by Alexey Stepin , Yaroslav Lyssenko
02/07/2008 | 01:04 PM
As we’ve repeatedly pointed out in our earlier reports, the leading GPU developers ATI and Nvidia take significantly diverse approaches to designing new GPUs. ATI Technologies, currently the graphics division of AMD, has long been putting an emphasis on the speed of math1ematic computations and considering other parameters such as texture sampling and rasterization performance as of secondary importance. This developer had been sticking to the 3:1 concept until the arrival of unified-architecture cores as this ratio of math1ematical instructions to texture operations was thought optimal.
Of course, the ratio of 3:1 was not a developer’s whim but the result of the analysis of many factors, particularly that a lot of gaming projects were developed for several platforms at once. Game consoles had limited amounts of graphics memory and game developers had to use low-quality textures, putting a stress on visual effects created by means of math1ematics-heavy shaders. The product of this development approach was the ATI R580 core that contained 48 pixel shader processors and only 16 TMUs, yet became the best GPU with a non-unified architecture. R580-based graphics cards, ATI Radeon X1900 XTX and later ATI Radeon X1950 XTX, deservedly held the title of the highest-performance solutions with DirectX 9.0c support.
It is also then that the drawbacks of the 3-to-1 concept became apparent because high display resolutions and full-screen antialiasing were widely used in the PC world. Moreover, shaders used to imitate materials contained a lot of texture sampling operations anyway, which put a load on the TMUs and memory subsystem. Nvidia responded to ATI’s success with the release of the G71 core that differed from the previous G70 in thinner tech process and higher clock rates. Nvidia’s combination of 24 pixel shader processors per 24 TMUs proved to be no less, and sometimes even more, efficient than ATI Technologies’ 48/16 formula.
The computing capacity of GPUs has increased considerably with the arrival of the new generation of graphics cores with unified architecture. Execution processors now have to process three types of shaders, and it is impossible to tell the ratio of computing units to texture sampling and filtering units. The performance of streamed processors can only be estimated by indirect evidence now.
Today we’ll discuss the GeForce 8800 GTS graphics card, currently the most math1ematically advanced solution from Nvidia.
In our review of the Nvidia G92 GPU we noted that the new chip seemed to contain more functional subunits than were declared for the Nvidia GeForce 8800 GT graphics card. This would help the developer easily create a product for the higher price niche than the GeForce 8800 GT 512MB to close the gap between the latter and the GeForce 8800 GTX. It could be done by unblocking the originally disabled eighth module with 16 shader processors and 4 TMUs to get a total of 128 processors and 32 TMUs. That’s the way the GeForce 8800 GTS 512MB has come up.
Click to enlarge
Click to enlarge
You can see it only differs from the earlier G92-based products in having 128 active shader processors and 32 TMUs rather than 112 and 28, respectively. It also has higher clock rates. The amount of rasterization units has remained the same at 16.
As you know, the GeForce 8800 GT 512MB proved to be as fast as to challenge the GeForce 8800 GTX in some tests. It made the weaker GeForce 8800 GTS 640MB and GST 320MB cards obsolete and their production was discontinued. However, Nvidia has decided to reuse the suffix GTS even though the new graphics card is going to be superior to the GeForce 8800 GT 512MB and deliver about as high performance as the GeForce 8800 GTX.
This similarity of names may be somewhat confusing for inexperienced users especially if the seller has got old models of GeForce 8800 GTS. You should be careful when shopping because the new and old versions of Nvidia GeForce 8800 GTS are absolutely different products with a huge difference in performance.
The cards resemble each other externally, so you should look at the amount of graphics memory: the old, G80-based, card comes with 320 or 640 megabytes of memory and a 320-bit memory bus whereas the new model, based on the 65nm G92 chip, carries 512MB of onboard memory and has a 256-bit memory bus.
The official price of the GeForce 8800 GTS 512MB is within $299-349 but the potential deficit of G92 cores may pump the price up for a while. Pre-overclocked versions are going to be even more expensive but today we are going to talk about Leadtek’s version of the card that has standard clock rates.
The box with the WinFast PX8800 GTS 512MB resembles the box of the WinFast PX8800 GT Extreme we described in an earlier review. It is painted the same colors and has a picture of a robot, too. Perhaps not very original, this design is attractive anyway.
Product specs are printed on the box but the mention of PCI Express 2.0 support is missing although we know that not all G92-based graphics cards are compatible with mainboards that have PCI Express older than version 1.1. The number that indicates the amount of graphics memory is emphasized with yellow color so that you didn’t make a mistake when shopping.
The contents of the box are packed meticulously: the graphics card is fixed in pieces of foam rubber to avoid damage during transportation. Discs with drivers and software are placed into a small “Utility” box together with a user manual. The card comes with the following accessories:
The accessories are the same as included with the WinFast PX8800 GT Extreme. They are good overall especially with a copy of Neverwinter Nights 2, a highly popular, even though not very new, RPG. The packaging should be appreciated for its reliability if not for design.
High-performance solutions from Nvidia used to be manufactured at contracted facilities at the company’s request and GeForce 8800 GTX/GTS cards could not come out with a unique PCB design. It’s different with G92-based products. We have already seen versions of GeForce 8800 GT based on non-reference PCBs, particularly from Gainward, but the first wave of GeForce 8800 GTS 512MB is going to be mostly based on the reference design. The Leadtek WinFast PX8800 GTS 512MB is a typical representative of this first wave:
The card looks more imposing than the Nvidia GeForce 8800 GT thanks to the cooler that resembles the one that used to be installed on GeForce 8800 GTX and old models of GeForce 8800 GTS. The cooler has dual-slot form-factor, of course. The card has the same PCB as the GeForce 8800 GT, though:
The GPU of the Nvidia GeForce 8800 GTS 512MB containing more active subunits and working at higher clock rates, the power circuit has been reinforced and now has all the four phases working. The GPU and memory are powered via Primarion PX3544 and Intersil ISL6549CBZ controllers, respectively. The power connector is a standard 6-pin PCI Express 1.0 plug with a load capacity up to 75W.
Like the Nvidia GeForce 8800 GT 512MB, the Leadtek WinFast PX8800 GTS 512MB uses Qimonda HYB18H512321BF-10 memory that is rated for a voltage of 2.0V and a frequency of 1000 (2000) MHz. The real frequency is lower at 970 (1940) MHz providing a bandwidth of 62GB/s across a 256-bit memory bus. Qimonda chips rarely work at frequencies above their rated one, so there is not much hope for good overclocking.
Like with the GeForce 8800 GT, the GPU is revision A2. This sample is dated the 45th week of the last year, i.e. November 4-10. The main domain frequency is 650MHz, the shader domain frequency is 1620MHz. These are the reference frequencies of GeForce 8800 GTS 512MB. Considering that the GPU has all its 128 shader processors and 32 TMUs active and working, we can expect the new card to be as fast as the GeForce 8800 GTX/Ultra or even faster. The amount and/or bandwidth of the memory subsystem is the only potential bottleneck because the number of ROPs cannot be a limiting factor in modern applications as you can see in tests.
The left part of the PCB is exactly like that of the Nvidia GeForce 8800 GT. The standard configuration implies two DVI-I ports, so the card doesn’t have a chip that would support the new DisplayPort interface. This sample of the card cannot transfer S/PDIF audio over HDMI, but there are places on it for the appropriate components including an internal S/PDIF connector, so we are surely going to see GeForce 8800 GTS 512MB with support of this feature.
Besides the DVI ports the card has a universal 7-pin mini-DIN connector for analog video output in Composite, S-Video and YPbPr format. This feature isn’t very important nowadays. As opposed to GeForce 8800 GTX/Ultra, the GeForce 8800 GTS 512MB doesn’t support triple-SLI configurations as it has only one MIO connector. Although this card is going to deliver high performance, multi-GPU systems are not as popular as to make the developer redesign the PCB to support triple-SLI.
Using the same PCB design, the new card differs but little from the reference GeForce 8800 GT 512MB. In fact, the single difference is the additional phase in the power circuit. The cooling system is a much more interesting thing.
The reference single-slot cooler Nvidia developed for the GeForce 8800 GT proved to be not quite good. It even didn’t cope with its job in cramped system cases with poor ventilation. The GeForce 8800 GTS 512MB being based on a G92 chip in which all 128 shader processors are active and having higher clock rates than the GeForce 8800 GT, Nvidia has to return to the solution that was tested on the GeForce 8800 GTX and older GeForce 8800 GTS to avoid overheat-related problems.
Notwithstanding a differently shaped casing and a different base, this is still the same design with a blower and a heavily ribbed aluminum heatsink. It exhausts the hot air out of the system case through the slits in the mounting bracket. Three heat pipes are utilized to distribute heat evenly in the heatsink. Two of the pipes transfer the heat right to the spot from where the hot air is exhausted. The third pipe connects the base with the edge of the heatsink in front of the fan. The sole of heatsink, which contacts with the GPU die, is made from copper, of course. Ordinary dark-gray thermal grease is used as a thermal interface.
The base the rest of the cooler components are installed on has juts at the places of contact with the memory chips and load-bearing elements of the power circuit. Fabric pads soaked in white thermal grease are used there to ensure proper thermal contact.
The cooler has a Delta BFB1012L fan (5.67W, 0.48A, 12V) with PWM-based speed regulation. Its speed in 3D mode is normally not higher than 1000-1100rpm, which is perfectly silent. This indicates high efficiency of the cooler. The whole arrangement is secured on the PCB with a number of spring-loaded screws and additionally fastened to the mounting bracket to prevent the cooler from misaligning and damaging the GPU die. The rectangular casing of the cooler is embellished with a picture of the robot. The fan bears a Leadtek logo. These are the only things that differentiate the card from the reference one.
So Nvidia secured its new product against overheat with a high-performance cooler that may cope with the card even at overclocked frequencies. This cooler features a clever design, a high quality of manufacture and high reliability.
We measured the level of noise produced by the new reference cooler from Nvidia installed on the GeForce 8800 GTS 512MB with a digital sound-level meter Velleman DVM1326 using A-curve weighing. The ambient noise in our lab was at 36dBA and the noise level at a 1m distance from the working testbed with a passively cooled graphics card inside was 43dBA. Here are the numbers:
The new cooler is indeed a worthy successor to the coolers of GeForce 8800 GTX/GTS being absolutely silent in every operation mode. Although the fan speed is adjusted dynamically during work, it is never above 700-800rpm. The GPU temperature is 82-85°C in 3D mode, and in idle mode it doesn’t go beyond 60ºC. That’s not high considering the low fan rotation speed so you shouldn’t worry, especially as the hot air is exhausted out of the system case.
Our overclocking attempt bore very good results: we managed to raise the graphics chip frequency from 650MHz to 760MHz without losing graphics card stability. In this case the shader processor frequency increased to 1900MHz. it was possible mostly due to new cooling system. As for the memory, the results here were not as good: without any extreme overclocking methods such as volt-modding, for instance, we could only get to 1050 (2100) MHz maximum speed. As a result, we decided not to benchmark the card at overclocked speeds, because in low resolutions the possible performance gain from overclocking is not critical, while in high resolutions it will be limited by the memory frequency anyway.
To test the performance of Leadtek WinFast PX8800 GTS 512MB in games we assembled the following standard test platform:
According to our testing methodology, the drivers were set up to provide the highest possible quality of texture filtering and to minimize the effect of software optimizations used by default by both: AMD/ATI and Nvidia. Also, to ensure maximum image quality, we enabled transparent texture filtering - Adaptive Anti-Aliasing/Multi-sampling for ATI Catalyst and Antialiasing – Transparency: Multisampling for Nvidia ForceWare. As a result, our ATI and Nvidia driver settings looked as follows:
For our tests we used the following games and benchmarks:
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. The only exception was Enemy Territory: Quake Wars game where we disabled the built-in fps rate limitation locked at 30fps.
Games supporting DirectX 10 were tested in this particular mode. With a few exceptions, the tests were performed in the following most widely spread resolutions: 1280x1024/960, 1600x1200 and 1920x1200. If the game didn’t support 16:10 display format, we set the last resolution to 1920x1440. We used “eye candy” mode everywhere, where it was possible without disabling the HDR/Shader Model 3.0/Shader Model 4.0. Namely, we ran the tests with enabled anisotropic filtering 16x as well as MSAA 4x antialiasing. We enabled them from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia ForceWare drivers
Performance was measured with the games’ own tools and the original demos were recorded if possible. Otherwise, the performance was measured manually with Fraps utility version 2.9.1. We measured not only the average speed, but also the minimum speed of the cards where possible.
The following solutions also took part in our today’s test session besides Leadtek WinFast PX8800 GTS 512MB:
This game doesn’t support display resolutions of 16:10 format, so we use a resolution of 1920x1440 pixels (4:3 format) instead of 1920x1200 for it.
The WinFast PX8800 GTS 512MB is never slower than the Nvidia GeForce 8800 GTX notwithstanding its 256-bit memory bus and the enabled FSAA. It doesn’t mean yet that the GeForce 8800 GTX is now obsolete, like the older versions of GeForce 8800 GTS, but the beginning is very reassuring.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
The game’s system requirements are not really modest but the new version of GeForce 8800 GTS managed to offer an average frame rate higher than 100fps even though in the lowest of the tested resolutions. There is a 20% gap between it and the GeForce 8800 GTX. This gap is getting smaller at the higher resolutions, up to 6% at 1920x1200. Perhaps the picture would be different with FSAA due to higher memory bandwidth requirements, but BioShock doesn’t support FSAA in DirectX 10 mode and the WinFast PX8800 GTS 512MB is the winner.
Call of Juarez proves that the shader processor frequency alone is not the only performance-related factor. Beside it, the graphics card must have enough of graphics memory and high memory bandwidth like those of the Nvidia GeForce 8800 GTS 512MB. The Leadtek card is only 7% slower than the GeForce 8800 GTX, though. The new card is suddenly very slow at resolutions above 1280x1024. It is not because 512 megabytes of local graphics memory is not enough – the ATI Radeon HD 3870 with as much memory on board doesn’t behave like that. It must be due to inefficient memory management by the driver.
Anyway, single graphics cards do not allow playing Call of Juarez in DirectX 10 mode with comfort. Can multi-GPU platforms do that? We’ll find out in our upcoming review.
The Leadtek WinFast PX8800 GTS 512MB is one step behind the Nvidia GeForce 8800 GTX at 1280x1024 but equals it at the higher resolutions. Considering the high minimum of speed, you can play the game with 4x FSAA at every resolution, yet it is only at 1280x1024 that you get the really comfortable frame rate of 50-60fps desirable for games of this genre.
The game being too hard at its Very High level of detail, we benchmarked the cards without FSAA to get a more playable speed.
It is only at a resolution of 1920x1200 that the lower memory bandwidth and, probably, the smaller amount of ROPs make the GeForce 8800 GTS 512MB fall behind the GeForce 8800 GTX. Until that resolution, the new card is only slightly inferior in terms of minimum speed. Like with Call of Juarez, today’s single graphics cards cannot be used to play this game at the highest graphics quality settings even if you turn FSAA off.
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 Nvidia GeForce 8800 GTS 512MB and Nvidia GeForce 8800 GTX are roughly similar, the former being slightly better at low resolutions and the latter, at higher widescreen resolutions starting from 1920x1200. Both cards are fast enough for you to play Quake Wars with the most popular antialiasing method, 4x multisampling.
Like Battlefield 2142, this game does not support resolutions of 16:10 format. So, we use 1920x1440 (4:3 format) instead of 1920x1200 in this test.
The Leadtek WinFast PX8800 GTS 512MB has a comfortably high frame rate at every resolution and is no worse than the Nvidia GeForce 8800 GTX in terms of performance. It is, however, better than the ex-flagship with its smaller size and lower power draw.
The second episode of Half-Life 2 doesn’t have anything new to tell us: the GeForce 8800 GTS 512MB from Leadtek is equal to the Nvidia GeForce 8800 GTX and even better than it by 2-3fps at high resolutions notwithstanding the less advanced memory subsystem. It’s an excellent result that allows playing the game comfortably with enabled FSAA at 1600x1200 or even 1920x1200, but the average frame rate is lower than 50fps if you use the latter resolution.
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.
FSAA is not supported together with the dynamic lighting model and the memory load is relatively low, but the Leadtek WinFast PX8800 GTS 512MB cannot overtake the Nvidia GeForce 8800 GTX. The gap is small, varying from 5% to 8% depending on display resolution. Anyway, the resolutions of 1600x1200/1680x1050 and 1920x1200 are playable with the frame rate never lower than 27-30fps.
Forcing FSAA from the graphics card’s driver doesn’t produce any effect as yet. That’s why the game is tested with anisotropic filtering only.
Like in many other tests there is no big difference between the new GeForce 8800 GTS and the GeForce 8800 GTX. Both provide an average frame rate higher than 80fps.
The low speed of the WinFast card at resolutions higher than 1280x1024 is roughly the same as the speed of the reference GeForce 8800 GT 512MB card. It is probably to insufficiently high memory bandwidth. You can solve the problem by reducing the graphics quality settings below the maximums as it doesn’t affect the visual quality much but the average speed of modern graphics cards at the High level of detail is too low to play this game normally.
Here, we see the WinFast PX8800 GTS 512MB outperform the Nvidia GeForce 8800 GTX at every display resolution. A fan of the Tomb Raider series won’t be disappointed if he buys a GeForce 8800 GTS 512MB for playing this game.
The game being very demanding and its long-range weapon mode differing but slightly from ordinary first-person shooters, few modern graphics cards can provide an acceptable speed at 1280x1024. Besides the Nvidia GeForce 8800 GTX and overclocked versions of GeForce 8800 GT 512MB, the GeForce 8800 GTS 512MB is now capable of that, too.
The current version of the game doesn’t support FSAA, so we performed the test with anisotropic filtering only.
The GeForce 8800 GTS 512MB is 13% ahead of the GeForce 8800 GTX at 1280x1024 but the gap shrinks to 7% and then 3% as the display resolution grows up. You shouldn’t replace your GeForce 8800 GTX if you’ve already got it as this card is going to serve you well for a while yet. But if you are just looking for a graphics card and are ready to shell out up to $400 for it, your choice is obvious now.
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.
It is the amount and frequency of shader processors that determine the performance in closed scenes of this game. The WinFast PX8800 GTS is unrivalled as the consequence. It is 1fps behind the GeForce 8800 GTX at 1920x1200 yet provides a higher minimum of speed.
It is different in open scenes: the GeForce 8800 GTS 512MB is ahead at 1280x1024 only. The GeForce 8800 GTX then catches up with the leader at 1600x1200 and overtakes it at 1920x1200. Note that the GeForce 8800 GTS 512MB has the same result as the GeForce 8800 GT 512MB. It means that the speed is indeed limited by the memory bandwidth.
From a practical standpoint, there is no difference between the Nvidia GeForce 8800 GTS 512MB and GeForce 8800 GTX when it comes to playing this game but the newer card is superior is terms of power consumption and size.
The new add-on to Company of Heroes is tested in DirectX 10 mode only since it provides the highest quality of the visuals.
Leadtek’s card is somewhat slower than the Nvidia GeForce 8800 GTX at high resolutions which we wouldn’t recommend using due to very low minimum speed. The gameplay won’t be really smooth even at 1280x1024.
The game having a frame rate limiter, you should consider the minimum speed of the cards in the first place.
Every top-end graphics card delivers the highest possible speed in this game, which makes it impossible to compare them in this test.
This is one of those games where the Leadtek WinFast PX8800 GTS 512MB is limited by the bandwidth and/or amount of the local graphics memory. It is almost as fast at high resolutions as the reference GeForce 8800 GT 512MB that has fewer shader processors and TMUs.
The new card is only 7% behind the Nvidia GeForce 8800 GTX at 1280x1024 but has a higher minimum of speed. This test resembles CoH: Opposing Fronts: you can try to play at 1280x1024 with enabled FSAA, yet slowdowns are unavoidable.
This is what you could expect to see considering that 3DMark uses a resolution of 1024x768 without FSAA by default. The GeForce 8800 GTS 512MB is only slower than the ATI Radeon HD 3870 and by a very narrow margin, too.
The outcome of the gaming tests run at higher resolutions with 4x FSAA is more true to life. The WinFast PX8800 GTS 512MB is far ahead of the GeForce 8800 GTX at low resolutions but feels limited by its lower memory subsystem performance at higher resolutions. Anyway, the GeForce 8800 GTS 512MB seems to be a worthy successor to the GeForce 8800 GTX.
The Nvidia GeForce 8800 GTS 512MB scores over 11,000 points, leaving behind both Nvidia GeForce 8800 GTX and ATI Radeon HD 3870.
The new graphics card wins the individual tests, too. It enjoys a bigger advantage in the SD3.0/HDR tests while in the SM2.0 tests the Nvidia GeForce 8800 GT 512MB scores only 146 points less than the leader.
The victory of the GeForce 8800 GTS 512MB in the SM2.0 tests was to be expected due to its new TMU architecture as well as to the higher frequency of the main domain, 650MHz as opposed to 575MHz of the Nvidia GeForce 8800 GTX.
The new card is not so confident in the SM3.0/HDR tests. It equals the GeForce 8800 GTX in the first test but loses 5-6% in the second test, which is not so very demanding.
Our tests show that the new graphics card from Nvidia, GeForce 8800 GTS 512MB, is a highly successful product in many respects. It not only sets a new performance level in its price category but also brings about many improvements over the older cards.
Particularly, we mean the return to the highly efficient cooler that used to be installed on older models of GeForce 8800 GTS. As a result, the new card is not only faster but also more reliable. You don’t have to worry about overheat with the new cooler as you had to with the reference GeForce 8800 GT cooler. The level of noise is very low, too.
Our today’s tests show that this new graphics card from the $299-349 price category is indeed as fast as the Nvidia GeForce 8800 GTX, which comes from the $549-649 sector, in most of games despite having less memory and lower memory bandwidth. The only exception is the resolution of 1920x1200 where the influence of the memory subsystem parameters on performance is substantial especially with enabled FSAA. The two cards also differ greatly in certain games, mostly those that use DirectX 10: Call of Juarez, Lost Planet and World in Conflict. Interestingly, these three games have one thing in common. All of them have very high system requirements, and none of existing graphics cards can deliver a comfortable frame rate in them at the highest graphics quality settings. In Crysis, the most technically perfect game of today, the new card is only inferior to the GeForce 8800 GTX at 1920x1200, and we don’t think it is because it felt a lack of graphics memory but rather because the available memory was not efficiently utilized by the ForceWare driver.
In other words, Nvidia really offers you the performance of a GeForce 8800 GTX for $299-349. It is quite possible that, like with GeForce 8800 GT, there will be pre-overclocked versions of GeForce 8800 GTS 512 or even versions with 1GB of memory, and such cards won’t leave a chance to the GeForce 8800 GTX. The G80 generation will have to leave the scene to better solutions. That’s the law of the market, and the next generation of GeForce cards, with the index of 9800, is already looming in the horizon (see this news story for details).
Alas, it is going to be impossible to purchase a GeForce 8800 GTS 512MB for $299 in near future. Nvidia can try to solve this problem directly, i.e. reducing the price of the G92, but also by recommending its partners to use cheaper components (see this news story for details). Anyway, it is already clear that there is no reason why you may want to buy a GeForce 8800 GTX today. But if you’ve got such a card already, you don’t have to replace it with a GeForce 8800 GTS 512MB as the latter delivers roughly similar performance.
As for the specific card from Leadtek, it received highly positive verdict from us, although it is not a unique products but just a copy of the reference card from Nvidia with Leadtek’s logotype stickers on the cooler. If you are not hunting for any unique high-tech features, the WinFast PX8800 GTS 512MB is going to be a good choice, and you’ll get a full copy of Neverwinter Nights 2 game with it as a bonus.