NVIDIA GeForce 8800M Preview


8800M Series Line-up & NVIDIA Benchmarks

Banner.jpg

The 8800M series will come in two flavors, a 8800M GTX and a 8800M GTS. Both the GTX and the GTS are based on the G92M GPU and they are functionally identical. They both come with 512MB of GDDR3 and they even share the same clock frequencies. The only difference lies in the number of available stream processors. The 8800M GTX has 6 stream processor clusters for a total of 96 stream clusters (16 stream processors per cluster) while the GTS only has 4 clusters for a total of 64 stream processors.

Both the 8800M GTX and GTS will have a core clock frequency of 500MHz, shader frequency of 1250MHz and a memory frequency of 800MHz. This means that the new 8800M GTX should actually be rather similar in performance to the original GeForce 8800 GTS desktop graphics card, which has the same number of stream processors, shares the same core clock frequency and has similar shader and memory frequencies. However the 8800 GTS has a memory bus width of 320 bits while the 8800M has a more conventional 256-bit bus like its 8800 GT brother, which means the 8800 GTS will have a slight memory bandwidth advantage over the 8800M GTX.

All of these numbers might be a bit hard to follow so we've summarized the main differences between the relevant graphics cards in the table below.


   GeForce
8600M GT
 GeForce
8700M GT
GeForce
8800M GTS
GeForce
8800M GTX
GeForce
8800 GTS
GeForce
8800 GT
 GPU Core
G84M  G84M G82M G82M G80 G82
 Stream Processors
 32 32 64 96 96 112
 Core Clock (MHz)
 475 625 500 500 500 600
 Shader Clock (MHz)
 900 1250 1250
1250 1200 1500
 Memory Clock (MHz)
 1400 1600 1600 1600 1600 1800
 Maximum Memory
 512MB 512MB 512MB 512MB 640MB
512MB
 Memory Interface
 128-bit 128-bit 256-bit
256-bit 320-bit
256-bit
 Memory Bandwidth (GB/sec)
 22.4 25.6 51.2
51.2 64.0
57.6
 Texture Fill Rate (billion/sec)
 7.6 10.0
16.0
24.0 24.0
33.6


As you can see, the GeForce 8800M GTX leaves the 8700M GT far behind and actually has very similar performance specifications to its desktop-bound brother, the 8800 GTS. They have nearly identical clock frequencies and share the same texture fill rate, although the 8800 GTS has the advantage when it comes to the amount of onboard memory and memory bandwidth. It's pretty safe to assume that the GeForce 8800M GTX would perform similarly to the 8800 GTS, however, the performance of the 8800M GTS is more of a mystery. It falls in between the 8800M GTX and the 8700M GT when it comes to performance specifications but actual performance is harder to judge. 

NVIDIA has provided us with a couple of benchmarks that you can see below. We'd like to note that these benchmarks were performed by NVIDIA. We had no control over how the tests were conducted and we only have very limited knowledge of the test setup so you should take these results for what they are worth. However, they should still be useful for getting a rough idea of how the different currently available NVIDIA mobile parts perform with respect to each other. The benchmark results are presented in their original format, the way we received them, although they have been cropped. You can click on each graph to view its original, completely unaltered form.



This first benchmark is a 3DMark06 run at 1280x1024. This benchmark was run on a Windows XP machine. Here, we get our first glimpse at the performance of NVIDIA's new flagship mobile chips. The two 8800Ms completely decimate the rest of the field. The GeForce 7950 GTX and 8700M GT can only challenge the 8800Ms when they are in SLI. That indicates that the new 8800Ms are nearly twice as powerful as anything that came before them.

However, the playing field wasn't completely level. As you can see if you click on the graph to view the original data, the different graphics cards in this test were actually on different test systems configured with different processors. That skews the results somewhat, in the 8800M's favor in this case. The two 8800Ms were in systems with 2.8GHz Merom processors while the 8700M was paired with a 2.2GHz Merom and the 7950 was partnered with a 2GHz Merom. That gives the two 8800Ms the upper hand since they had several hundred more MHz to work with. So does that make these results useless? Not entirely. While we suspect the SLI setups might be able to best the two 8800Ms if they had an even playing field, the new 8800Ms are certainly out of reach of a lone 8700M or 7950 GTX, even if they had equal footing.



In this, and the remainder of the benchmarks, the 7950 GTX has been eliminated. Now all three of the remaining graphics cards are paired with 1.8GHz Merom processors and the 7950 GTX has been replaced by a 8700M paired with a Merom running at 2.8GHz. The goal of this benchmark seems to be to show how much the 3DMark06 results are affected by a 1GHz difference in processor frequency. As you can see, the difference isn't huge. This indicates, as it should, that 3DMark06 is GPU bound. It also tells us that the results in the previous benchmark aren't too far off despite the unequal footing.



Here we see F.E.A.R. at 1920x1200 with 4x Anti-Aliasing and 8x Anisotropic Filtering. It's unclear how these results were obtained but we assume that F.E.A.R.'s built-in benchmark was used. In any case, we see the two 8800Ms clearly dominating the 8700M by a large margin. Both the 8800M GTX and the GTS easily perform twice as well as the 8700M GT at 1.8GHz and they even manage to double the score of the 8700M GT when it has a 1GHz CPU frequency advantage.



The last benchmark NVIDIA provided us is S.T.A.L.K.E.R. at 1920x1200 with 4x Anti-Aliasing and 8x Anisotropic Filtering. Once again, we are not privy to the exact details of the system setups and how the benchmark was conducted, however, we do know the processor that was used and obviously the graphics cards too. Here we see a similar result as with the F.E.A.R. benchmark. Both 8800Ms completely trump the 8700M GT, achieving double its frame rate. Also interesting is the fact the 8700M GT was indifferent to the frequency of the processor, posting the same result regardless of a 1GHz difference in processor frequency.

Related content