Overclocking an 8600m-GT: results and analysis

Well, this is my first thread.

I was motivated to bring in this brief report because: (1) I am intrigued by how good can be the overclocking performance, and (2) there is an amazing lack of information regarding overclocking results VS stock clocks.

The programs I have used are:
A) For overclocking (OC) and stability:
-RivaTuner: for overclocking.
-AtiTool: for stability measurement.

B) For benchmarking:
-3DMar06: standard benchmark.
-Half Life 2: Lost Coast (HL2:LC): A semi-old engine but very popular. I used its in-built Video Stress Test.
-World in Conflict (WiC): A modern game taxiing both the CPU and GPU. I used its in-built Benchmark.


TESTING ENVIRONMENT:
My specs are on my signature. My 8600m-GT is DDR2, so the stock clocks are 475/400 core/memory. All measurements were done at a resolution of 1280x800. I am using the 174.12 driver. All measurements were done under the same normal life conditions (always the antivirus (AVG) on and a couple of webpages open).

METHODOLOGY:
I increased by 15 MHz the core clock, then run AtiTool for artifacts for 10 minutes, if no artifact then I increased 15MHz the memory clock and run the AtiTool for artifacts for 10 minutes. I kept repeating until I reached the levels I wanted to benchmark: 10%, 20%, and 30% above the stock clocks. Once on those levels I run the AtiTool for 1 hours looking for artifacts, after that I run the three benchmark programs. The games (HL2:LC and WiC) benchmarks were run at least three times to ensure consistency of the results.

RESULTS:
I was able to overclock 10% (525/440), 20% (571/481) but not 30%. The maximum stable clocks were ~28%/31% clocks (609/525), and also I tried a little lower clock: 609/505 (~28%/26%). AtiTool was stable for more than 1 hour in the four overclocked settings. The temperature increase was from 69°C (no OC) up to 73°C for the maximum OC.

The following table summarizes the results:
__Overclocking _________3DMark06_____HL2:LC________WiC
__(core/memory)____________________(avg FPS)__(min/avg/max FPS)
A) 475/400 (+0%/0%)_____3543_________65.79 _____12/22/39

B) 525/440 (+10%/10%)___3871_________72.8_______13/24/42

C) 571/481 (+20%/20%)___4177_________77.86______13/26/46

D) 609/505 (+28%/26%)___4354_________82.26______13/26/47

E) 609/525 (+28%/31%) ___4455_________84.56______13/26/49


ANALYSIS:
The 3DMark06 benchmark show increases of 9.25%, 17.89%, 22.89% and 25.74% VS the 10%, 20%, ~27% and ~29.5% OC.

HL2:LC frames per second (FPS) increases are more close to the GPU OC: 10.66%, 18.35%, 25.03% and 28.53%.

WiC is a different story. The 10% OC gives about 10% increase in the FPS. The 20% OC gives another 10% only to the average and maximum FPS. Further OC's (~ 27% and ~29%) gave only a meager 2% and 6% increase only in the maximum FPS. At this moment my speculation is that since this game is both CPU and GPU taxiing, then there is a limit the effective GPU OC, 20% in this case.


CONCLUSIONS:
First of all, regarding the 8600m-GT: I was impressed by its OC capability. I knew 10% could be achieved, 20% was reasonable, but be able to go past 25% is already a great achievement. I know there are reports of 8600m-GT OC'ed at almost 40%, but I am not sure how long they can remain stable.

Second, I learned that OC'ing results from different games (or engines, to be more general) cannot be used as proof of the general increased performance, not even 3DMark06. Look at HL2:LC and WiC, while the first can follow the OC (no heavy CPU load), WiC can only increase effectively up to 20% on its average value.

Finally, the big question: is overcloking really worthy? I think I have no clear answer. Always the dilemma goes in those lines: can the performance increase really reflect something in the games? (in other words: are the extra FPS really seen/felt by the user?)
- From my results it is very palatable to see a continuous increase in HL2:LC, but I doubt I will see an increase from 65 to 84 FPS.
- On the other hand, a 20% OC for WiC is much better. I know (using FRAPS) that the real average during real mission lies in the middle of the average and maximum values of the benchmark, while the real minimum is like 5-6 FPS less than the average from the benchmark. Therefore, the effective 20% increase will ensure I will not go below 20 FPS and thus will not experience any kind of lag in the game.



Please feel free to comment or ask questions on this brief report. If you have found something interesting about overclocking, please share it with the community, so we can learn together from our personal experiences/experiments.

 

Hi.

nice work, will come in handy overclocking my GDDR2

regards

John.

 

Good job. +rep. I'm not completely sure if such comparisons haven't already been done previously though.

 

I love this analysis that you did. I'll definitely be doing this for my new notebook that I should be getting (in the next 3 days).
http://www.newegg.com/Product/Product.aspx?Item=N82E16834115
With your analysis, I know exactly where to start overclocking from (safely). I think I'll make the new clock speeds permanent (by flashing the GPU's BIOS).


What I now want, is a tool to enable me to overclock my 1.66GHz Core2Duo processor.

 

Hi.

I thought we were only up to 174.31, but you say you are using 176.12

Is this a mistake, if not give us a link to it.

thanks

John.

 

Upppss, thanks for the info. I'll edit. Its the 174.12!!!

 

Thanks for the analysis. Problem is that the WiC benchmark isn't really 100% representative of in-game performance. I find the game in general has siginificantly better performance than what the benchmark shows.

 

Precisely, like I mentioned in the conclusions. Actually the benchmark used is with many special effects and zoom in into the action, therefore the benchmark is really for the most stressful parts of the game (like when having 5 or 6 air attacks and in zoom in).

 

I think this will be very useful for others. Good job. Can you put up your system specs and OS also? Thanks

 

I really hope these can help other people. My system specs and OS are on my signature.

 

bookmarked this under so i can look it up when i need it later, thanks for good guide.

 

What res was the 3dmark? Nice work by the way. +Rep.

 

Thanks Forerunner. My resolution on everything, including 3DMark06, was 1280x800.

 

You beat me to the punch, but I was doing something similar mainly for the c90 though. hehe. FOr me I found increasing the core and memory by 10 gave me around a 20 3dmark boost every time.

 

At which resolution you found this boost? 1280x800 or higher?

 

1280 x 1024. I got something special cooked up.

 

Very curious. You say that at 1280x1024 you get aproximately +1 mark in 3DMark06 for each MHz you increase the clocks. My results, at 1280x800 gives like 3.52 marks increase per MHz. I compared my 3DMark of normal cloks VS the max clocks I got stable.

The funny thing is that 1024/800=1.28, so there is more than just an increase in pixel load.

 

My 8600M GT is cranked up to 620/1240/550 and its stable. Max temperature is around 80 Celcius. Going to 650/1300/575 starts producing artifacts, so technically i could leave it at 560 mhz memory clock, but i left it at 550 mhz just to be safe. Just wondering if anyone managed to OC their DDR2 8600 GT at even higher clocks and remain stable?

 

I can't even push my cards memory past 450 mhz. Core is stable to 625.

 

Very nice work.


I'd like to add that I'm sure that benchmarks vary according to the design of the GPU cooling system and airflow of the notebook.

 

I'm stable until 630/490 and get 3950 in 1280 x 1024.

Benchmarks without overclocking are in link in signature.

Directions for overclocking in signature.

 

Great guide, wish I had read this a couple days ago. Has my vote for a sticky.

 

Thank you. I will see if in a week I can do some updates to the guide.

 

Thanks SonDa5. About the benchmarks you mention: are you saying that the benchmark at some specific clocks vary with the cooling system design?, or you mean that the maximum achievable clocks vary with the cooling system?

 

Figured I'd throw my 3dMark06 scores into the thread... (Notebook specs are in profile)

All tests are with driver version 174.16, and at 1280x1024 unless otherwise specified.

Stock: 3356
OC @ 575/460: 3916
OC @ 590/475: 4015
OC @ 600/500: 4128

Temperature never rose above 60°C on the GPU for any of these tests, no special cooling pads used. An LG display, and a GPU that overclocks very nicely and with reasonable temps. Personally I'm damn happy with those scores, my 3dMark06 score is a 23% boost over stock. I test at 1280x1024 since it's default, and since 90% of the benchmarks posted on the web are at that resolution so it makes it easier for me to compare.

Edit: Okay, got a few more gaming benchmarks to post. OC benchmarks are using my 600/500 speeds.

Counter-Strike: Source | Video Stress test @ 1680x1050 2xAA 4xAF

Stock trial 1: 52.20 @ 58 C
Stock trial 2: 58.15 @ 59 C

OC trial 1: 73.39 @ 59 C
OC trial 2: 73.89 @ 60 C

1680x1050 8xMSAA 16xAF

Stock: 41.31 @ 58 C
OC: 53.68 @ 62 C

A few important notes here on framerates. While on 8xAA/16xAF, the when OC'd it never dropped below 30 FPS that I could tell and it always looked smooth. Anybody who has run the CS:S video test knows the very beginning with the suspended tiles is the most graphically intensive portion--OC'd it was 29-35 FPS here (cl_showfps 2), stock it was 20-24. This was the only place in the entire stress test that the OC'd trial dipped under 50 fps. As any gamer knows, it's not about averages or peak framerates, it's about minimum framerates. The stock clocks reached unplayable choppiness during the most intensive stress test, the OC'd version was smooth the whole way. Overall I'm very happy with this card and it's overclocking, I hit 67 C during some CoD4 action but that's not that hot.