The new HD6870/6850 bench faster than a HD5870?

[wonderwrench]

I was just looking at the high end video card charts and find it strange that the HD6870/HD6850 have higher scores than a HD5870 or GTX 470. The HD6870 and HD6850 are nowhere near as fast as a HD5870 or GTX 470 in any reviews I have read on the net. Anandtech review, Toms hardware review. I'm guessing this is because the HD6870/6850 are running much higher clock speeds than previous cards and PerformanceTest's 3d test includes windowed 3d and several low resolution 3d tests. Regardless the charts are not reflecting the actual performance of these new cards. I think its time for new 3d tests that more accurately reflect gaming performance of the cards being tested. Can anyone say PerformanceTest 8!

I'm unsure what should be done in the meantime. People that are not tech savvy could easily be mislead by the charts as they are no longer accurate. I downloaded several baselines to compare and find that the 3d simple test results for the HD6870/6850 are much higher than other cards that perform similar in real life scenarios, aka 3d gaming. The DX10 tests are slightly higher but not enough to matter. What would the video card charts look like if the simple 3d tests were dropped? If the chart then reflects the relative 3d performance correctly it would be a usable workaround until PT 8 comes out.

Bill

[David (PassMark)]

We have had a look at the individual results. At least looking at the charts today the GTX 470 is the fastest of the cards you mentioned above.

At the moment we only have a very low number of samples for the 6870 & 6850, so the results will jump around a bit until we get hundreds or 1000s or samples, like we have for the other older cards.

So given that the GTX 470 is currently where you think it should be (at the top), the remaining issue is that the HD5870 is slightly lower than the HD6850/70.

The review you referred to at Tom's hardware doesn't actually contain any benchmarks for the HD5870. So it doesn't directly prove anything. Although I agree that if you imagined that the HD5870 was in Tom's review it might be positioned above the HD5850, which would also place it above HD6850. ().

In the Anandtech review does cover all the cards, and like our charts the HD6850/70 and HD5850/70 are all pretty close. Often within 10% of each other.

While we don't see a 10% variation from current real life games as being a huge problem, it would definitely be nice to get closer. Never going to be perfect however as different games place different loads on the CPU.

There will also definitely be changes in PT8 to the 3D test. For example we would like to use DirectX11.

[Idonno]

I also have to wonder if there is any way to determine whether someone who uploads his results is running a CrossFire/SLI configuration. I am running 2-HD 6870's in CF and I saw nothing in the results to suggest that fact.

If it is the case that PassMark can't see and report a CrossFire/SLI configuration then the ratings would make more sense.

Due to the relatively low cost of the HD 6870's many people (myself included) bought two for a CrossFire configuration instead of a more expensive single GTX 580 with a slightly lower all around performance.

I believe this would account for the discrepancy in the scores.

[wonderwrench]

I also have to wonder if there is any way to determine whether someone who uploads his results is running a CrossFire/SLI configuration. I am running 2-HD 6870's in CF and I saw nothing in the results to suggest that fact.

If it is the case that PassMark can't see and report a CrossFire/SLI configuration then the ratings would make more sense.

Due to the relatively low cost of the HD 6870's many people (myself included) bought two for a CrossFire configuration instead of a more expensive single GTX 580 with a slightly lower all around performance.

I believe this would account for the discrepancy in the scores.

PT only tests single cards in SLI/CF systems so that cant be it.

Bill

[David (PassMark)]

PT only tests single cards in SLI/CF systems so that cant be it.

Unless AMD or nVidia have (finally) released a SLI/CF profile for our code. But we aren't aware of this happening.

[Idonno]

Unless AMD or nVidia have (finally) released a SLI/CF profile for our code. But we aren't aware of this happening.

That would certainly go a long way towards making sense of where the HD6870's rank in PTs test results.