3DMark score variance anomaly

[Tan3l6]

Testing with 3DMark Fire Strike I get strange variance in final score. It's around 300 points variance around 41700 or 38700. Not anything in the middle.

The comparison shows mainly CPU and combined scores being 20% apart. Same machine different runs, and repeatable. Sometimes it's 41k other times 38k.

What might cause it?

Spoiler

 

System specs:

Spoiler

 

[alyen]

Is the system doing anything in the background?

Could the CPU or GPU be throttling? How much time between each run?

I haven' bother to use Fire Strike for so long.

 

[Mark Kaine]

honestly,  while i really like firestrike... the whole benchmark suite doesn't seem too reliable,  ie there are definitely some bugs...

 

its also more useful to find actual issues (eg. overheating etc) than to say "oh look how good my scores are"... that is generally utterly irrelevant in regards to how a system performs in actual workloads. 

 

 

That said, i would try some other benchmarks,  simply to see if there's still an issue...

 

Prime 95 for cpu, superposition for gpu... (typically max or close to max settings) 

 

And if that still isnt conclusive,  my fav: Prime95 + superposition *at the same time* (just keep your temps in check and be ready to abort, because, man, its gonna be *really* hot! )

 

 

just a suggestion... what i would do... to see if there really is a problem...

 

*edit: and by abort i mean prime95, cause that is likely the main cause for heat and power draw in my experience.

[Mark Kaine]

ps: also post a link to the firestrike results? im curious to check a few things  

[Tan3l6]

12 hours ago, alyen said:

Is the system doing anything in the background?

Could the CPU or GPU be throttling? How much time between each run?

I haven' bother to use Fire Strike for so long.

 

System is idle in both cases. Not throttling - temps are fine.

One for the instance I ran one day, other the day after.

[Tan3l6]

8 hours ago, Mark Kaine said:

ps: also post a link to the firestrike results? im curious to check a few things  

41783 
 

38707

[Mark Kaine]

3 minutes ago, Tan3l6 said:

41783 
 

38707

I'll look a bit more into it, but this just seems like normal benchmark variance to me?

 

if you look at it overall this is a difference from 98% to 99%, so literally 1%.

 

this is similar to like cinebench, that can vary easily about 100-200 points too between runs, for no apparent reason. 

 

 

on the surface it all looks completely normal.

 

 

 

 

[Mark Kaine]

 

 

the result with lower score has the cpu about 3c hotter and clocked 100mhz lower... that is literally it... normal boost behavior...

 

and thats why i said these benchmarks dont necessarily tell you much, even when they're fairly accurate. 

 

 

so if you do this next time make sure the temps are the same between  runs, then the results should also be very similar! 

[Tan3l6]

7 minutes ago, Mark Kaine said:

on the surface it all looks completely normal.

Yeah, but the 20% of difference in combined test seem odd to me. And the difference is always 20%

The total score difference is 3000 points - that also seems a tad too much for a margin of error.

But In general I don't notice anything massively different in games, so it's not a critical problem.'

 

  

2 minutes ago, Mark Kaine said:

the result with lower score has the cpu ablut 3c hotter... that is literally it... normal boost behavior...

 

and thats why i said these benchmarks dont necessarily tell you much, even when they're fairly accurate. 

 

 

so if you do this next time make sure the temps are the same between  runs, then the results should also be very similar! 

The GPU temps should affect more the GPU score though...

[Mark Kaine]

Just now, Tan3l6 said:

Yeah, but the 20% of difference in CPU and combined tests seem odd to me. And the difference is always 20%

The total score difference is 3000 points - that also seems a tad too much for a margin of error.

But In general I don't notice anything massively different in games, so it's not a critical problem.

seems to depend on temps, see above,  there's a 3c/100mhz boost diff. 

[Mark Kaine]

11 minutes ago, Tan3l6 said:

The GPU temps should affect more the GPU score though...

 

yeah, but there's still this temp temp diff... 

 

seems really like totally normal variance to me, its not even margin of error,  its simple boost behavior. 

 

this is why people use LN2 when going for scores, the temp variance should be minimal in that case .