GTX970 SLI with overclocked hex-core... GPU-benchmarks lower than expected

crinkleshoes · 4 Jan 2015 at 19:40

Hey,

I am sure I am missing something obvious, however my benchmark results are a good 10% lower than I would expect them to be both when stock and overclocked and I am wondering where the bottleneck may be.

- All NVCP settings setup to performance - tried with altering the windows settings to performance too (although that last one was done just for the annoying win 7 performance warning message)
- G-Sync/v-sync disabled for benchmarks
- Tried with various combinations of settings/overclocks... same pattern repeats
- Memory benchmarks seem on-par with expectation (actually exceeding in latency)
- CPU benchmarks seem on-par with expectation
- Appears to only be GPU limitation
- SLI-bridge is connected (I also tried connecting it the other way around just to be sure - no "please connect bridge" message appearing from CP).
- Correct PCI-E ports used according to motherboard manual.

System config:
Asus X99-S
5820k @ 4.5 and 4.625, cache @ 3.5
16GB quad channel ddr4 @ 2666 at timings ranging from 15-15-15-38-2T to 12-13-13-26-1T
2x GTX970 running at 1450/8000 for most of these comparisons... although I have also run a few benchmarks at 1500/8000

Corsair H110 keeping temps well under thermal throttling speeds (sub-80C in all these benchies).

Any recommendations would be greatly appreciated. The system is running everything well at 2560x1440@144Hz - all CPU/memory bound things are exceptional.

But these benchmark scores are showing some kind of GPU bottleneck & I would like to solve it... not because I "neeeeeeeeed" desperately higher scores... it just seems as though something may be missing. You can see clearly it's the GPU score holding back the firestrike total score, physics score looks OK to me.

04-01-2015 18-07-19 by CosmicLogos, on Flickr

04-01-2015 18-07-24 by CosmicLogos, on Flickr

04-01-2015 20-01-35 by CosmicLogos, on Flickr

04-01-2015 20-01-50 by CosmicLogos, on Flickr

04-01-2015 20-18-05 by CosmicLogos, on Flickr

04-01-2015 01-06-36 by CosmicLogos, on Flickr

Thanks

Freddie1980 · 4 Jan 2015 at 19:45

When you get run the benchmark are you running Pricission/Afterburner to check what clockrates your actually getting? It could be one of the cards (probably the top one) is clocking lower due to higher temps.

Dirk Diggler · 4 Jan 2015 at 19:46

I had a similar query yesterday. Cut a long story short, even with G-Sync disabled the monitor was capping the max fps.

I run the benchmark on my second non G-Sync monitor and the score jumped from 2500 to 2800 in Heaven.

crinkleshoes · 4 Jan 2015 at 19:48

Yes... the clock speeds I have quoted are from afterburner. I have been checking the graph following benchmark completion. Both are running at a steady 1450 & keeping quite cool... the fans aren't even spinning up to max speed in the process.

Faster boost clocks do improve the scores slightly... only I have settled on these clocks as a happy 24/7.

crinkleshoes · 4 Jan 2015 at 19:48

String said:
I had a similar query yesterday. Cut a long story short, even with G-Sync disabled the monitor was capping the max fps.

I run the benchmark on my second non G-Sync monitor and the score jumped from 2500 to 2800 in Heaven.

Interesting to hear, thank you... I will give that a go.

setter · 4 Jan 2015 at 19:50

Do indeed seem a bit low, particularly standard fs. Closest one i could find to yours was this one.

20703. GFX Score 26983. 970. Air cooled. Physics Score 18708. CPU 5820K @ 4.7 Water cooled - SS-89 - Link

No idea on gpu clocks there, but cpu is a little higher.

crinkleshoes · 4 Jan 2015 at 20:09

String said:
I had a similar query yesterday. Cut a long story short, even with G-Sync disabled the monitor was capping the max fps.

I run the benchmark on my second non G-Sync monitor and the score jumped from 2500 to 2800 in Heaven.

Worth a try, but no real change

Plus note... peak FPS in heaven screenshot above is over 144fps (over 200 in fact).

setter said:
Do indeed seem a bit low, particularly standard fs. Closest one i could find to yours was this one.

20703. GFX Score 26983. 970. Air cooled. Physics Score 18708. CPU 5820K @ 4.7 Water cooled - SS-89 - Link

No idea on gpu clocks there, but cpu is a little higher.

P4Clock · 4 Jan 2015 at 20:15

set fans to 100% in AB and re-run

crinkleshoes · 4 Jan 2015 at 20:31

Makes absolutely NO sense... but the score has dropped in doing that?!

andybird123 · 4 Jan 2015 at 21:23

Sound like a power limit issue, what is yours set to?
The fans drawing more power takes a tiny bit from what the card has available

What 970's are they exactly?

Dirk Diggler · 4 Jan 2015 at 21:27

Is your core clock solid all the way through the test? Or, is it dropping 13hz at a time?

Gripen90 · 4 Jan 2015 at 21:45

Close all background programs you may have running in the system process bar. Close things like anti virus, steam, origin etc. Have a lot of these processes running in the background takes CPU time and reduce benchmark scores.

Also I don't think you score low at all - you just need some more GPU OC.
This is what I get with Core i7 4930K@4GHz, MSI GTX 970 Gaming SLI @1300/1903MHz, Forceware driver 347.09. The cards boost to a constant of 1502MHz.

FireStrike Extreme
SCORE 9779 with NVIDIA GeForce GTX 970(2x) and Intel Core i7-4930K
Graphics Score 11388
Physics Score 14744
Combined Score 3814
http://www.3dmark.com/fs/3633649

FireStrike
SCORE 17992 with NVIDIA GeForce GTX 970(2x) and Intel Core i7-4930K
Graphics Score 24628
Physics Score 14300
Combined Score 6833
http://www.3dmark.com/fs/3666773

Unigine Heaven 4.0

What really makes a difference is overclocking the GPU - it means much more than OC'ing the Ram. I Oc'ed my ram first from 1753 to 1815 it gave 2 FPS more, then I OC'ed from 1815 to 1903Mhz that only gave 0.7FPS.

SHOGUNXSHINOBI · 4 Jan 2015 at 22:29

My results are as follows- FIRE STRIKE EXTREME
SCORE
9666 with NVIDIA GeForce GTX 970(2x) oc to 1567mhz, if remember correctly, and my Intel Core i5-4670K oc to around 4.6ghz
Graphics Score 25411
Physics Score 9080
Combined Score 7817.
Lol my physics score gets a good hiding compared to yours, nearly double figures.
And your gpu score does not look to off considering your oc.

Gripen90 · 4 Jan 2015 at 22:32

SHOGUNXSHINOBI said:
My results are as follows- FIRE STRIKE EXTREME
SCORE
9666 with NVIDIA GeForce GTX 970(2x) oc to 1567mhz, if remember correctly, and my Intel Core i5-4670K oc to around 4.6ghz
Graphics Score 25411
Physics Score 9080
Combined Score 7817.
Lol my physics score gets a good hiding compared to yours, nearly double figures.
And your gpu score does not look to off considering your oc.

Your GPU, Physics and combined score aren't from running FireStrike Extreme but regular FireStrike

JustCallMeVlad · 4 Jan 2015 at 22:35

Damn those scores are very low compared to mine

Are you testing with 100% fan?

Gripen90 · 4 Jan 2015 at 22:41

GUYS ! His GPU core clock speed in "only" 1224MHz ! that is what makes the big difference.

Dirk Diggler · 4 Jan 2015 at 22:47

Gripen90 said:
GUYS ! His GPU core clock speed in "only" 1224MHz ! that is what makes the big difference.

Not according to his OP. He's saying 1450 and sometimes 1500, with 8000 mem?

JustCallMeVlad · 4 Jan 2015 at 22:48

Gripen90 said:
GUYS ! His GPU core clock speed in "only" 1224MHz ! that is what makes the big difference.

Its not reading his boost clock correctly. Mine states 1478 but was really at 1556x

Nickolp1974 · 4 Jan 2015 at 22:54

@ OP are you changing your resolution to 1080, if your leaving at 1440 your gpu is working harder to scale down and your score will take a hit.

varkanoid · 4 Jan 2015 at 23:14

I know people dont like using it but I found out in Furmark that it actually tells you on the benchmark screen if the overclock is limited by power / volts / temps / ov I found it quite usefull.