Undervolting affecting FPS....?

[Phaedrus567]

** Sorry if this is in the wrong forum, but it seemed like the bizzaro world alternative to overclocking and seemed appropriate. **

I'm currently trying to undervolt my CPU then North Bridge then maybe GPU then RAM?
This is my first time, so I'm obviously trying to learn the in's and out's.

My system Configuration:
Phenom 2 x4 945 with Stock Cooler
Gigabyte GA-MA790GPT-UD3H
2x 2gig G. Skill Ram
Sapphire Radeon 5850 1gig

Testing Tools:
Prime 95
Furmark 1.70
OCCT: GPU
HWMonitor


Rundown of situation:
Undervolting CPU seemed easy. I was getting down to 1.1125 volts at 3000 clock and prime 95 with small FPU went without much issue. Then tried the same with the Northbridge; Got to about 1.250 prime 95 blend tests (with cpu still at 1.1125). (This is only for P-State 0)

Tried OCCT:GPU (45 minute run, shader 8 complexity, error checking full screen) (Prime 95 Blend in background) and it would restart sometime in the test - maybe 5 minutes in.

Read that Furmark was preferable over OCCT, it seemed to run better, but then would restart or blue screen around 800 seconds. (Furmark settings: 8x MSAA, Post-Processing, Stability Test, 1920x1080 Full-screen)

Now retrying my CPU and NB voltage combos with Prime 95 blend and Furmark and while they run, I've noticed my average FPS drops with my CPU voltage.... that doesn't seem right considering undervolting is suppose to keep clock speed otherwise it would be underclocking.

Finally, I read that Furmark was crippled by catalyst software.


SO,
- What do you think is happening?
- What could I have done wrong?
- Is there a better method of testing?
- Is GPU undervolting possible (it's the one I haven't seen much when searching online and undervolting in general doesn't seem to get much current hype)


Thank you and sorry about the long post, I try to be concise... I just don't do it very well.

 

[daft]

gpu undervolting doesn't happen much because their usually fairly low voltage as is. though, you can try MSI afterburner

http://event.msi.com/vga/afterburner/

you should first try to raise your voltage up a tad with your cpu and NB just to be sure and run the tests again

 

[Phaedrus567]

The real goal is efficiency, so if the GPU is pretty well in tune that just one less thing for me to deal with right now.


I'll run a few tests of Prime95 and Furmark as listed above and post what furmark is reporting for FPS.
I usually run each test around 20 minutes, but in this case i may drop it down if the average comes out quickly.

 

[Phaedrus567]

Ok, so I pulled an April fools on myself I think

Results
CPU:1.1125 NB:1.2000
Min: 41 Avg: 50 Max: 62 Temp: 69°C

CPU:1.2000 NB:1.2000
Min: 39 Avg: 51 Max: 81 Temp: 73°C

CPU:1.3000 NB:1.2000 (Stock Defaults)
Min: 42 Avg: 55 Max: 69 Temp: 75°C



There is some movement there, not sure if it's enough to be worried about considering I only ran the test until the GPU temp stabilized (about 5 minutes or so).


Any other GPU test to match with Prime95 to see if Furmark is accurate?

What about running the Heaven Unigide Benchmark for stress testing (as this is mostly a gaming machine)?

 

[daft]

occt and furmark are the 2 for gpu stress testing for max heat. i would suggest going longer on p95. at least an hour. don't just go until temperatures stabilize, you only do that to see if you can go higher with the voltages.

best thing to do is to run p95 for at least an hour, then furmark for another hour. this is the best way to test for problems. your NB probably cant go down any further, but thats alright, it will maybe save a watt or 2.

 

[Phaedrus567]

Ok, working with Furmark 1.8.0 now and running Prime 95 Blend tests (same time).

First 1 hour test I've done had good results, but if I would like to clarify some points that other forums seem to be in disagreement about.


- daft, you said to test the two programs one at a time. Others I've read say do them together. I can see the rational on both sides but I wonder which is the more realistic one for gaming, encoding and general web surfing. I can't imagine my CPU going 100% on all 4 clocks for a long time, unless I ran Folding@Home maybe? On the other hand, it would be nice to know how that maxed out the system is stable.

- Should I see much difference in Tests completed between the "workers" in Prime95? I had 32, 41, 40, 40 on this last test. That spread surprised me.



The 2 watts would be nice, but the main reason to lower my NB was because I heard this motherboard has trouble with all ram slots being filled. I wanted to try and help it by keeping it cool with undervolting. Oh well, I'll see what I can do.

 

[daft]

well, you can run them both at the same time, though, it doesn't stress one core (the core being used by furmark) nearly as much, so stability cant be 99% certain like if it were with p95. though, it would maximize the temps in the case and thus test it for temperatures. either way is fine. i just prefer separate myself.

the workers in p95 are essentially different fft sizes to test. having them random helps with stability testing. so don't worry about that.

with your motherboard issues with the ram slots. it usually has something to do with the bios, and if it is with the NB, it would most likely be because its not getting enough voltage. though, i can say for certain

 

[Phaedrus567]

Just thought I'd finish out this thread.

Daft, thanks for your input. I'll keep in mind your suggestion about the ram and the NB voltage when/if I upgrade later.

I've only finished testing my stock P0 state for my Phenom 2 x4 945.
CPU voltage: 1.3000 down to 1.1125
Northbridge: 1.2000 down to 0.9875 (was a little surprised by this)

Performance doesn't show any difference in FPS and stock speed hasn't changed (3Ghz).

Prime 95 Blend Test and Furmark (Stability Test, Displacement Mapping, Post FX, Full screen 1920x1080 and MSAAx8) running at the same time
54 FPS average at both voltage settings
Wattage dropped from 346 to 297 average.


Using Heaven Unigine Benchmark (DX11, Shaders High, Tessellation Extreme, AA and AS at 4, Full screen at 1920x1080)
Wattage dropped from 206 to 177.5 average


Not too shabby, now, what else can I undervolt