8800 GT: OC experience?

[zombie67]

These are the stock settings for my 8800 GTs:

Core clock: 600
Memory clock: 900
Shader clock: 1500

What are reasonable OC values to use. I am looking for increased performance, with stability. Does anyone have any experience over clocking a 8800 GT?

[Mad Matt]

This is how I clocked my 9800GT which is AFAIK a rebranded 8800 (G92 chip): GPU 624 MHz RAM 900 MHz Shader 1809 MHz. http://www.techpowerup.com/gpuz/kh4dp/

Think I had some stability problems so I went back to stock GPU 600 and Shader 1792 or so.

Hope this helps. :-h

[Fire$torm]

I was running two of them for about two years. Got them used from Keith [MM]. Using Afterburner I had them OC @705Mhz Core. Take a look at my GPU-Z report (here (http://www.techpowerup.com/gpuz/kphky/)).

Edit: Note that the 8800s had Zalman Vf1000 (http://www.newegg.com/Product/Product.aspx?Item=N82E16835118037) coolers installed.

[zombie67]

Thanks for the info. Does there need to be a constant relationship between the three clocks? Or they can me moved independently? Does POEM care about memory clock speed?

[Fire$torm]

Thanks for the info. Does there need to be a constant relationship between the three clocks? Or they can me moved independently? Does POEM care about memory clock speed?

For the clocks I always kept the Shader linked to the GPU Core (Default in AB) as I didn't want to muck things up. RAM clock can go lower as long as it does not effect the project. I never crunched POEM with my 88s so I cannot answer that question. Sorry.

F$

[zombie67]

I decided to OC all three cards by 2%, to start with. I increased the core clock to 612, and the shader clock to 1530. Interestingly, even though the shader setting says 1530, GPU-Z is saying that it is running at 1512. Same with all three. And I left the memory clock at stock speed. I did this all with nvidia system tools.

I'll let them run like this for several days, and see what impact it has on the credits/day. As you can see, they have been very constant at stock speeds.

http://boincstats.com/en/charts/58/user/gainPerDay/201/chart.png

[Mad Matt]

Thanks for the info. Does there need to be a constant relationship between the three clocks? Or they can me moved independently? Does POEM care about memory clock speed?

I think fiddling with CUDA memory is complicated. There was little effect on some (I think DirT) and notable on others (IIRC GPUGRID). Never considered it worth getting into detail.

Definitely increasing GPU clock has little effect on total output, the shaders matter. And in the end you can trade off GPU clock for shader clock. Not sure why shader seems to prefer 54MHz steps. You can use different steps, but for some reason they seem working quite reliable.

1242 Mhz + 54 MHz = 1296
1350
1404
1458
1512
1566
1620
1674
1728
1782
1836

[zombie67]

I decided to OC all three cards by 2%, to start with. I increased the core clock to 612, and the shader clock to 1530. Interestingly, even though the shader setting says 1530, GPU-Z is saying that it is running at 1512. Same with all three. And I left the memory clock at stock speed. I did this all with nvidia system tools.

I'll let them run like this for several days, and see what impact it has on the credits/day. As you can see, they have been very constant at stock speeds.

Well, adding the 580 completely messed up that chart. In any case, I looked at the individual charts for the three machines with 8800 GTs. Two of the charts are below. The third is meaningless because I stupidly merged hosts during the trial run. No change that is noticeable. Sure, a 2% increase is not much. But I think a more significant influence on the daily credits, is just when a particular task finished WRT the dateline. I will try another 2%. If it still works, and if it still makes no difference, I will go back to stock. Might as well save electricity. That is real $.

http://boincstats.com/en/charts/58/host/gainPerDay/128863/chart.png

http://boincstats.com/en/charts/58/host/gainPerDay/125616/chart.png

[trigggl]

Might as well save electricity. That is real $.

That's what I'm thinking. There's the law of diminishing returns. No need to create more heat and burn more energy for minimal gain. I've down-clocked my 9800 GT and it does just fine on POEM. That card gets too hot for the level of performance you get out of it and they don't over-clock well anyway.

[Mad Matt]

None of my cards of this generation had voltage controls. If you have, that would be worth considering and go for some hefty undervolting. If these cards should use more than 50% of the TDP, I'd reserve OC for the big guns.

[Fire$torm]

That's what I'm thinking. There's the law of diminishing returns. No need to create more heat and burn more energy for minimal gain. I've down-clocked my 9800 GT and it does just fine on POEM. That card gets too hot for the level of performance you get out of it and they don't over-clock well anyway.

Is that was a stock cooler? My 88s with the Zalman VF1000 @705 OC always ran cooler then my stock 4850s did near stock clocks.

[trigggl]

Is that was a stock cooler? My 88s with the Zalman VF1000 @705 OC always ran cooler then my stock 4850s did near stock clocks.

Affirmative, that ridiculous "cooler" is still on it.

[John P. Myers]

Definitely increasing GPU clock has little effect on total output, the shaders matter.


Very true. I used to drop the core below stock speed to save on heat, leave memory at stock and only OC the shader. Core does 100% nothing on the older Nvidia cards.

[Mad Matt]

Very true. I used to drop the core below stock speed to save on heat, leave memory at stock and only OC the shader. Core does 100% nothing on the older Nvidia cards.

There's a nice app for instant feedback on changes, unfortunately it does not work with cards after GTX 295. http://cuda-z.sourceforge.net/

[DrPop]

So the only reason we increase the core clock now is because it's hard linked to the shaders? On the old cards they were not linked, right? I actually have an old 8800 GT from one of our buddies in the old Dell rig, and I should try this on it! ;)
What's a good "safe" max temp for these older GPUs to handle?

Sent from my MB860 using Tapatalk 2

[Mad Matt]

So the only reason we increase the core clock now is because it's hard linked to the shaders? On the old cards they were not linked, right? I actually have an old 8800 GT from one of our buddies in the old Dell rig, and I should try this on it! ;)
What's a good "safe" max temp for these older GPUs to handle?

Exactly. This was my first big blow when holding the 570 in my hands and I realized shader and GPU now are linked, just as on ATI... x-( For crunchers it just was perfect before. 'Just' been missing a project with return. :rolleyes:

I really forgot the temps I had, sorry. But they should not get really hot with POEM. If they do nevertheless, I would not OC them as this would indicate they are using more juice compared to big new cards.