors and other power packs seem to relate to the memory voltage regulator. One more Volterra VT1165MF controller is located in an unusual place near the CrossFireX connectors. It seems to be responsible for powering the card’s memory.
The Volterra chips provide for a more flexible power management by ATI PowerPlay technology. Some AMD partners will probably offer the user the opportunity to increase the GPU and memory voltages from software. Overclockers would appreciate this feature as they wouldn’t have to volt-mod the card in order to achieve the same result. There are two 6-pin PCIe 1.0 power connectors on the PCB which indicates a rather low power draw of the RV870. However, the installation of an 8-pin PCIe 2.0 connector is possible.
The card carries eight GDDR5 memory chips from Samsung (K4G10325FE-HC04) with a capacity of 1Gb (32Mb x 32), voltage of 1.5V, and a rated frequency of 1250 (5000) MHz. They work at a reduced frequency of 1200 (4800) MHz on the Radeon HD 5870. Coupled with the 256-bit memory bus, the card’s memory bandwidth is 153.6GBps. That’s quite high in comparison with previous-generation single-chip cards but the dual-chip Radeon HD 4870 X2 and GeForce GTX 295 have a memory bandwidth of 230.4GBps and 224GBps, respectively. It means the Radeon HD 5870 may be slower in some resource-consuming games like Crysis at high display resolutions.
The new chip resembles the R600 externally because its die, like the R600 die, is turned around by 45 degrees. Despite the horrendous complexity, it is rather small at only 334 square millimeters, thanks to 40nm manufacturing process. The die is protected with a metallic frame on the GPU package. The marking says this sample of RV870 was manufactured rather recently, on the 33rd week, i.e. in mid-August, of 2009.
The new version of GPU-Z knows about the RV870 and reports the same info about the Radeon HD 5870 as AMD’s official specs. The core indeed has 1600 ALUs grouped into 20 SIMD cores, 80 texture processors, 32 raster back-ends and is clocked at 850MHz.
The ordinary Radeon HD 5870 can output video content to three monitors simultaneously, so it has two DVI-I ports, one HDMI and one DisplayPort. The rest of the mounting bracket is perforated for ventilation. The Radeon HD 5870 supports CrossFireX and offers two appropriate onboard connectors. Theoretically, you can unite four such graphics cards into a single graphics subsystem that would deliver unprecedented performance in modern and upcoming games.
The cooling system of the Radeon HD 5870 is a time-tested, simple and clever solution that has been used by both GPU developers for a long while. In other words, it is a rather large aluminum heatsink connected with heat pipes to a copper base that contacts with the GPU die. There are four heat pipes here which should ensure uniform distribution of heat.
As usual, the aluminum frame the heatsink and fan are installed on takes heat off the memory chips and power transistors of the voltage regulator. Elastic pads ensure proper thermal contact and a layer of traditional dark-gray thermal grease is used for the GPU. The heatsink is cooled by a blower with red impeller that becomes very loud at increased speed.
The photo shows the air duct for some of the airflow to move at an angle of 90 degrees to the mounting bracket.
Thus, only part of the hot air leaves the cooling system in the ordinary way. The other part is blown into the system case through the slits in the cooler casing:
This is an unavoidable compromise because the Radeon HD 5870’s mounting bracket is populated more densely than those of previous-generation Radeon HD cards and does not have much place for vent holes.
The whole contraption is fastened to the PCB with a few screws. An X-shaped back-plate ensures proper contact between the GPU die and the cooler’s heat-exchanger. As we’ve mentioned above, the metallic plate that covers the back of the PCB is a purely decorative element and has no practical function. There is nothing to cool there.
Summing it up, there are no special innovations in the design of the cooling system installed on the Radeon HD 5870. It is a rather simple and long-known design that has repeatedly proved its worth in real applications. The question is how hot the RV870 is going to be. Even though manufactured on 40nm tech process, it consists of over 2 billion transistors. Doesn’t the cooler get too loud maintaining a safe GPU temperature? We’ll check this out in the next section.
The new ATI Cypress core is unique in its own ways. Particularly, it is terrifically complex, so we are highly interested in its power consumption properties. We measured the card’s power draw on a testbed configured as follows:
Following our standard procedure, the 3D load was created by the first SM3.0/HDR test from 3DMark06 running in a loop at 1600x1200 with forced 4x FSAA and 16x AF. Additionally, we used OpenGL FurMark. The 2D load was emulated by the 2D Transparent Windows test from PCMark05. We’ve got the following results:
So, the power draw is not higher than 107 watts in 3DMark06 which is amazing for a graphics card whose GPU consists of over 2 billion transistors. Nvidia’s G200b can’t even dream of such power efficiency. However, this benchmark is already obsolete and we only use it as we’ve accumulated a large amount of data on graphics cards from previous generations. FurMark provides a different picture. It makes full use of all GPU resources and the new card’s peak power consumption is as high as 160 watts. On the other hand, FurMark is far more intensive than real-life games, so the practical power draw is going to be closer to the first number than to the second. The Radeon HD 5870 is also extremely economical in 2D mode, setting a new record for its class.
The cooling system did well despite our apprehensions.
The temperature of the new Radeon HD 5870 is far from dangerous. The GPU was no hotter than 75°C even during long test sessions. This is all due to the 40nm tech process the RV870 Cypress is manufactured on.
The reference point for our noise measurement tests is 43dBA which is the level of ambient noise in our testbed as measured at a distance of 1 meter from the testbed with a passively cooled graphics card inside. When we installed our Radeon HD 5870, we got the following results:
The card is actually silent in idle mode, but gets loud after working long under load. However, it is never louder than the Radeon HD 4870 X2 which is the loudest gaming graphics card of all, according to our tests. The high level of noise is the payment for the high cooling efficiency and low GPU temperature. The aerodynamics of the reference Radeon HD 5870 cooler is far from ideal, too, adding to the noise. The results are acceptable overall, but we will surely see quieter versions of Radeon HD 5870 equipped with nonstandard cooling solutions.
The Radeon HD 5870 was no good at overclocking. We quickly saw image artifacts even after a 30MHz increase in the GPU frequency. Our card did pass 3DMark Vantage at 900MHz GPU and 1300 (5200) MHz memory, but the visual artifacts negate that achievement. Perhaps we failed because we had dismantled the card and replaced its thermal grease, so we will check out the overclocking potential of the Radeon HD 5870 again in our upcoming reviews. Right now, let’s benchmark its performance in games.
We are going to investigate the performance of Radeon HD 5870 graphics card using the following universal testbed:
The graphics card drivers were configured in the following way:
The list of benchmarks includes the following gaming titles and synthetic tests:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the ordinary user doesn’t have to know how to do it. We made a few exceptions for selected games if that was necessary. We are going to specifically dwell on each exception like that later on in our article.
Besides Radeon HD 5870 we have also included the following graphics accelerators to participate in our test session:
We ran our tests in the following resolutions: 1280x1024, 1680x1050, 1920x1200 and 2560x1600. Everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.
Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 2.9.8. In the latter case we ran the test three times and took the average of the three for the performance charts.
Starting from version 1.3 the game allows to record and reproduce a demo. Unfortunately, this method does not report the bottom frame rate.
The new card has a good start, proving that new-generation single-chip solutions are quite competitive to dual-chip cards of the previous generation. The Radeon HD 5870 is either ahead of the Radeon HD 4870 X2 or comparable to it. It also looks good against the GeForce GTX 295 except for 2560x1600 where Nvidia’s solution is obviously faster. Anyway, the Radeon HD 5870 makes the game playable at that resolution, too, while being quieter than the Radeon HD 4870 X2 and considerably more economical.
GPU generations come and go but the dream of a comfortable speed in Crysis at the highest graphic quality settings remains a dream. For all its 1600 ALUs, 80 texture and 32 raster processors, the Radeon HD 5870 gives up even at 1680x1050, being unable to maintain a comfortable bottom speed. We have to wait for a Radeon HD 5870 X2. Take note that the GeForce GTX 295 is somewhat faster at 1280x1024, but this makes no practical difference between the two cards.
We disabled the integrated frame rate limiter in the game console for the sake of comparing the cards. The game’s built-in benchmarking options do not provide information about the bottom speed, so there is no such info in the diagrams.
The new Radeon HD 5870 is somewhat better than its dual-processor predecessor in this game, but slower than the GeForce GTX 295 at low resolutions. There is no point in making comparisons here, though. The game is perfectly playable at any resolution because each card is faster than the integrated frame rate limit set at 30fps.
The new card was unable to outperform the GeForce GTX 295 but was much better than the Radeon HD 4870 X2, especially at 1920x1200 where its bottom speed is higher. Owners of large monitors or GeForce GTX 295 cards should not worry about upgrading, and the GeForce GTX 285 provides a larger reserve of speed, yet AMD’s effort cannot be underestimated: the RV870 is indeed equal to two RV770 chips!
The game runs on the Source engine and has an integrated benchmark, but the latter does not report the bottom speed information.
The Radeon HD 5870 goes ahead of the Radeon HD 4870 X2 starting from the resolution of 1680x1050. The gap is growing larger, reaching an impressive 33% at 2560x1600! The increase in average performance from 80 to 110fps cannot be noticed in practice, yet AMD’s achievement is indisputable.
To achieve a playable speed in this game we disabled FSAA and such resource-consuming options as Sun rays, Wet surfaces and Volumetric Smoke. We use the Enhanced full dynamic lighting (DX10) mode for our test and additionally enable the DirectX 10.1 mode for the ATI cards.
The new Radeon HD 5870 is not too brilliant in comparison with the GeForce GTX 295, but the gap is small at 2560x1600, the AMD solution even providing a higher bottom speed. That’s an impressive result for a single-chip graphics card.
The Radeon HD 4870 X2 retains its leading position in this game. The new-generation Radeon HD competes with the GeForce GTX 295 and loses at low resolutions. The gap is smaller at 1920x1200 and 2560x1600 while the overall level of performance is very high. An extra 5-8fps can make no difference if the cards are as fast as 130fps or something.
The Radeon HD 4870 X2 keeps its first place while the Radeon HD 5870 is fighting with the GeForce GTX 295. The newcomer gains the upper hand at 1680x1050 and 1280x1024 and loses at 1920x1200 and 2560x1600, although keeps the frame rate high enough for comfortable play. Considering the lack of typical drawbacks of multi-GPU solutions, the Radeon HD 5870 is the obvious leader in the top-performance class.
The Radeon HD 4870 X2 is somewhat ahead at 2560x1600 but the Radeon HD 5870 is so close behind that there is no practical difference. The newcomer has the same result as the GeForce GTX 295 at that resolution.
The Radeon HD 5870 has no rivals in the new game from the Need for Speed series. It enjoys a huge lead at every resolution including 2560x1600. This may be due to the driver, but it is also clear that the game does not run well on multi-GPU solutions as you can see by comparing the results of the GeForce GTX 295 and GTX 285 as well as of the Radeon HD 4870 X2 and HD 4890. Being a classic single-chip solution, the Radeon HD 5870 does not care about that and turns in an indisputable win.
We use the in-game benchmarking tools that do not allow to measure the bottom frame rate. We also enable DirectX 10.1 support for ATI’s solutions.
The game is incompatible with Adaptive AA algorithms utilized by AMD’s solutions and their performance plummets when this antialiasing option is turned on. However, we can see that the Radeon HD 5870 is almost as fast as its dual-chip predecessor.
We enabled DirectX 11 support for Radeon HD 5870. We have also taken a few screenshots from the game for your reference.
High level of detail
Very High level of detail
We couldn’t notice any significant visual differences. Moreover, the difference between High and Very High detail modes is really minor.
The Radeon HD 5870 is third behind the Radeon HD 4870 X2 and GeForce GTX 295 but all of them have a low bottom speed even at 1280x1024. Oddly enough, it is the GeForce GTX 285 that has the best bottom frame rate and ensures the most comfortable gaming experience.
The Radeon HD 5870 falls farther behind the Radeon HD 4870 X2 as the resolution grows up, but ensures a comfortable play up to the resolution of 1920x1200, just like the previous-generation flagships. But the new card does not depend on the software multi-GPU support and consumes much less power.
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all display resolutions to provide a full picture.
It is good that the new card is better than the Radeon HD 4870 X2 because in 3DMark Vantage it cannot rival Nvidia’s products which support PhysX. The Radeon HD 4870 X2 is obviously an outdated solution now. It has played its part in the history of desktop 3D graphics and has to leave the scene now.
The gap between the Radeon HD 5870 and Radeon HD 4870 X2 is getting smaller as the resolution grows up in the first test. In the second test, the gap is getting bigger. Note that the advantage of the GeForce GTX 295 is much smaller in the second test, especially at high resolutions.
History tends to repeat itself at newer levels. Not so long ago AMD focused on rather simple and inexpensive GPUs and was right: Nvidia with its monstrous G200 had to assume a defensive position after a series of marketing blows from the former ATI Technologies. In the sector of top-end solutions AMD offered dual-chip models and was right once again because the Radeon HD 4870 X2 has remained an unrivalled performance leader for a long time. And what do we have now?
Ironically, ATI actually went Nvidia’s way in developing its new-generation GPU and has created a chip consisting of over 2 billion transistors. But while Nvidia’s G200 was based on rather inappropriate 65nm technology, the RV870 Cypress is manufactured on advanced 40nm tech process. Nvidia’s solution was too progressive for its time and brought a lot of problems to its developer whereas the RV870 is most appropriate. It comes when the production of such complex chips has become not only possible but also economically justifiable. Being 50% more complex than the Nvidia G200, the new GPU from AMD is 1.4 times smaller even than the 55nm version of the G200! As we also know now, the RV870 Cypress is the world’s first GPU to support new DirectX 11 graphics standard, while Nvidia’s products do not offer such support as yet. So, instead of steady evolution of the Radeon HD series, the new card is a revolutionary breakthrough that has no counterparts in the industry at the moment.
The first sample of the graphics card based on ATI new GPU leaves a very positive impression. The Radeon HD 5870 is economical and has good acoustic specs. And it is indeed capable of competing with the previous-generation flagship, the dual-chip Radeon HD 4870 X2, and even beating it in real-life applications. Let’s look at the detailed picture.
There is usually no point in comparing such top-end graphics cards at 1280x1024, yet the Radeon HD 5870 is ahead of the Radeon HD 4870 X2 by an average 3%, being much slower in two tests only, Street Fighter IV and BattleForge. The new card from AMD is also good against the GeForce GTX 295, scoring 4 wins and one draw against 8 losses among which only two are serious losses. However, the loss in Enemy Territory: Quake Wars makes no practical difference while in Tom Clancy’s H.A.W.X. you can just turn off Adaptive AA and boost the frame rate.
The maximum lead over the Radeon HD 4870 X2 is 75% and can be observed in NFS: Shift where the Radeon HD 4870 shows the highs of classic single-chip architecture that does not depend on software optimizations. The new card is slower than the GeForce GTX 295 by an average 3%, Tom Clancy’s H.A.W.X. being the only test where the gap is really large.
The picture does not change much at 1920x1200: the Radeon HD 5870 still beats the Radeon HD 4870 X2 in most of our tests. But it also beats the GeForce GTX 295 in six tests and loses seriously in Clancy’s H.A.W.X. only. And that loss is not due to any deficiencies of the Radeon HD architecture as we have explained above.
When it comes to 2560x1600, the Radeon HD 5870 is still good, but inferior in many tests to the Radeon HD 4870 X2 because of the lower memory subsystem bandwidth. The gap is not large, however, and the new card’s frame rate is still comfortable except in such demanding games as Crysis Warhead. The same goes for the comparison of the Radeon HD 5870 with the GeForce GTX 295. Anyway, the new ATI card seems to be a better buy. It is more economical than the dual-chip flagships of the previous generation and is free from their main drawback – the dependence on software support for multi-GPU technologies. The Radeon HD 5870 always delivers its full speed and unlike Radeon HD 4870 X2 it will never experience a situation when half of its computational capacities are idling. Moreover, the new solution supports DirectX 11, which is a great reserve for the future.
So, the classic single-chip architecture has returned with new capabilities and new level of performance, making the Radeon HD 4870 X2 obsolete. Nvidia finds itself lagging behind once again. ATI Radeon HD 5870 is undoubtedly the best single-chip gaming graphics card today. It is as fast as dual-chip solutions of the previous generation but is free from their drawbacks. It also supports the new industry standards DirectX 11 and DirectCompute 11. Moreover, AMD will evidently not stop there and following their strategy will eventually announce a really monstrous Radeon HD 5870 X2 which performance we will only be able to estimate when we get a second sample of Radeon HD 5870.