by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
06/28/2007 | 03:48 PM
Thanks to its G80 graphics processor Nvidia was not only the leader in the sector of top-performance desktop graphics solutions but actually the single supplier of truly modern and fast graphics cards until the recent release of the ATI Radeon HD 2900 XT. The G80-based GeForce 8800 series covered price ranges from $299-349 (GeForce 8800 GTS 320MB) to $599-$649 (GeForce 8800 GTX). The May 2 announcement of the GeForce 8800 Ultra extended this coverage further up to over $800.
Yes, the GeForce 8800 series brought the user extremely high performance as was confirmed by our tests but Nvidia didn’t have anything to fight ATI’s inexpensive solutions with in the lower market segments. Nvidia’s products on the previous-generation G71 processor had good performance but were inferior to AMD/ATI’s solutions in capabilities and image quality. Particularly, the GeForce 7 series could not support FP HDR and full-screen antialiasing simultaneously and provided a much lower quality of anisotropic filtering in comparison with graphics cards from the Radeon X1950 and X1650 families.
And so Nvidia wanted a new chip that would become the core of a new graphics card family targeted at the $150-229 price range because this segment accounts for some 75% of total sales of discrete graphics solutions. The new chip had to be not only fast but also technically advanced considering the imminent arrival of new Radeons. And this chip, or rather a family of two chips, a faster G84 and a slower G86, was introduced on the 17th of April.
A variety of cost-reducing measures was taken, of course, and the G84 came out inexpensive but lacking a lot of the G80’s features, being equipped with only 32 unified shader processors, 16 texture-mapping units, and 8 raster operators. Thanks to its relative simplicity (289 million transistors as opposed to the G80’s 681 million) and the use of 80nm tech process the main core frequency could be increased to 675MHz and the shader processor frequency to 1.45GHz which made up for the performance loss ensuing from the reduction in the amount of the chip’s functional. However, our tests in real-life applications showed that even the fastest model of the series, GeForce 8600 GTS, could only compete with the GeForce 7900 GS, not the 7950 GT, and was overall slower and, accordingly, a less appealing buy than the ATI Radeon X1950 Pro.
Thus, the long-established tradition that new-generation graphics solutions priced at $199 deliver performance of one-year-old flagship products was broken. The GeForce 8600 series had every chance to follow the fate of the GeForce FX 5600 which had never taken off due to its low performance and had been quickly replaced with the faster GeForce FX 5700. Perhaps we’ll indeed see a successor to GeForce 8600 in the third or fourth quarter this year, but a number of manufacturers was known to offer pre-overclocked GeForce 8600 GTS even at the time of the announcement of the new series. Such cards are faster and more appealing in the potential buyer’s eyes.
Today we will show you one such card, produced by MSI, and will see what performance you can expect from it in modern games.
The card came to our lab in its retail package with a full set of accessories. It is a small black-and-silver box designed in a rather hackneyed fantasy style.
There is a black cardboard box inside the colorful wrapping. It has a translucent cap you can see the card through. The card lies in a special foam-plastic container that protects it from damage during transportation. The accessories lie in a cardboard compartment nearby. Here they are:
The kit includes everything you need to use the card. The user manual comes in two versions: a black-and-white brochure and a large colorful poster. Such posters have become popular among graphics card vendors recently although they don’t seem as convenient to us as traditional booklets.
Anyway, the MSI product is packaged well and comes with everything necessary, yet this is not the card’s key feature. What is it? You’ll learn shortly.
AMD/ATI and Nvidia follow a more liberal policy with respect to their mainstream solutions, like GeForce 8600 GTS, than to their top-end products. The latter are shipped to the GPU developers’ partners ready-made and the only thing the graphics card supplier can do is to experiment with the cooling system. With mainstream products, there is a much larger field for experiments. Unique PCBs can be developed, for example.
Although the MSI card is based on the etalon PCB design, the card itself differs from the original with its red solder mask, yellow DVI connectors, and a different marking. MSI either uses PCBs it makes itself or buys from an OEM. We suspect the card is made at MSI’s own production facilities.
The PCB carries a revision A2 G84-400 processor with full HDCP support. Our sample is dated the 9th week of 2007, i.e. late February or early March. The main graphics core frequency is 700MHz, not much higher than the reference frequency of 675MHz. The shader domain frequency was left intact at 1450MHz. The frequency gain is so small that you can’t expect great performance benefits, but the card is equipped with an advanced cooler and can theoretically do better if you try to overclock it further on your own. The GPU configuration is standard for a GeForce 8600 GTS: 32 universal shader processors capable of executing any type of shaders, 16 TMUs and 8 raster operators.
The G84’s external bus being 128 bits wide, the card has only four memory chips (Samsung K4J52324QE-BJ1А, 512Mbit, 16Mx32) that provide a total of 256 megabytes of graphics memory. The memory chips are rated for a frequency of 1000 (2000) MHz, but work at 1050 (2100) MHz on the MSI card. This results in a small growth of memory bandwidth, from 32GB/s to 33.6GB/s, which can hardly affect the card’s gaming performance much, but the good cooler allows to overclock the card more if you want.
The MSI NX8600GTS-T2D256E-HD-OC is identical to the reference card otherwise. Its main feature is not the red color of the PCB but the advanced cooler it is equipped with.
According to our tests the GeForce 8600 GTS consumes only 47W under load and does not require a sophisticated cooling system. It can do quite well with a single-slot cooler. The MSI card has higher GPU and memory frequencies than the reference card, but the frequency growth is small and the card can hardly generate more than 50 watts of heat, but the company targets this product at enthusiasts and overclockers and installs a rather advanced cooler on it.
The cooler blocks the neighboring slot as in faster G80-based solutions and has a similar design. Contacting with the GPU die through a layer of dark-gray thermal grease, its copper sole transfers the heat to the aluminum heatsink via a heat pipe. The heatsink is blown at by a 2.2W fan. The hot air is exhausted through slits in the card’s mounting bracket. As opposed to the reference card, the fan uses a 3-pin connection via an adapter.
The cooler has an aluminum base all of its elements are fastened on and the juts in this base have elastic pads for better contact with the memory chips. The whole arrangement is secured on the PCB by means of four threaded poles. A protective plastic bar is put on these poles on the reverse side of the PCB. There are also spring-loaded round nuts with slits for a flat-headed screwdriver there. You cannot tighten the nuts more than necessary as the screwdriver’s tip will press into the pole. Anyway, you should be very careful if you remove the cooler or mount it back again. The nuts are fragile and their heads may break off as they did on our sample of the MSI NX6800GT-T2D256E (for details see our article called MSI NX6800GT-T2D256E Graphics Card Review ).
The cooler’s casing does not touch the mounting bracket and some of the hot air will get into the system case. Anyway, this cooler should be more than enough to keep a GeForce 8600 GTS cool even at overclocked frequencies. We’ll now check out how noisy this cooling solution is.
We measured the level of noise produced by the MSI card with a digital sound-level meter Velleman DVM1326 using A-curve weighing. At the time of our tests the level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from a working testbed with a passively cooled graphics card inside was 43dBA. We got the following results:
The cooler employed by MSI is somewhat quieter than the reference one, yet still not completely silent. Its noise spectrum contains a distinct note you can easily catch out of the noise from other system components. We guess it is just the sound of the air passing through the heatsink. It is not irritating and is completely lost when the system case is closed. The cooler’s fan is working at the same speed in every operation mode.
We couldn’t help trying to overclock the MSI card considering its overclocker-friendly cooler. We managed to increase the GPU frequency from the default 700MHz to 760MHz. This is 85MHz higher than the frequency of the reference card (675MHz). The frequency delta between the main GPU domain and the shader processor domain is about 2.071 on the MSI NX8600GTS-T2D256E-HD-OC, so the shader domain was clocked at 1757MHz after that. This is a considerable increase which should show up in gaming tests.
The memory subsystem was overclocked quite well considering its characteristics, from the default 1050 (2100) MHz to 1160 (2320) MHz. We achieved a good frequency growth and tested the card in that mode too to see how its performance was affected.
To test the performance of MSI NX8600GTS-T2D256E-HD-OC we assembled the following standard test platform:
Since we believe that the use of tri-linear and anisotropic filtering optimizations is not justified in this case, the AMD and Nvidia graphics card drivers were set up to provide the highest possible quality of tri-linear and anisotropic texture filtering. We have also enabled transparent texture filtering. As a result, our AMD and Nvidia driver settings looked as follows:
We selected the highest possible graphics quality level in each game using standard tools provided by the game itself. The games configuration files weren’t modified in any way. Performance was measured with the games’ own tools or, if not available, manually with Fraps utility version 2.8.2. We also measured the minimum speed of the cards where possible.
We performed tests in 1280x1024/960, 1600x1200 and 1920x1200 resolutions. The games that didn’t support 16:10 ratio were run in 1920x1440 resolution. We used “eye candy” mode everywhere, where it was possible without disabling the HDR or Shader Model 3.0. Namely, we ran the tests with enabled anisotropic filtering as well as MSAA 4x. We enabled them from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of Catalyst and ForceWare.
MSI NX8600GTS-T2D256E-HD-OC will be competing against the following graphics accelerators participating in our test session:
For our tests we used the following games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
Quite expectedly, the factory overclocking has a very small effect on the card’s performance. At the highest frequencies, however, the MSI speeds up as hard as to outperform the Radeon X1950 Pro at 1600x1200. This has little practical value, though, because the average speed of 35-37fps is not comfortable enough for this game genre.
The MSI card is slow even at the max frequencies in 1920x1200 as it has only 8 ROPs and a 128-bit memory bus. The Radeon X1950 Pro is free from these disadvantages and delivers a better result.
The speed of the Radeon X1950 Pro is slower than on the AMD/Windows XP platform we used in our earlier reviews. It may have something to do with the adaptive antialiasing option we didn’t use earlier. Another possible reason is that the Catalyst driver for Windows Vista is not yet optimized enough.
The MSI card and the reference GeForce 8600 GTS have almost identical speeds, while the further overclocking of the NX8600GTS makes it as fast as the Radeon X1950 Pro in 1280x1024. Anyway, you can’t play comfortably because the average frame rate is only 23fps. You’ll have to lower the level of detail and, probably, disable Shader Model 3.0/HDR to play this game on a mainstream graphics card.
The Radeon X1950 Pro does not slow down because of the enabled adaptive antialiasing option as it did in Battlefield 2142.
The Radeon X1950 Pro is far slower than every version of GeForce 8600 GTS when transparency antialiasing is enabled. No wonder as the Pier map abounds in vegetation and the foliage is rendered using such textures. The overclocking gives a boost to the MSI card but doesn’t add more playable resolutions. You can still play comfortably in 1280x1024 and, perhaps, in 1600x1200 at the eye candy settings.
There are few transparent textures on the Research map but the graphics card’s shader processors have got a lot of work to do. The Radeon X1950 Pro is ahead here. It is only at the overclocked frequencies that the MSI NX8600GTS OC gains the upper hand at 1280x1024. It’s better with average performance than on the Pier map but the resolution of 1920x1200 is still unplayable because you can’t always remain in closed scenes as you are playing Far Cry. You’ll have to go out into the open sooner or later and suffer a performance hit.
The Radeon X1950 Pro does not suffer from the above-mentioned problem in the HDR mode and is quite competitive against every version of GeForce 8600 GTS except for the MSI card at the frequencies we set ourselves. The resolution of 1280x1024 is playable on each tested card. 1600x1200 is playable, too, even though the speeds are close to being uncomfortably low.
The overclocked frequencies bring no benefits to the MSI card in high resolutions while in 1280x1024 its speed is less than 10% higher in comparison with the reference card. The Radeon X1950 Pro performs slower with enabled adaptive antialiasing, so the former outsider GeForce 8600 GTS is now ahead, at least in low resolutions which are playable with more or less comfort on mainstream graphics cards.
Using the deferred rendering technique, this game is incompatible with full-screen antialiasing. There are only anisotropic filtering results here.
The choice of the graphics card to play the Tom Clancy’s series of tactical shooters depends on your personal taste because the MSI NX8600GTS OC and the Radeon X1950 Pro are equally sluggish in Ghost Recon Advanced Warfighter although ensure a high enough minimum of speed. The former card is more future-proof as it supports DirectX 10. The further overclocking of the GeForce 8600 GTS is useless for this game, resulting in a speed gain of less than 2fps.
The Radeon X1950 Pro and GeForce 8600 GTS are both slower in comparison with themselves on our older testbed and tested with our older methods. Their standings have remained unchanged, however, except for 1280x1024 where the Radeon X1950 Pro is considerably faster than the standard GeForce 8600 GTS. The overclocked MSI NX8600GTS OC almost overtakes its opponent, though.
Note that there is almost no reserve of speed in the resolution of 1600x1200 when transparency antialiasing is turned on, so you may want to play the game at 1280x1024 to have the maximum image quality.
Despite the OpenGL-based game engine and enabled FSAA, the Radeon X1950 Pro is preferable to every version of GeForce 8600 GTS. The average speed is low, however, so you cannot play the game comfortably with 4x FSAA and at the maximum level of detail.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. So, we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
The GeForce 8600 GTS suffers a heavy performance hit with the new driver settings whereas the Radeon X1950 Pro stays on the same level. This must be some flaw in the ForceWare driver or the consequence of our enabling transparency antialiasing. The game is not very fast on the Radeon, yet it is quite unplayable on the GeForce 8600 GTS.
The cards behave almost as they did on the older testbed, equaling each other. The overclocked frequencies of the MSI card do not bring in any advantages. 1280x1024 is still the highest comfortable resolution.
The GeForce 8600 GTS does not surpass the Radeon X1950 Pro in average speed anymore, but is still ahead of it in terms of minimum speed and thus suits better for playing this game.
The MSI card is 6-7% ahead of the reference card, which has no practical value for the gamer. Our overclocking increases the gap to 18% but this has little effect on the minimum speed which is a more important factor here than the average frame rate. Full comfort can only be achieved by replacing your mainstream-class graphics card with something better, like a GeForce 8800 GTS 320MB.
We try to get the best quality of graphics from each game, so we chose HDR in favor of FSAA and benchmarked the cards with anisotropic filtering only.
Having a considerably higher computing power the Radeon X1950 Pro is beaten by the GeForce 8600 GTS anyway. The latter’s performance isn’t exactly brilliant, yet the average frame rate is never lower than 30fps (35fps with the overclocked card from MSI). You won’t have full comfort because the minimum of speed is still too low, below the critical mark of 25fps, yet the game is playable and with the highest possible image quality.
The current version of Gothic 3 does not support FSAA, so we benchmarked the cards using anisotropic filtering only.
The GeForce 8600 GTS and Radeon X1950 Pro have similar average speeds but Nvidia’s card has a higher minimum of speed than AMD/ATI’s one. However, neither card can ensure comfortable playing conditions. The overclocked frequencies of the MSI card can’t save the day even though they provide a 14% boost in the average speed in comparison with the reference card.
The game permits to use FSAA starting with version 1.04, but its support for HDR is still deficient. Therefore we tested the cards at the eye candy settings with 4x FSAA.
The GeForce 8600 GTS in its every version fails this test at the new, higher-quality driver settings. 15fps is the minimum you need to play this game, but even the overclocked MSI card cannot reach this limit, although it is only 0.4fps behind the Radeon X1950 Pro. If you own one of these cards, you have to disable transparency antialiasing to achieve a more or less acceptable speed. We also recommend disabling FSAA unless you’ve got a GeForce 8800 series card.
The game looks much worse without HDR (although this is an arguable point), so we test it with enabled FP HDR.
The Radeon X1950 Pro is just as fast as it used to be while the GeForce 8600 GTS has got somewhat slower in terms of minimum speed after our transition to Windows Vista and to the use of transparency antialiasing. Anyway, both cards can be used to play Oblivion at the highest level of detail in 1280x1024, judging by these numbers.
When tested in the open game scenes, the Radeon X1950 Pro is slower than its opponent, but not as much as when the transparency antialiasing was disabled. Its speed remains the same whereas the GeForce 8600 GTS has slowed down, especially in terms of minimum speed. The overclocked versions of the card don’t save the day, so you may want to disable the Transparency Antialiasing option in ForceWare to have a comfortable frame rate in TES IV.
There is a significant increase in the average speed of the GeForce 8600 GTS which cannot be only due to the different configuration of our testbed. It must surely be the new driver. The minimum of speed has got lower, however, even though not below comfortable level. Thus, the GeForce 8600 GTS is now preferable to the Radeon X1950 Pro in the mainstream category when it comes to playing X3. This card allows playing at 1600x1200 with enabled FSAA. On the other hand, the Radeon X1950 Pro ensures a higher minimum of speed.
The game having a frame rate limiter, you should consider the min speeds of the cards in the first place. This parameter determines your playing comfort in Command & Conquer 3.
It is the GeForce 8600 GTS that is better in terms of minimum speed although the Radeon X1950 Pro delivers a higher average frame rate. We are talking about 1280x1024, of course, because the min speed of every card, except for the GeForce 8800 GTS 320MB, is not higher than 19fps.
We tested the game in the pure speed mode with enabled anisotropic filtering only, because it has problems when you turn on FSAA.
Speeds have grown higher in Company of Heroes as opposed to the older testbed, especially the speed of the GeForce 8600 series. Yet even enjoying a 50% advantage over the Radeon X1950 Pro at 1280x1024, these cards still cannot make the resolution of 1600x1200 playable. Overclocked to highest frequencies the MSI card has a min speed of only 19fps. The Radeon X1950 Pro offers comfortable gaming conditions in 1280x1024.
The GeForce 8600 GTS performs worse here, perhaps due to its driver or to the transparent texture multisampling. The Radeon X1950 Pro, on the other hand, is never slower than 30fps and allows playing at 1280x1024 with enabled 4x FSAA and adaptive antialiasing.
The MSI NX8600GTS OC has more points than the reference card while the further overclocking helps it score over 12,000 points and get close to the GeForce 8800 GTS 320MB. These scores refer to the resolution of 1024x768 without antialiasing. Let’s better view the results of individual tests instead.
The Radeon X1950 Pro loses to the GeForce 8600 GTS in the second test only. It is equal to its opponent or even outperforms it, depending on the display resolution, in the first and third test. This may be explained by the abundant vegetation rendered in the second test. Coupled with enabled adaptive antialiasing it provokes a performance hit on the AMD card. The MSI NX8600GTS-T2D256E-HD-OC isn’t much better than the reference card but is keeping in second place, behind the GeForce 8800 GTS 320MB, when overclocked “manually”.
The MSI card is only 234 points away from the original GeForce 8600 GTS. When overclocked further, the NX8600GTS OC increases the gap to 643 points and notches a score of over 6,000 points.
The Radeon X1950 Pro seems to lose the SM2.0 tests due to low performance in the second test as we’ll see shortly. In the SM3.0/HDR tests the AMD card outperforms both the ordinary GeForce 8600 GTS and the pre-overclocked version from MSI.
We expected to see the AMD Radeon X1950 Pro lose the second test, but it is inferior to all versions of GeForce 8600 GTS in the first test of this group, too. MSI’s factory overclocking is not very effective due to the small frequency gain.
The MSI card needs additional overclocking to overtake the Radeon X1950 Pro in both SM3.0/HDR tests. The two cards differ by a negligible 0.2fps in the second test. So, everything seems normal here except for the first test from the SM2.0 group.
The results of our test session on the new platform with the latest AMD and Nvidia graphics card drivers showed that the situation in the Mainstream arena has changed dramatically. If Nvidia GeForce 8600 GTS used to look not very impressive against the background of Radeon X1950 Pro, now the situation is very different.
Since we run several different levels for some of the games and also taking into account the minimum performance numbers, Nvidia solution based on G84 GPU used to prove an indisputable leader only in 6 tests, including both 3DMark versions. In other cases, we either saw parity with AMD Radeon X1950 Pro, or a significant advantage of the latter over our hero. Now with the new drivers and settings GeForce 8600 GTS wins in 10 tests, which is more victories than it used to be, even though it is still only a half of all our tests.
Besides, despite a more advanced testbed, we have also detected a performance drop that may have been caused by enabled transparent texture anti-aliasing as well as imperfections of the Nvidia and ATI drivers.
As for the MSI NX8600GTS-T2D256E-HD-OC graphics adapter we talked about today, it made a very favorable impression. First of all, it features very powerful and quiet cooling system, and second, it boasts pretty good overclocking potential. The combination of these two factors makes this solution a very attractive purchase for computer enthusiasts with limited finances. The card is very modestly pre-overclocked by the manufacturer, and doesn’t provide a significant performance gain, but the efficient cooling solution allows pushing its speed to much higher limits for the sake of much greater performance boost. The only thing we wish were better, is the accessories bundle, but in this price range every additional game in the box may change the price tag of the product significantly and discourage potential users from buying it.
All in all, with the price of around $180, MSI NX8600GTS-T2D256E-HD-OC may be very appealing, but if you have another $100-$120 at your disposal we would recommend taking a look at GeForce 8800 GTS 320MB that will provide much higher gaming performance.