by Alexey Stepin , Yaroslav Lyssenko
02/28/2008 | 04:25 PM
Announcing the G92 graphics core, Nvidia introduced two GeForce 8800 GT cards with 512 and 256 megabytes of graphics memory on board and priced at $259 and $199, respectively. However, it was only the more expensive model that came to market in mass quantities. There was a shortage of G92 chips, and the 512MB version was not freely available while its retail price exceeded the recommended one greatly. On the other hand, there was no serious alternative to the Nvidia GeForce 8800 GT 512MB that could provide comparable performance at a similar price.
This situation changed with the release of ATI Radeon HD 3800 on November 19, 2007. Knowing beforehand that even the senior model of the new series wouldn’t be competitive against the Nvidia GeForce 8800 GT 512MB, AMD launched an attack in the lower market segment setting the recommended price of the ATI Radeon HD 3870 and ATI Radeon HD 3850 at $219 and $179, respectively. Our tests showed that the new card really provided good performance in the below-$250 sector and Nvidia had to respond somehow. Although the GeForce 8800 GT 256MB had been expected to ship in mass quantities in late November, first such cards only emerged in December. Nvidia had lost its leadership in the mentioned price sector by that moment.
The problem with the availability of GeForce 8800 GT is being solved steadily, and the new inexpensive version of the card, with 256MB of memory, has appeared in large quantities on the market. But is it efficient under modern conditions? A few years ago even 256 megabytes of memory was excessive as the existing games could not use up all of it. 3D games are constantly developing, though, and today the optimal amount of graphics memory is 512 megabytes while some latest projects, e.g. Crysis, can use even more if you run them at the highest graphics quality settings together with full-screen antialiasing.
Today we’ve got an opportunity to check out the performance of the Nvidia GeForce 8800 GT 256MB in popular games to compare it with its two most dangerous market rivals, ATI Radeon HD 3870 and HD 3850. We are going to see what effect the amount of graphics memory has on performance of Nvidia’s mass solutions and if 256 megabytes is enough to compete with the ATI Radeon HD 3870. Prices of G92-based products still being very high, the GeForce 8800 GT 256MB has to rival that ATI solution as long as the price of the former card differs greatly from the recommended one. We will also see if the ATI Radeon HD 3850 can show its worth in comparison with the Nvidia GeForce 8800 GT 256MB and prove the correctness of ATI’s approach to designing graphics processors.
In our today’s review the Nvidia GeForce 8800 GT 256 is represented by a graphics card from Sparkle that goes under the incomprehensive name of SF-PX88GT256D3-HP. We’ll be referring to it as Sparkle GeForce 8800 GT 256MB especially as its parameters are exactly those of the reference card.
The Sparkle GeForce 8800 GT 256MB comes to retail in a rather small upright box. It resembles some products from Gigabyte:
The box design is one of the best we’ve seen lately as it presents a nice variation after all the hackneyed robots and monsters painted in garish colors. The calm combination of white, blue and silver looks appealing and quite charming. There is a transparent window in the box you can look at the card through: that’s handy for the customer but graphics card makers do not use this solution often lately.
All the basic technical features of the product are listed on the face side of the box, particularly its support of PCI Express 2.0. As you know from our reviews, Nvidia’s G92-based cards are not quite compatible with old mainboards.
The graphics card is firmly fixed in a plastic tray that protects it against any troubles during transportation and storage. It is accompanied with the following accessories:
The accessories are scanty, yet sufficient for using the Sparkle GeForce 8800 GT 256MB normally. The lack of a DVI → HDMI adapter is the only really disappointing thing. There are no games included, but not all users care about them. So while the package design is top-class, the accessories are only normal.
The Sparkle GeForce 8800 GT 256MB resembles the GeForce 8800 GT 512MB and uses the reference PCB design. The card is blue rather than green as most cards manufactured for Nvidia are.
We could find no differences in the PCB wiring. A four-phase Primarion PX3544 controller is responsible for the power supply of the GPU and one of its phases is not installed just like on the Nvidia GeForce 8800 GT 512MB. An Intersil ISL6549CBZ chip is responsible for the memory chips. Like the reference GeForce 8800 GT 512MB, this card is equipped with one 6-pin PCI Express 1.0 power connector.
The Sparkle card is equipped with GDDR3 memory manufactured by Quimonda. The HYB18H256321BF-14 chips have a capacity of 256Mb (8Mb x 32) and a voltage of 1.8V. Eight such chips make up a 256MB memory bank accessed across a 256-bit bus. For an access time of 1.4 nanoseconds the rated frequency is 700 (1400) MHz, and this is the frequency the memory is actually clocked at (it is also the same as the memory frequency of the reference GeForce 8800 GT 256MB). So the memory bandwidth of the junior GeForce 8800 GT is considerably lower at 44.8GB/s as opposed to 57.6GB/s of the senior model. This may have a negative effect on the card’s performance at high resolutions and with enabled FSAA. Already clocked at their rated frequency, the memory chips seem to lack any overclocking potential, but we had a nice surprise with them as you’ll learn shortly. Some makers may install faster memory on their versions of GeForce 8800 GT 256MB because the GPU developer doesn’t impose strict limitations in this respect.
The GPU is revision A2 like the GPUs of all the versions of GeForce 8800 GT 512MB we have tested before. This sample is dated the 41st week of the last year (October 7 to 13). The GPU configuration is the same as on the GeForce 8800 GT 512MB: 7 active computing blocks with a total of 112 unified shader processors, 28 TMUs and 16 ROPs. To remind you, each of the G92’s TMUs has two address and two texture filter units, which theoretically means a total of 56 TMUs. However, real applications imply the use of tri-linear and anisotropic filtering, and these 28 TMUs may prove to be even less efficient than the 32 TMUs of the G80 chip in which there are two filter units per each texture address unit. The main domain frequency is 600MHz, the shader domain frequency is 1500MHz. The Sparkle GeForce 8800 GT 256MB doesn’t differ in anything from the reference GeForce 8800 GT 512MB in this respect.
The seat for a DisplayPort chip is left empty as the card is equipped with two standard DVI-I connectors supporting dual-link mode. The Sparkle supports the audio-over-HDMI feature by means of the 2-pin white plug, located near the MIO connector. It can be connected to the SPDIF connector of your sound card. An external adapter can probably be used as well, like with the Gainward Bliss 8800 GT 512MB GS GLH, but neither this adapter nor a DVI-I → HDMI adapter is included into the box. The 7-pin mini-DIN port outputs analog video in S-Video, Composite and YPbPr formats. The latter format is the analog version of HDTV which has lost popularity due to the arrival of HDMI.
The original cooler installed on the GeForce 8800 GT was been criticized harshly by the users. Trying to make the card compact, the developer chose a single-slot format for the cooler. Relying on the 65nm tech process, the company underestimated the heat dissipation of the new graphics core and made a number of flaws in the cooler design. Particularly, the cooler was equipped with a very small fan while the heatsink was not large enough to cool a chip with a heat dissipation of 80W. As a result, the cooler often failed to perform properly and some cards even died from overheat.
Releasing the GeForce 8800 GT 256MB, Nvidia made some corrections and installed a second, improved, version of the cooler on the whole series. This is the cooler you can see on the Sparkle card.
The new cooling system is almost the same as the older one externally but the larger fan is noticeable. There are, however, more changes than visible to the eye. The new version of the cooler has a different shape and area of the copper piece that contacts the GPU die. The configuration of the heat pipes has been optimized. The heatsink area has been enlarged considerably by reducing the distance between the ribs and increasing their number.
The shape of the base has become different, too. The holes near the fan have been enlarged, probably to improve the cooling of the PCB. Instead of Nvidia’s traditional fabric pads soaked in thermal grease there are elastic pink-colored thermal pads now. The cooler’s casing is still metallic, but it is now fastened to the base with screws rather than latches.
With all the changes in the design, the new cooler seems to be more efficient than the older one, yet you still have to take care about proper ventilation of your system case at large.
We measured the level of noise produced by the Sparkle card, which has a larger fan than in the reference cooler of the GeForce 8800 GT, with a digital sound-level meter Velleman DVM1326 using A-curve weighing. The level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from the working testbed with a passively cooled graphics card inside was 43dBA.
The new version of the reference cooler is indeed quieter and feels virtually silent to the eye. The cooling performance hasn’t worsened. On the contrary, the GPU temperature is 85-88°C under load. This is not low, but the temperature would easily be higher than 90°C with the older cooler, especially if the system case was not ventilated properly. The new, improved, cooler is installed on all new batches of GeForce 8800 GT – you can tell it by the diameter of the fan and the hole in the casing.
The card did well at overclocking. We managed to increase the GPU frequency to 700MHz, and the shader domain frequency rose up to 1750MHz. The memory chips were stable at 1000 (2000) MHz. This frequency gain implies a serious performance increase, so we will test the card at the overclocked frequencies, too. Besides everything else, this will help us better see how the memory amount limits the performance of the GeForce 8800 GT 256MB.
The card proved to be free from compatibility problems. It worked normally with all PCI Express 1.0a mainboards we tried it on.
To test the performance of Sparkle GeForce 8800 GT 256MB 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.
We have also included the results for the following graphics accelerators:
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.
This traditional opener of our test program says that the Nvidia GeForce 8800 GT 256MB doesn’t feel a lack of graphics memory. It delivers high performance at resolutions up to 1600x1200 and good performance (for its class) at 1920x1400. The card is roughly equal to the ATI Radeon HD 3870 although has a high minimum of speed at low resolutions. The memory frequency becomes an important factor starting from the 1600x1200 mode, and the overclocking provides a considerable performance boost, over 20% at 1920x1440.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
There is no shortage of graphics memory in BioShock because there is no FSAA. The Sparkle GeForce 8800 GT 256MB is far faster than the ATI Radeon HD 3870 and becomes as fast as the GeForce 8800 GT 512MB when overclocked (except that it has a somewhat lower minimum speed then).
The amount of graphics memory is the crucial factor for Call of Juarez. The Sparkle card fails this test even in comparison with the ATI Radeon HD 3850. It’s not even the amount but the inefficient use of the available memory by Nvidia’s driver. We noticed this problem quite a long time ago, but there have been no improvements.
The Sparkle card loses its ground starting from 1600x1200, but that’s not a big problem considering its official price. The worse thing is that the card cannot keep the minimum speed even at 25fps in the 1280x1024 mode, which makes it unsuitable for running Call of Duty 4, at least with enabled FSAA.
The game being too hard at its Very High level of detail, we benchmarked the cards without FSAA to get a more playable speed.
The application is far more demanding even among other DirectX 10 projects. It refers to the graphics memory as well. 512 megabytes of the bare minimum for Crysis, and the low results of the Sparkle GeForce 8800 GT 256MB are not surprising at all. The overclocking brings about no visible effect because it is indeed the memory amount that limits the card’s performance here.
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.
Using MegaTexture technology, Quake Wars appreciates large amounts of graphics memory. 256MB should be enough for low resolutions, but the GeForce 8800 GT 256MB suffers from inefficient memory management again. The overclocking helps improve the performance at 1280x1024 but is virtually useless at the higher resolutions. The ATI Radeon HD 3850 can run the game normally at 1600x1200 but the Sparkle card is limited to 1280x1024.
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 Nvidia GeForce 8800 GT 256MB is good at 1280x1024 and 1600x1200. The Sparkle card is as fast as the ATI Radeon HD 3870 at the default clock rates and nearly matches the GeForce 8800 GT 512MB at the overclocked frequencies. The overclocking doesn’t help at the higher resolutions because the speed bottoms out below playable level.
The only resolution at which the Nvidia GeForce GT 256MB delivers comfortable performance is 1280x1024. People who buy such graphics cards usually have a 19” monitor with a native resolution of 1280x1024 or 1440x900 pixels, so that’s not a problem at all. When overclocked, the Sparkle delivers a playable average frame rate at 1600x1200, too, yet the minimum speed is rather low.
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.
Nvidia’s GeForce 8800 cards are traditional strong in this game, and the Sparkle GeForce 8800 GT 256MB is quite good at 1280x1024. At least it delivers a higher average frame rate than the ATI Radeon HD 3870. At the higher resolutions the card doesn’t have enough graphics memory to perform fast. Interestingly, overclocking has but a small effect here irrespective of the resolution.
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.
The game has moderate system requirements and runs fast enough on the Sparkle GeForce 8800 GT 256MB at every resolution. The frame rate is playable even at 1920x1200. The choice between Nvidia’s solution and the ATI Radeon HD 3850 depends only on your personal preferences here.
The Sparkle GeForce 8800 GT 256MB fails this test irrespective of the clock rates. We’ve already seen this behavior, provoked by inefficient memory management, in Call of Juarez and Crysis. The problem roots in the software, in the ForceWare driver, because ATI Radeon HD cards with as much graphics memory on board behave differently and are not that slow in comparison with their 512MB versions.
This game has moderate graphics memory requirements. Judging by the average frame rate, the Sparkle GeForce 8800 GT 256MB delivers superb performance at every resolution. There is a problem about the minimum speed, though. This parameter can be lifted up to acceptable level at 1600x1200 by means of overclocking.
It’s similar to what we’ve seen in Crysis, Call of Juarez and Lost Planet. The Nvidia GeForce 8800 GT 256MB is very slow and doesn’t accelerate much even at the overclocked frequencies. This graphics card cannot be used to run Hellgate: London at such settings.
The current version of the game doesn’t support FSAA, so we performed the test with anisotropic filtering only.
The minimum speed of the Sparkle GeForce 8800 GT 256MB and ATI Radeon HD 3850 is low as they have only 256MB of local memory. You cannot play the game comfortably at these cards because the speed may bottom out to 13-16fps.
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 Sparkle GeForce 8800 GT 256MB is but slightly inferior to the Nvidia GeForce 8800 GT 512MB and copes well with the task of ensuring comfortable performance even at 1920x1200. Still, we think the ATI Radeon HD 3870 looks preferable as it doesn’t feel a lack of graphics memory in some other games.
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 flaws of the ForceWare driver can be observed once again: very low performance of G92-based cards equipped with 256 megabytes of graphics memory. The ATI Radeon HD 3850 is not much better for practical purposes, though. The ATI Radeon HD 3870 and GeForce 8800 GT 512MB are the only cards that can run the game normally, yet their minimum speed doesn’t reach 20fps, either.
The game having a frame rate limiter, you should consider the minimum speed of the cards in the first place.
The Sparkle GeForce 8800 GT 256MB is the only card whose minimum speed falls below the comfortable level as soon as 1280x1024. The overclocking helps solve the problem up to the resolution of 1920x1200 where the lack of graphics memory becomes the crucial factor. The cheaper and more available Radeon HD 3850 looks preferable here.
The Sparkle GeForce 8800 GT 256MB is a disappointment here due to the same reason as it was in Call of Juarez, Crysis, Lost Planet, Hellgate and Company of Heroes: the driver cannot manage the available graphics memory efficiently. The ATI Radeon HD 3850 doesn’t behave like that, which means that memory management is implemented better in the Catalyst driver. Overall, each card, including the GeForce 8800 GT 512MB, is rather slow in this test.
The GeForce 8800 GT 256MB takes last place when working at its default frequencies even though the test is run at 1024x768 without FSAA. This is not a lack of graphics memory because the overclocked frequencies (700MHz core and 1000 (2000) MHz memory) help it beat the GeForce 8800 GT 512MB and even the ATI Radeon HD 3870.
The Sparkle GeForce 8800 GT 256MB is considerable faster than both Radeon HD 3800 models in the first and third tests, but falls to the level of the ATI Radeon HD 3850 in the second, math1ematics-heavy, test. When overclocked, the Sparkle wins two out of the three tests and also wins the resolution of 1280x1024 in the second test. These results agree with the overall 3DMark05 scores.
The GeForce 8800 GT 256MB is confident in 3DMark06, outperforming the ATI Radeon HD 3850 and loses only 430 points to the ATI Radeon HD 3870. FSAA is not enabled by default, so the overclocking helps reach the 11,000 mark and take first place.
The Sparkle card is ahead of the ATI Radeon HD 3850 and even of the ATI Radeon HD 3870 in the SM2.0 tests. The results of the SM3.0/HDR are predictable, too, as the Sparkle, being a copy of the reference card from Nvidia, cannot compete with the ATI Radeon HD 3850 anymore. AMD’s solution is inferior to Nvidia’s one in TMU performance but is far superior to it in sheer computing power which is necessary to execute the complex shaders the SM3.0/HDR tests abound in.
The Sparkle card beats the ATI Radeon HD 3800 in the first test easily due to the specifics of the test, but its advantage over the ATI Radeon HD 3870 is only 5% in the second test.
According to the individual SM3.0/HDR tests, the ATI Radeon HD 3850 beats the non-overclocked Sparkle in both cases. When overclocked, the Sparkle only goes ahead of the ATI Radeon HD 3870 in the second test.
The results of the individual tests agree well enough with the overall scores, though.
The Sparkle SF-PX88GT256D3-HP graphics card is a good product overall except for its scanty accessories. Packaged into a pretty box, it features an improved cooler and overclocks easily.
However, the card’s gaming performance – and that of any copy of the reference GeForce 8800 GT 256MB – is not quite good. The GeForce 8800 GT 256MB may be far behind the GeForce 8800 GT 512 in those applications where there is but a small difference between the ATI Radeon HD 3850 and Radeon HD 3870.
This mostly shows up in modern games that use DirectX 10 capabilities, except for BioShock that runs on the thoroughly optimized Unreal Engine 3. We observed a considerable performance hit in such games as Call of Juarez, Crysis, Lost Planet, Hellgate: London, CoH: Opposing Fronts, and World in Conflict. These are all DirectX 10 applications and all of them are so demanding that mainstream graphics cards usually cannot provide high enough performance in them. In some other games the Nvidia GeForce 8800 GT 256MB feels a lack of graphics memory at high resolutions, 1600x1200 and higher, this graphics card is not actually meant for.
This is already the second time that we see Nvidia’s GeForce 8 solutions suffer a performance hit when their local memory is reduced twofold. The first time we observed that with the 8800 GTS 640MB and 8800 GTS 320MB, and now we see the same with the 8800 GT 512MB and 8800 GT 256MB. We can only make guesses as to why the graphics cards behave like that:
We’ve combined the performance data on the Nvidia GeForce 8800 GT 256MB and two models of ATI Radeon HD 3800 into the following table:
Nvidia’s solution seems to be equal to the ATI Radeon HD 3870 at 1280x1024, but AMD’s card is preferable at 1600x1200. Although the latter comes at a higher recommended price ($219 as opposed to $199), retail prices of G92-based products are still too high, and the customer will often choose between the Nvidia GeForce 8800 GT 256MB and ATI Radeon HD 3870. The choice is obvious here.
The Nvidia GeForce 8800 GT 256MB doesn’t look good in comparison with the ATI Radeon HD 3850, either. The AMD card is cheaper even in terms of recommended price. At the resolution of 1280x1024 such cards are actually meant for the AMD card is but rarely inferior to the Nvidia solution, for example in S.T.A.L.K.E.R. and Call of Duty. But it also allows playing some games comfortably at 1600x1200. If these two cards cost the same money, the choice would be based on the buyer’s personal gaming preferences, but with the current prices the ATI Radeon HD 3870 seems preferable to the GeForce 8800 GT 256MB.
Users who cannot afford to spend more than $200 for the graphics card should consider the ATI Radeon HD 3850 in the first place. Thus, the popularity of the GeForce 8800 GT 256MB depends directly on its retail price. If Nvidia brings it down to the level of the ATI Radeon HD 3850 or lower, the product will be demanded by gamers with limited budgets who do not plan to play at resolutions higher than 1280x1024. In this case the Sparkle card we’ve described here may be a good choice, particularly due to its good overclockability and lack of extra features that would increase the price of the product.