lo all,

 

we have all experienced the drop in Physx score when running high CPU overclocks. I had a quick session yesterday as I got hold off three 980 cards and the air pretesting at 4.6GHZ CPU OC gave me a higher Physx score as when running the CPU at over 5Ghz.

 

It seems the sweetspot for a good Physx score is around 4.9GHz, though we have seen some top benchers having decent outputs at 5.4 and beyond.

 

Physx scores:

 

• 4.6GHz was around 21.5K
  
• 4.9GHz score is 22.6
  
• 5.3Ghz score is 21.5K
  

 

The combined score is also with hit and miss... The 3D11 benchmark is not stable in output for me...

 

3D11 version was version 1.0.1 and the latest System Info 3.338 installed

 

What I've tried:

 

• Running different memory dividers ranging from 2600-3000MHz: no go
  
• Running different Uncore dividers from 4000-4600Mhz: no go
  
• Tried to run the Physx test before without any GPU OC, than OC GPUs and do a full run. The Physx score drops 200-300 points during a full run.
  

 

What still to do ? Do we downclock the CPU OC for the Physx Test and than up again for the combined one ? Is it some software setting or the new System Info that hammers the score. Too bad this part of the test is so determinant for the final score. A good Physx/Combined score easily makes up for 50-100MHz less on the GPU clocks

Edited December 30, 201410 yr by Leeghoofd

A possible solution is to downclock the CPU only for the first few seconds of the physics test and then put the clocks back up. Although I've seen people do hassle-free full runs at 5.7GHz++ so it must be fixable.

Well if you look at the bot many scores with high clocked CPUs have a weak Physx score Sam, even with some of the top russian OCers it is an issue... Maybe Petrovic has a fix

Well if you look at the bot many scores with high clocked CPUs have a weak Physx score Sam, even with some of the top russian OCers it is an issue... Maybe Petrovic has a fix

 

It only happens with multi card set ups, you have to do as Sam says and down clock for the the start then jump it back up. It's usually 2 or 3 multis to get it to score right. :celebration:

so quickly set eg Multi at 49, than during the PhysX run up again to normal speed ?

This worked for me: Open benchmark and run physx test only. After it's done run the complete test.

Tried that Roman, think with the new system info we are s c r e w e d; Difference between a seperate run and the full run is 200-300 points less for the full one... but I'm scoring over 1.5K too low anyway for the CPU clocks

Edited December 30, 201410 yr by Leeghoofd

so quickly set eg Multi at 49, than during the PhysX run up again to normal speed ?

 

Exactly. :celebration: pause feature of Asus boards FTW.

So all is good with single card?

 

Here's mine with cpu/cache at only 4.5 / 4.375, 25k should be doable at 4.6 I think so something is wrong for sure.

 

http://hwbot.org/submission/2705326_l0ud_sil3nc3_3dmark11___performance_geforce_gtx_980_23526_marks

So all is good with single card?

 

Here's mine with cpu/cache at only 4.5 / 4.375, 25k should be doable at 4.6 I think so something is wrong for sure.

 

http://hwbot.org/submission/2705326_l0ud_sil3nc3_3dmark11___performance_geforce_gtx_980_23526_marks

 

I've not seen the problem with a single card.

 

If it stutters in the slightest at the start, that will be enough to drag the score down.

This issue has been around for a long time. I remember having the same issues trying 4Way at Campus Party in Berlin in 2012.