A question for the GPU benchers

new boy

new boy

new boy

Soldato
Joined
25 Jun 2009
Posts
3,495
Location
Weston-super-Mare
So its a well known fact that pushing your GPU memory to the highest clocks it can manage does NOT necessarily give you the best benchmark scores. Something to do with error correction or whatever, it doesn't matter why, but its true.

Once you have finished with testing, and found what mem clocks give you your highest scores in a benchmark, does that mem clock give you your highest scores in all benchmarks? Or does the error correction stuffs work differently in different scenarios?

basically, if I take the time to tweak my GPU mem to give me the best scores in heaven, is that then always giving me the best FPS possible in games, or not?

I know it might seam like a waste of time to a lot of you, but there's defiantly a % or so up for grabs by doing a final tweak, and many of us have spent a lot of money for <10% gains (980/290x)

Thanks
 
Hey new boy.

In my tests, a memory overclock did nothing for my score in heaven or valley. I raised the core to 1515MHz and subsequently raised the memory with no difference to my score. It got to the point where my score got worse.

Here are the reults I came to doing my recent project log.

"Unigine Heaven 4.0
Unigine Valley 1.0
Luxmark Room
Tomb Raider
Hitman Absolution
Crysis 3
Watch_dogs
Alien Isolation
and 3D Mark Advanced

I will just post the overclock results, don't want this getting too heavy. These tests where applicable were performed at 1080p, and max settings were used where applicable except from any cinematic effects like motion blur etc...

Results:

Heaven 4.0:
CPU and GPU overclocked = Score 1443 @ 57.3fps
Improvement over stock = 149 points (11.5%)
Improvement over my previous system = 502 pts + 19.9fps (53%)

Valley 1.0:
CPU and GPU overclocked = Score 2441 @ 58.3
Improvement over stock = 239 points + 5.7fps(10.9%)
Improvement over my previous system = 411 points + 9.8fps (20.2%)

Luxmark:
CPU and GPU overclocked = Score 1657
Improvement over stock = 111 points (7.2%)
Improvement over my previous system = 116 points (7.5%)

Tomb Raider:
CPU and GPU overclocked = 51.8 average fps using built in benchmark
Improvement over stock = 5.6 fps (12.1%)
Improvement over my previous system = 11.1 fps (28%)

Hitman Absolution:
CPU and GPU overclocked = 48 average fps using built in benchmark
Improvement over stock = 2.1 fps (4.6%)
Improvement over my previous system = 5.2 fps (12%)

The Hitman Absolution results look really bad but by using stock clocks on both systems the 970 was around 28% higher scoring. This game does not perform well regardless of which hardware is being used.

Crysis 3:
CPU and GPU overclocked = 38.2 average fps using Fraps and my own consistent method
Improvement over stock = 3.7 fps (10.7%)
Improvement over my previous system = 9.2 fps (32%)

Watch_Dogs:
CPU and GPU overclocked = 51.4 average fps using Fraps and my own consistent method
Improvement over stock = 5.8 fps (12.7%)
Improvement over my previous system = 25.8 fps (100%)

Could not believe this so I tested it quite a few times. This result could be down to drivers but it's still pretty epic. I can actually play this game now!

Alien Isolation:
CPU and GPU overclocked = 214.9 average fps using Fraps and my own consistent method
Improvement over stock = 26.3 fps (5.4%)

I added this in for a laugh. Is this a bad console port or something? The fps is crazy for a game which looks so good!

Firestrike 1.1:
CPU and GPU overclocked = Score 10292
Improvement over stock = 1569 points (18%)

Sky Diver 1.0:
CPU and GPU overclocked = Score 24439
Improvement over stock = 3900 points (20%)

Cloud Gate 1.1:
CPU and GPU overclocked = Score 22283
Improvement over stock = 4805 points (27%)"

Bear in mind these results include CPU overclocks too. If you want to see the log, I'll post the link to it on the OCUK forum.

See my signature for the specs :)
 
Sorry, I only just saw this.

That's nice info, but it doesn't answer my question I'm afraid. I feel a little bad because its obviously taken you some time dig that out and reply, but I appreciate it none the less my friend :cool:

Normally, while chasing memory OCs, if you up them by 10Mhz a time, you will see something like this in say... heaven.

fps

39.9
41.1
41.3
41.4
41.6
41.3
41.5
CRASH

You you know the 41.6fps is the best you can get in that app. I justed wanted to know if whatever caused the dip after 41.6 fps, happens at the same place in every application, or if its application specific?

I'll test it myself I think son, but I just cant for the life of me nail down a 100% stable OC on my 290x that I'm happy with lol.

EDIT - In heaven 4.0. memory OCing did rather a lot for my score. Like, 1 fps for just 30mhz.... strange
 
This will not happen in every application as each program is different and may stress the card in different parts.

At the moment I don't feel I need a solid over clock for my 290x yet, it's not struggling, I don't think I'll have it long enough to have to find a stable OC tbh!
 
savagesam117 said:
This will not happen in every application as each program is different and may stress the card in different parts.

At the moment I don't feel I need a solid over clock for my 290x yet, it's not struggling, I don't think I'll have it long enough to have to find a stable OC tbh!
Click to expand...

Ah, that's... maybe good? Less tweaking to do. Its annoying, I cant QUITE get a nice balance of voltage and speed that can run crysis 3 AFK for 12 hours (can loop heaven endlessly though.

Currently trying at stock just to make sure its not my CPU or RAM that's crashing Crysis.
 
new boy said:
Ah, that's... maybe good? Less tweaking to do. Its annoying, I cant QUITE get a nice balance of voltage and speed that can run crysis 3 AFK for 12 hours (can loop heaven endlessly though.

Currently trying at stock just to make sure its not my CPU or RAM that's crashing Crysis.
Click to expand...

Well I can run the tomb raider bench at a much higher clock than I can ever run the heaven bench so it shows the different stresses. It will differ, you may get a spike in game that's hectic putting stresses further than what the bench is running for example. So it will crash. It's best to run a lower 24/7 OC so you don't run the risk of crashes in game, obviously running a benchmark is totally different :)
 
You must log in or register to reply here.
Back
Top Bottom