Higher overclock = Lower 3DMark11 score?

paradigm

paradigm

paradigm

Caporegime
Joined
26 Aug 2003
Posts
37,514
Location
Leafy Cheshire
I'm just playing with my new i7-3820 build, and am seeing how quick I can get it, whilst remaining stable, relatively low voltage, and keeping load temps below 60c.

I've noticed that my 3DMark11 scores seem to nose dive after 4.75GHz (I've tried 4.875GHz, 4.96GHz, and 5GHz, all the same), the benchmark runs through OK, and tools like OTTC, IBT, and superpi/prime are OK also, but my 3DMark11 scores seem to get worse.

It's mainly the graphics tests (which I find odd, as they don't utilise the CPU), the Physics test seems to go up (CPU only), and the Combined test drops slightly.

I can't work out what I need to change (voltage or settings wise) to keep the scores on the up?

:confused:

Anyone shed any light?
 
Still royally confused by this. I've changed some of the voltages and settings, then done some long IBT runs (100 runs on HIGH), and it's quite happily sitting there, under 60ºc without any problems at all.

3DMark11 however still drops marks based solely off GPU performance, which leads me to think it's a BCLK problem, but according to the specs, changing the CPU Strap to 125 should leave every other bus at their stock values? So in theory it shouldn't be related?

:confused:
 
A friend had a very similar problem, he's said that for him turning off all temperature monitoring (VRM etc) sorted it - apparently it runs very hot on X79.
 
GPU clocks are slightly lower on the low result, not by much though so unsure how much effect that would have. Not sure what else it could be :(
 
I have the R4E board so things may be different with that.But when you change buses you must change memory speed also yes??? 100 blk gives 1600,1866,2133,2400 etc etc divider 125 gives 1666, 2000, 2333mhz divider so it could be a memory speed thing. Also when you change memory speed if all teh subtimings are on auto then these will change too so you go up these slacken again effecting the score. With more cards in the system memory bandwidth and speed becomes more important than ever when benching.
 
Last edited:
paradigm said:
Memory speeds are kept as close to 1600 as possible (at 125 strap they are at 1666).
Click to expand...

Have you checked the timings like 8 pack said? I noticed this problem during my OC, the XMPprofile would cause issues with overock performance somehow?? So I set it all manually!
 
The timings will have been adjusted then by the XMP or the AUTO settings of the board during memory training. Its what X79 does in line with the CPU IMC. Slackened subs can make a lot of difference to benchmark scores.
 
8 Pack said:
The timings will have been adjusted then by the XMP or the AUTO settings of the board during memory training. Its what X79 does in line with the CPU IMC. Slackened subs can make a lot of difference to benchmark scores.
Click to expand...

Thinking about it, I don't believe this to be the problem (obviously I can't prove this, as I'm at work), but as I said at the start, the CPU/Physics scores have gone UP as expected with clockspeed.

The problem is only related to PCIe throughput (maifested as FPS drop in the GPU only tests), which leads me to believe that it's more to do with the chipset/cpu voltage/some kind of throttling.
 
I still think it is mempry related mate. Test and see. Bandwidth is needed with crossfire config.
 
So is 5ghz easily obtained? I was waiting for ivy bridge i'd never thought of SB-E as I didnt realise you could overclock the fsb so to speak until ive looked today I just saw the 43x multiplier on the 3820 and thought that would mean 4.3ghz max.
 
Pretty straight forward, yes. 4.5-4.7 is pretty easy, infact for 4.6 I was fine leaving all the voltages on auto!

5GHz required a bit of tinkering with the CPU/RAM voltages, LLC, and SA voltages, but it's nothing a few hours reading won't arm you with the info for :)

Seems my problem was related to the VRM thermal threshold. I've upped it a touch and added some more direct airflow.
 
paradigm said:
Pretty straight forward, yes. 4.5-4.7 is pretty easy, infact for 4.6 I was fine leaving all the voltages on auto!

5GHz required a bit of tinkering with the CPU/RAM voltages, LLC, and SA voltages, but it's nothing a few hours reading won't arm you with the info for :)

Seems my problem was related to the VRM thermal threshold. I've upped it a touch and added some more direct airflow.
Click to expand...

thanks the only thing that was putting me of SB-E was that I thought it was like SB with a BLCK that you couldnt touch :D hence why I thought 4.3ghz was going to be the max.

So the question is what do I go for :)
 
You must log in or register to reply here.
Back
Top Bottom