Rauf

Nah, card is pretty much maxed out in Port Royal.

Maybe need to rebench this on cuda 10 sometime.

Thanks!

I suggested something like low end, mid end and high end categories for gpu benchmarks many times during PJ's rule. But there was never any interest in implementing it. Maybe something for the future? Evc2 should open up many non xoc cards for playing with voltage and/or power limits. On the other hand. Prices are what they are. Same for cpus, many are priced crazy high, and cpus require way more binning than gpus. If the cost is very high, and reward points wise is low, there is little incentive to bench it. I'm not saying 3d should be worth more points than 2d. It's impossible to decide which bench is harder or require the most effort. I think all benchmarks should be valued the same and awarded the same points. But the overall number of rankings you can compete in for 3d are way lower than 2d, so it would be nice to see a few more 3d benchmarks for points. Or introduce low, mid and high end categories

As far as I know the points distribution will be adressed in the upcoming changes. Some benchmarks have had strange points boosts since last rev, like 03, gpupi for cpu etc. But this will surely be fixed.

These benchmarks are all eol. 3dmark (FS and newer) suite does not have the timer issues, which means that the problem is not the sysinfo itself. And there is little chance that UL fixes eol benchmarks. Also, we allow lod and tess, so we don't follow the same rules anyway. I don't see any other benchmark here that works 33% of the time or so. Please understand that benching 3d11 for example is pretty challenging. You need high cpu clocks, high ram which often is sensitive to cold. Then you have gpu, which if you are lucky can do full pot. But you only have a short while to bench before vram gets too cold because of full pot temps. And you might crack after a short while even if you are succesful in getting to full pot. Now let's say you manage to pull everything of the above off. Maybe have remounted 2, 3 or 4 times to try to get to full pot. And once you do your score is invalid because of an issue that isn't even a real world problem. This problem applies to 3 out of 10 global gpu benchmarks. If you bench for hw points there are at least 7 benchmarks that have this issue. For me this is a big problem.

Nice!

Depends on how you define "works". Links with lod are no longer public, which means they have no real value. Sure you can screenshot the validation page yourself, but we already provide one screenshot. If one can be faked then why not the other? And then we have the issue where 2 out 3 amd runs will result in false positive for timer issues. So in reality you need luck to get a valid score. I don't think this situation is working very well. Is Benchmate an option if you still want more security than screenshots? Talking everything from 03 to 3d11.

Load time is not dependant on storage performance. It's all tweak and cpu perf. And it's only a small part of the score anyway.

What version do you mean? Sysinfo or benchmark? Because benchmark version is not conclusive, just which major subversion. For instance 1.0, but not which 1.0.xxx

After seeing the new benchmarks in place for a while now. And benching 3D pretty heavily lately I have some feedback regarding globals. It's pretty clear the total number of relevant rankings/points is waaay higher for cpus. Just counting the most popular rankings for Geekbench 3 multi there are more than all 3d combined (counting for 3090/6900xt where sli is almost non existant and so rare on hwbot that it does not give points anyway). In this regard I think we need more 3d benchmarks with globals enabled. Lets bring back some of the removed benchmarks If catzilla is fixed so that it actually runs with mb/cpu drivers installed, why not. Also, maybe consider setting fixed points for sli categories as they are very rare here on hwbot. But in 3dmark hof they are still filling up the top 100.

What is the point of running systeminfo on legacy benchmarks where lod and tess is allowed anyway? 01-06 ran just fine without sysinfo for years. But for everything except 01 with vali link for top scores at least. Just make sure they are run on platform and/or OS that does not have timer bug = No issues. System info on amd for legacy benchmarks is a mess... All years benching 03-06 and vantage and 3d11 I have never seen any issues running on old sysinfo or completely without. Why put your trust in sysinfo that is not made for these old benchmarks. Most likely they won't catch anything anyway. Like 03 running black screen, but fps in subtests will give it away.

Great scores!

Great score and great gpu!

Well done! Nice to see some proper 3D benching and putting those KP to use.

Another one, great card!

Nice clocks! Try win 7 next time

Will I get 200 votes for $100? I think I will vote hard in the 2D categories Pcmark 04, pcmark 07, pcmark vantage etc incoming

Very strong card, well done!

I know. But if result details (subtests fps) are not needed maybe rules should say "benchmark window with result". Or maybe just benchmark window, I think most will understand what part of it is needed But for the legacy 3d we still need details. Because often vali links will not be visible. I think it will look and feel simpler with the new layout and information

Some thoughts: Example screenshot with red boxes for the important parts. We need to know where lod is allowed and not allowed. Don't like gpu-z monitoring is mandatory. It's too much for the noobs. All the pros do it anyway. But many times it will not be correct and most benchmarks log it in vali file anyway. Max boost clocks don't say much in many circumstances and is easily tricked if anyone wants to. Result details are not possible with new 3dmark layout.

Btw, vantage extreme does not scale much at all with cpu score. Isn't Intel better?

Nice!

Crotch pic :)

So, I ran my 6900XT on LN2 and EVC2 first time yesterday. Ended up being a pretty long session because as others have reported there are quite a few things to balance to get good efficiency. It does not get easier on LN2 For me MPT worked fine in combination with EVC2 power adjustments. Like chispy reported you have to balance volts and clocks because those two decide power, and the power limit is the biggest problem when benching these cards. But on LN2 you also have to factor in temperature, going to cold can hurt your score. But the biggest factor is voltage. You will have to play with voltage set in AMD-driver in combination with offset in EVC2. To be clear (or confusing maybe...) 1050mV set in driver plus 100mV in EVC2 is NOT the same as 1100mV set in driver plus 50mV EVC2! These two settings, even though actual voltage is similar, will make a huge difference in performance. Also, the voltage and offset that works for FS does not work for FSX and yet another combination is needed for FSU ? And then we have the factor that good voltage combination for game tests in Fire strikes does not always mean you will have good combined! This is especially true for normal Fire strike.