Out of the many new parts that NVIDIA recently released, consumers have paid particular attention to the company's new line of mainstream graphics cards. Although NVIDIA officially released these cards last month on the 17th, there was so much going on at the time in terms of product releases that we weren't able to give the cards as much attention as we would have liked. Taking this into consideration, we have decided to revisit the GeForce 8 mainstream video cards and see the current state of the market.

The launch of the NVIDIA mainstream cards last month wasn't met with the same enthusiasm shown towards the GeForce 8800 when it launched back in November. Expectations were high for the new midrange cards and many expected them to show similar performance levels, naturally scaled down to fit the price range, as the GeForce 8800. The performance of the new cards, however, ended up being considerably lower than expected and thus many viewed the newcomers as poor performers.

Many also feel that the technical characteristics of the new cards are inadequate for the mainstream market segment. The 128bit memory bus seems rather small compared to the 384 bit bus width the 8800 GTX comes with or even the 320 bit bus of the 8800 GTS. The amount of stream processors, though, falls victim to the largest criticism. The GeForce 8600 line of cards only comes with 32 stream processors. Comparatively, this is a humble amount when looking at the 128 stream processors of the GeForce 8800 GTX and the 96 of the GeForce 8800 GTS. In truth, much of this criticism brings up some valid points, and we'll revisit them in the conclusion of this article.

In this roundup we will be analyzing 10 mainstream cards that are based on the NVIDIA GeForce 8600 GTS, 8600 GT, or 8500 architecture from six different manufacturers. Below is a table of the solutions we are taking a look at:

The ASUS GeForce 8600 GTS has the highest operating frequency of the 10 cards we are analyzing. The Gainward GeForce 8500 GT Gold Edition comes with the lowest overall clock speeds, the deciding factor being memory frequency. In addition, the Gainward GeForce 8500 GT is the only video card in the article that uses DDR2 memory.

We tested a total of two XFX cards in this article. Although we already looked at these cards in a previous article, we included them for completeness and to ensure that readers are given a larger variety of performance numbers to look at.

The 8600 GT uses the same PCB as the 7600 GT. This means that compared to the XFX GeForce 8600 GTS card it is slightly smaller. Because its power requirements are lower, the card does not require an additional 6 pin PCIe connector.

The 8600 GTS has a brace to strengthen the PCB. The cooling solution of the card is simple and based on the NVDIA reference design. There is a 6 pin PCIe connector for the 8600 GTS that provides additional power as the slot cannot provide all the power this card requires.

Looking at the technical characteristics, the XFX 8600 GTS has a faster clock speed than the XFX 8600 GT. The XFX 8600 GTS comes overclocked at 730 MHz for the GPU, and 2260 MHz memory. The 8600 GT, on the other hand, has a clock speed of 620 MHz for the GPU 1600 MHz for the video memory. Keeping in mind that clock speeds ultimately do affect power draw, the lower clock speeds of the 8800 GT might potentially result in lower power consumption as a whole for the card.

The ASUS 8600 GTS is nearly identical to the NVIDIA reference board. The only difference that sets it apart is the S.T.A.L.K.E.R sticker on the thermal solution. ASUS has made no other physical changes to the reference board and the cooling solution is completely standard. The memory frequencies of the card, though, are pretty interesting. The GPU is clocked at 745 MHz, the default clock being 675 MHz. The memory, on the other hand, is clocked at 2290 MHz, the default clock for the 8800 GTS being 2000 MHz.

The board layout for the ASUS 8600 GTS is exactly the same as the XFX GeForce cards. The only thing done differently is the graphic on the cooling solution, which is entirely dedicated to S.T.A.L.K.E.R. The only area that the ASUS logo is even visible is on the GPU fan. As can be seen, the board does have a 6 pin PCIe connector.

The board layout for the ASUS 8600 GTS is exactly the same as the XFX GeForce cards. The only thing done differently is the graphic on the cooling solution, which is entirely dedicated to S.T.A.L.K.E.R. The only area that the ASUS logo is even visible is on the GPU fan. As can be seen, the board does have a 6 pin PCIe connector.

It is apparent that Foxconn's priority when designing this card wasn't personalization. The Foxconn logo is literally nowhere on the card.

Foxconn GeForce 8600 GTS

Gainward submitted three video cards for this test: one GeForce 8600 GTS, one 8600 GT, and one 8500 GT. All of the cards are from Gainward's Golden Sample line of cards, which means they come factory overclocked.

Externally, all three cards seem identical. They are very similar to each other and sport the same design. Looking closer, though, slight differences in the design are noticeable. The 8500 GT is the most apparent, since instead of having two DVI ports it has one DVI and one VGA. The two 8600 GT and GTS cards, on the other hand, have 6 pin PCIe power connectors. In any case, though, the design is completely original.

Underneath the cover is the heatsink, cooling fan, GPU, and video memory. The operating frequencies of the cards are as follows:

- Gainward GeForce 8600 GTS Golden Sample: 725 MHz GPU and 2200 MHz memory

- Gainward GeForce 8600 GT Golden Sample: 600 MHz GPU and 1600 MHz memory

- Gainward GeForce 8500 GT Golden Sample: 600 MHz GPU and 1400 MHz memory

The MSI NX6800GTS has the most unique cooling solution of all the cards we're reviewing today. The thermal design also happens to be quite large, and so the card ends up being a dual slot solution. Kindly enough, MSI includes a copy of Company of Heroes with the card. As indicated by the “OC” moniker, the video card has been designed with overclocking in mind.

The card makes use of a heatpipe to cool the GPU and move heat as far away from the GPU as possible. The heatpipe is circled in the photograph above.

The Gigabyte GV-NX86S256H uses a passive cooling system and is based on the 8600 GTS chip. This is another unique cooling solution, largely because it is completely silent. The card has a large heatsink that nearly covers the entire surface. Two heatpipes emerge from the GPU and lead to the radiator. Gigabyte has designed the cooling solution to move heat as far away as possible from the GPU and ensure that heat is properly dissipated. Since the cooling solution is completely passive, it can only be expected that the card comes clocked at reference speeds. The speeds are 675 MHz for the GPU and 2000 MHz for memory.

The card requires two free slots for proper installation since the thermal design is quite thick. Gigabyte calls the passive cooling solution on this card Silent-Pipe 3 because they claim it integrates three Gigabyte thermal designs.

The Gigabyte GV-NX86T256D is based on the GeForce 8600 GT. Since the 8600 GT generates less heat compared to the 8600 GTS, Gigabyte decided to use its slimmer Silent-Pipe II thermal design. Although it is hard to see, the heatpipes used in the heatsink are visible. The card comes clocked 600 MHz for the GPU and 1440 MHz memory.

In order to test the performance of the GeForce 8600 GTS, GT and 8500 GT cards, we used a motherboard based on the NVIDIA 680i SLI chipset. The processor of the test platform is an Intel Core 2 Extreme X6800 Socket LGA 775. The Core 2 Extreme X6800 is a dual-core chip with a clock speed of 2.93 GHz. The total memory of the test system was 2 GB. We used two G.Skill F2-6400CL4D-2GBHK memory modules, each with a capacity of 1 GB. The operating frequency of the RAM was 400 MHz and had timings of 5-5-5-15. The HD was the WD1600JS, which uses an SATA interface and has a total capacity of 160 GB with a 7200 RPM. The power supply unit for the test system was an Enermax Infinity 720 Watt unit.

Test Platform

• motherboard: Foxconn C55A01 ( nVIDIA nForce 680i SLI chipset, Socket 775 LGA)
• processor: Intel Core 2 Extreme X6800 2.93 GHz
• memory: G.Skill F2-6400CL4D-2GBHK, 2x1 GB
• power supply unit: Enermax Infiniti 720W
• operating system: Windows XP SP2
• NVIDIA Forceware Win XP 32 bit International 158.16 Drivers

Benchmarks were run at 1280x1024, 1600x1200, and 1920x1200 except for Oblivion, S.T.A.L.K.E.R., and Battlefield 2142. For games that lacked integrated benchmark utilities, we used FRAPS to record performance. In all benchmarks anisotropic filtering was set at 16x.

In the 3D Mark benchmark the clear winner are cards based on the 8600 GTS. They hold a visible advantage over cards based on the 8600 GTS and the 8500 GT. The ASUS 8600 GTS
outperforms all of the cards in its class, largely because of its 745 MHz GPU clock. The standard GPU frequency of the NVIDIA 8600 GTS 675 MHz. The performance difference between the highest performing 8600 GTS and GT card is 10%.

Shadermark emphasizes that the 8500 GT has much weaker performance than other tested card models. This can be seen in the line graph where the 8500 GT solution stands quite alone. For the most part, all other cards remain in relatively the same score range.

click on the image to view the picture at 2560x1600. File Size: 3.28 MB

click on the image to view the picture at 2560x1600. File Size: 2.55 MB

Half-Life Episode 1 is the latest expansion pack of Half-life 2. It is distributed through Steam, Valve's content delivery system. The benchmark is run at 1280x1024, 1600x1200 and 1920x1200 with anti aliasing at 4x in one test and HDR enabled in the other. Both tests have anisotropic filtering set at 16x. The benchmark is run using the HOCbench PC-Space demo.

click on the image to view the picture at 2560x1600. File Size: 2.66 MB

Prey is based on a modified version of the Doom 3 engine. We test using hocbench with the hwzone demo on a retail version of the game patched to version 1.3. The chosen resolutions are 1280x1024, 1600x1200 and 1920x1200. All testing is done with 4x anti aliasing and 16x anisotropic filtering.

The ASUS GeForce 8600 GTS is the lead performer in the Half Life 2 benchmarks. The main performance difference between the GTS cards can be attributed to clock frequency. The ASUS card
has the highest operating frequencies, and thus it scores 10-16% higher in the benchmarks compared to the numbers obtained by the standard clocked cards, which are the Foxconn and Gigabyte 8600 GTS boards.

The GeForce 8600 GT solutions offer performance comparable to standard clocked 8600 GTS cards at all resolutons, even with HDR enabled at 1600x1200. The same cannot be said for the 8500 GT video card. It isn't able to offer fluid gameplay at higher resolutions with filters enabled.

With filters enabled and 4x anti-aliasing and 16x anisotropic filtering Prey proves to be too overbearing for the cards in the test at resolutions of 1920x1200 and 1600x1200. Performance numbers are different for 8600 GTS solutions at 1280x1024. Again, a general trend we've been seeing is that the fastest cards receive the highest scores. For the 8600 GT and 8500 GT cards the tests indicate that either filters or resolution would have to be toned down to play at a more comfortable frame rate.

click on the image to view the picture at 2560x1600. File Size: 2.55 MB

This benchmark is conducted using a retail copy of the game patched to version 1.08. We run tests at 1280x1024, 1600x1200 and 1920x1200. Anti aliasing is set at 4x with 16x anisotropic filtering in one test, the other has soft shadows enabled and AA/AF disabled.

Similar to prey, F.E.A.R. is another game that demonstrates the trouble for mainstream cards in terms of performance. The frame rate is quite low in all tests, and no card is able to run the game at smooth frame rates at 1920x1200 or 1600x1200. The only viable option for smoother performance is to lower filters. The minimum scores of the test show that a huge bottleneck is present at higher resolutions. This can most likely be attributed to the amount of memory the cards have.

Enabling soft shadows and turning off AA and AF only repeat the scene of the previous test. Scores are higher compared to the previous benchmark, but in any case it is still not an ideal frame rate.

click on the image to view the picture at 2560x1600. File Size: 2.09 MB

For the Splinter Cell - Chaos Theory benchmark we test at resolutions of 1280x1024, 1600x1200 1920x1200 and 2560x1600. The benchmark uses the Ubisoft LightHouse demo and has HDR and anisotropic filtering at 16x enabled.

We again see a direct performance-GPU frequency relationship in this test. It is a general trend we have been seeing and is most obvious with the GeForce 8600 GTS cards. Since the 8600 GT cards have the same operating frequency their performance is aligned with each other. The 8500 GT once again receives the lowest score. It is only expected since it has the most modest technical specifications of the bunch.

Oblivion doesn't have an integrated benchmarking utility. To test performance, we use FRAPS in this benchmark.

click on the image to view the picture at 2560x1600. File Size: 3 MB

In Oblivion there is a clear difference between the performane of the 8600 GTS cards and those of the 8600 GT family. The test highlights a large performance difference between the two card classes.

The line graph shows how the cards perform over the 30 second test. Again, the difference between the 8600 GTS and GT cards is quite apparent.

S.T.A.L.K.E.R. is one of the newest additions to our testing lineup. It doesn't have an integrated benchmarking utility so again we have to use FRAPs. In this test we do not enable AA, as it would be too stressing for these cards. AF, though, is set at 16x. The test consists of a 26 second scene in which the character walks through a number of houses surrounded by rich vegetation.

click on the image to view the picture at 2560x1600. File Size: 658 KB

This test is by far one of the most stressing. We see nearly all cards suffer in terms of performance. At 1280x1024 the GeForce 8600 GTS cards have a minimum score between 18 and 21. Performance obviously gets worse moving to lower models. At 1600x1200, performance nearly completely collapses. This can be attributed to the small amount of memory.

click on the image to view the picture at 2560x1600. File Size: 2 MB

In this benchmark we see performance improve slightly for the 8600 GTS cards compared to the previous benchmark. The 8600 GT also performs at acceptable frame rates, however, the minimum frame rate is what causes most of the trouble. The 8500 GT scores the lowest, and the game isn't playable with it at 1600x1200 and 1280x1024 is borderline.

For space reasons, we are not able to show the entire duration of the benchmark. This is why the minimum values are not immediately obvious on the graph.

We have measured operation noise, GPU temperature and consumption of the analyzed cards in our usual method. Consumption is measured in three states: idle, with the CPU under stress using Orthos, and CPU/GPU under stress using Shadermark along with Orthos. Temperature is also measured in these three states. For the noise test we place a noise meter approximately 10 cm from the GPU fan. In order to reduce any errors, we temporarily disable all other system fans.

Based on our measurements the Gigabyte and Gainward 8600 GTS consume the most power out of all the GeForce cards. What raises a few more questions is that neither of those two cards are overclocked.

The two Gigabyte cards have the highest operating temperatures, but the data isn't all that negative considering that both cards are passively cooled. The Gainward thermal designs are not particularly efficient, as they immediately occupy the three spots right after the Gigabyte cards. The MSI card has the best thermal design, making it an ideal card for overclocking without needing to change the cooling solution. Its temperature raises only 7 degrees going from idle to stress.

The effectiveness of the MSI cooling solution has a price in terms of noise, even if the value is not all that distant from the other competing video cards.The Gaiward cards are particularly quiet as Gainward seems to have given this aspect priority over actual temperatures. Obviously, the Gigabyte cards are not part of the tests as they are completely silent thermal designs.

With its new mainsteam cards NVIDIA has been quite successful in redefining the performance point of the mainstream class. Looking back at previous versions of the card, it is apparent that the gap between the mainstream and enthusiast class has grown significantly. The large gap between enthusiast and mainstream cards can be ascribed to the overall lower performance of the G8x mainstream parts.

By taking a look at our performance benchmarks today, it is easy for us to say that the G8x mainstream range of cards would not really work well for an extreme enthusiast. For an enthusiast with a budget in mind, we strongly suggest that the lowest GeForce card that should be bought is a 320 MB 8800 GTS. Any GeForce part other than that would probably not suit an enthusiast well, especially if the card is meant for gaming a high resolutions with quality settings at their maximum. The 320 MB 8800 GTS would probably be the card of this generation that best matches the "historical performance" of the mainstream class.

For those of you who are willing to compromise a little and feel the new mainstream cards would be a good match for you, it is hard for us to establish a clear winner out of the cards we reviewed today. While for enthusiast cards manufacturers always embark with the undertaking of achieving the highest performance, the focus of mainstream cards are usually somewhat different. Don't take us wrong, manufacturers still focus on performance, however, it is a smaller priority than in the high end market. Instead, manufacturers like focus on other features and aspects of the video cards.

In terms of raw performance, the ASUS GeForce 8600 GTS obtained the highest scores in most of our benchmarks thanks to its overclocked GPU. The ASUS 8600 GTS held an average advantage of 10-13% compared to other cards in its class clocked at the NVIDIA base frequencies. Obviously, the benchmarks favored the card with the higher clock. Surprisingly enought, despite its higher operating frequencies, the card ended up having one of the lowest overall power consumption scores out of its class, especially at idle.

If you're not looking for raw performance, though, the cards we have tested today also come with some unique characteristics. The Gigabyte mainstream cards we took a look at all are equipped with a passive thermal design, making them completely silent and ideal for HTPC use. The silent cooling soluiton, though, means that overclocking wouldn't be a very good idea for the card, meaning that resolution and filters would probably have to be set at lower resolutions for fluid gameplay.

The MSI GeForce 8600 GTS, for its part, is also quite an interesting card. The cooler of the MSI NX8600GTS, in fact, comes out to be the most effective cooler out of the entire lineup of mainstream cards we tested in this article. This does, however, have a slight drawback as the cooler also turns out to be the loudest. The thermal headroom of this card, though, makes it very ideal for overclockers.