The #1 Fire Strike Extreme GTX 1660 Ti GPU score is blatantly physically impossible to achieve.

[operator97]

It is 13,568 points higher than the next best 1660 Ti graphics score (8982). Well over double.

They were certainly running a pretty hefty memory OC, but I see no realistic way that could possibly account for the massively better score, if you look at the often comparable OCs the cards used for various other results were running.

Would be very interested in a developer perhaps providing some input on what exactly the heck is going on in this instance. Hopefully not "unpatched vulnerabilities in the scoring system that people are actively exploiting".

[Huperspace]

https://www.3dmark.com/fs/25145489
lists a Quadro RTX A6000 as a secondary and according to https://www.3dmark.com/search#advanced?test=fs%20X&cpuId=&gpuId=1379&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=overallScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
the score is more the A6000 then the 1660Ti

[operator97]

Originally posted by Huperspace:

https://www.3dmark.com/fs/25145489
lists a Quadro RTX A6000 as a secondary and according to https://www.3dmark.com/search#advanced?test=fs%20X&cpuId=&gpuId=1379&gpuCount=0&deviceType=ALL&memoryChannels=0&country=&scoreType=overallScore&hofMode=false&showInvalidResults=false&freeParams=&minGpuCoreClock=&maxGpuCoreClock=&minGpuMemClock=&maxGpuMemClock=&minCpuClock=&maxCpuClock=
the score is more the A6000 then the 1660Ti

I didn't even notice the second card... seems like this run should certainly not be included in the single-card 1660 Ti rankings at all then, as no one *actually* using nothing but a single 1660 Ti will ever come anywhere close to it.

[operator97]

Also, on further inspection, the results in general for all benches are actually chock full of complete and utter BS... like, what the hell is this:

https://www.3dmark.com/fs/24960913

Graphics score of 25,606 on a 1060 6GB in Fire Strike? Really? Oh, wait, it probably has something to do with whatever the hell the 6GB secondary "Generic VGA" card is there.

TLDR I have like 200% less faith in the overall accuracy of 3DMark results than I did even like 2 days ago. Whatever this weirdness going on with secondary cards is clearly needs fixing. Any score that isn't actually achieved solely by the primarily listed card should *not* be considered valid.

[Na]

Any result with multiple gpu's where SLI/CF is not enabled should be invalid.

[operator97]

Originally posted by Sodium:

Any result with multiple gpu's where SLI/CF is not enabled should be invalid.

For sure. My question is what are they even doing to get the second cards doing work in these cases, though?

All I can think of (for Nvidia cards, at least) is that it must in some way relate to the "CUDA - GPUS" settings in Nvidia Control Panel. Possibly 3DMark isn't actively paying attention to what that's set to / what it actually does for 3D rendering in non-SLI multi-card setups.

[UL_Jarnis]

This is all simple;

Older tests (anything DX11) does not know which GPU was used if system has multiple devices present. This is not related to multiGPU in any way, these GPUs do not co-operate in rendering. It is just a case of one GPU rendering while the result page shows another GPU which did nothing.

This is because for DX11 tests (Fire Strike, Sky Diver and older), our online service makes a guess. If it guesses wrong, things like above can happen. If you report the result links to [e-mail] we'll fix them manually. No need to flip out, simply inform us and we'll sort out the individual cases when they turn up.

New tests (Time Spy and onwards) have GPU selector function that ensures the result file explicitly states which GPU rendered the test. No such issues can occur there. This was added fairly recently so database may still have some old results. We tried to fix most of them when the GPU selector was added, but may have missed some.

Due to the complexity of fixing the old tests we've chosen not to do so - they are for the Windows 7 era and will be eventually moved to unsupported, so the cost/benefit of such a fix was considered and we decided to use the resources on new things instead.

We recommend using latest tests - Time Spy, Night Raid, Wild Life and Port Royal depending on the system being tested - which do not have this limitation.

[UL_Jarnis]

And yes, I've manually fixed the results mentioned in this thread, if you run into others, just report them. Email is fastest, but we do also follow this forum.

[operator97]

Originally posted by UL_Jarnis:

And yes, I've manually fixed the results mentioned in this thread, if you run into others, just report them. Email is fastest, but we do also follow this forum.

I've come across a seemingly different issue, which really just looks more like actual cheating via some manner of BIOS flash / driver hack technique that makes "card X" appear to be "card Y". Here's an image I made that describes it pretty clearly.[i.redd.it]

[UL_Jarnis]

It is a corrupt result.

GRAPHICSTEST1_SCORE = 700.539673

So yeah, it did not run GT1 at 700fps...

Sadly on older DX11 tests runs can very very rarely fail in this way. Report result links to us and we'll flag them.