Question Radeon 7900 XTX gets very low precentile in Userbenchmark

[JayGau]

I saw something strange today when testing my PC with Userbenchmark. I got my Radeon 7900 XTX about 2 weeks ago and everything looked fine so far. My temperatures are good (it's not an overheating one) and I score slightly above average in 3DMark Time Spy. But for some reason Userbechmark says that the GPU is in the 6th percentile. How can I score better than average in Time Spy and at the bottom of the curve in Userbenchmark?

I know we should not rely too much on this benchmark since you never really know against what it compares you but I have never got a so low percentile with a GPU on this thing. I am not super worried since Time Spy gives me a good result but I would like to understand why the Userbenchmark GPU score is so low. The percentiles for the rest of the hardware are good by the way, only the GPU poorly benches.

 

[USAFRet]

but I would like to understand why the Userbenchmark GPU score is so low

Because UBM is a badly flawed tool.

https://ownsnap.com/userbenchmark-biased-admin/
https://www.tomshardware.com/news/userbenchmark-benchmark-change-criticism-amd-intel,40032.html
https://linustechtips.com/topic/1291540-userbenchmark-is-completely-biased-or-bs/

 

[drivinfast247]

I saw something strange today when testing my PC with Userbenchmark. I got my Radeon 7900 XTX about 2 weeks ago and everything looked fine so far. My temperatures are good (it's not an overheating one) and I score slightly above average in 3DMark Time Spy. But for some reason Userbechmark says that the GPU is in the 6th percentile. How can I score better than average in Time Spy and at the bottom of the curve in Userbenchmark?

I know we should not rely too much on this benchmark since you never really know against what it compares you but I have never got a so low percentile with a GPU on this thing. I am not super worried since Time Spy gives me a good result but I would like to understand why the Userbenchmark GPU score is so low. The percentiles for the rest of the hardware are good by the way, only the GPU poorly benches.

Do not use UB as a comparison tool! I bet with the XTX being a new card there are few benchmarks in the database for comparison. And the few in the data are heavily overclocked and using exotic cooling.

 

[JayGau]

Because UBM is a badly flawed tool.

https://ownsnap.com/userbenchmark-biased-admin/
https://www.tomshardware.com/news/userbenchmark-benchmark-change-criticism-amd-intel,40032.html
https://linustechtips.com/topic/1291540-userbenchmark-is-completely-biased-or-bs/

Yeah and it's why I wrote that we should never rely much on it. I actually haven't used it for a while because of that. I was just curious to see how my brand new card would do on it and I have never got a so low score but maybe I should just ignore it. I also checked a couple of reviews for the 7900 XTX and compared my score on Fire Strike Ultra with what they got and I score as good or better than they did. So something is obviously wrong with Userbenchmark.

 

[USAFRet]

Yeah and it's why I wrote that we should never rely much on it. I actually haven't used it for a while because of that. I was just curious to see how my brand new card would do on it and I have never got a so low score but maybe I should just ignore it. I also checked a couple of reviews for the 7900 XTX and compared my score on Fire Strike Ultra with what they got and I score as good or better than they did. So something is obviously wrong with Userbenchmark.

Plus, understanding what that "percentile" actually indicates.

example:
On a scale of 1-100, ALL tested components score between 91 and 100%
Most fall in 94-95-96%

Would you call something that scored at 91% as "bad"?
No, you wouldn't.
Yet UBM would call that 10th percentile or lower.

"OMG!! Something is broken!!!"

 

[JayGau]

Do not use UB as a comparison tool! I bet with the XTX being a new card there are few benchmarks in the database for comparison. And the few in the data are heavily overclocked and using exotic cooling.

That was one of my thoughts too. The stats are most likely not representative at all for the moment because the card got release only a month ago and if I remember well, there were only somehting like 330 results when I ran it this morning. If many of them are ultra enthusiasts who overclock as hell those numbers don't mean anything.

 

[JayGau]

Plus, understanding what that "percentile" actually indicates.

example:
On a scale of 1-100, ALL tested components score between 91 and 100%
Most fall in 94-95-96%

Would you call something that scored at 91% as "bad"?
No, you wouldn't.
Yet UBM would call that 10th percentile or lower.

"OMG!! Something is broken!!!"

That's absolutely true.

 

[norcalsc]

Do not use UB as a comparison tool! I bet with the XTX being a new card there are few benchmarks in the database for comparison. And the few in the data are heavily overclocked and using exotic cooling.

this

 

[zx128k]

Like using the top 100 scores from 3DMark Time Spy for checking stock system performance. Then wondering how the CPU temp is -40c. Userbenchmark really love AMD btw. As pointed out the RX 7900XTX has 392 samples at the moment. 4090 has 10k samples.