How inaccurate is GPUboss?

[Daniel Z.]

My friend is lecturing me about GPUs from info from GPUboss of all places lol

Edited July 23, 2017 by TheRandomness
Edited at the request of a user.

[The Sloth]

GPU boss sucks

 

CPU boss also sucks 

 

[TheGhzGuy]

About as accurate as your grandmother telling you how to operate a smartphone and how said phone functions. 

[Max_Settings]

If CNN were a GPU review site

[Conductor]

I agree, gpu, cpu, and ssd boss are terrible.

 

[ARikozuM]

I wouldn't trust it at all.

 

Same with UBM unless you know how to read the benchmarks correctly. 

[DrMacintosh]

"GPUBoss recommends the XXX  based on its compute performance and noise and power." 

 

That about sums it up. The funny thing is that it recommends cards with weaker performance than the card you are comparing it to  A LOT. 

[DrMikeNZ]

I keep some dice in a draw, those are more accurate than GPU/CPU boss.

[Daniel Z.]

Just now, charrst said:

GPUBoss, CPUBoss, etc have always seemed a bit sketchy. But on a similar note, I've come to trust UserBenchmark a decent amount. Any comments on whether that is misplaced trust?

they are better

[ARikozuM]

4 minutes ago, charrst said:

GPUBoss, CPUBoss, etc have always seemed a bit sketchy. But on a similar note, I've come to trust UserBenchmark a decent amount. Any comments on whether that is misplaced trust?

You have to choose two items that are similar (a 4th-gen with 6- or 7th-gen Intel or a 4th-gen with Ryzen). Look at the distribution of scores and be aware that the general curve is more important than the "absolute average", i.e. is there a large clump of systems on the left of the curve and a few outliers on the right skewing results? You also need to pay attention to points where the benchmark may have gone awry, i.e. system not performing optimally, and whether you'll actually benefit from the scores being shown, i.e. if you're gaming, you need higher single-core than multi-core performance and an overall balance would be better for creative/CAD suites. Please check that there have been a good number of benches ran. Had a user showing that a 1060 6GB was 1.4x faster than a 970 when the 1060 only had a few thousand of benches. As you can see, there are plenty of tests ran for the 1060, which gives us a more accurate feel for its performance in the synthetic benchmark. 

 

Above all else, use UBM to get a gross analysis and narrow down your results with non-synthetics.