Gpu2 - 9600gso

[ThunderRd]

Is it a G92 core or a G94?

It must be a G92... Newer GSO has a G94 core and 256/512 memory; and shaders are reduced to only 48. Sounds like a dog.

Nvidia must have been getting rid of some old stock G92s. Someone will eventually figure out how to hard mod it to use the shaders not in use. Do you have cojones? A $50 8800 equivalent sounds good.

[SteveI]

The one I just got has 96 shaders... and the one I want to get (384mb memory) also has 96 shaders.

Nvidia is playing games with old 8000 series GPUs and rebadging all over the place. I usually don't follow closely, but it seems like they are creating buying opportunities for cheap bastids like me.

The older 8 series is on the old fab process, and are probably less power efficient. It would explain the massive HSF on my card. Such a big assembly on a middle of the road budget card would be overkill. There are gems to be had, if you know what to look for. I saw 96 shaders and pulled the trigger with my fingers crossed.

[SteveI]

I think my GPU is G92.

I found a table of values for shader. It applies to 8800GS/9600GSO, which are supposed to be the same:

1512
1566
1622
1674
1728
1782
1836
1890
1944
2052
2106

Setting to something in between results in the shader running at the next lower value in the table. For a couple, I tested, the table seems right.... 65mhz apart seems like a lot, but represents 4% increments.

Edit: also looking at the above table and comparing my PPD gains, it is apparent that locking GPU/Memory, while increasing shaders only yields diminishing returns. Initially, 12% OC yields 7.5%, and at the end 3% only yields 1%. Once I find the max shader, I think I will investigate Core and memory OCs a bit. I suspect increasing all 3 should yield more linear improvements.

[Samuknow]

This is all good stuff. I am showing 4550 and 3987 respectfully on my gpu clients. A total of 8869 PPD with the CPU running also and my wife's rig.

[SteveI]

Sorry guys, I know this is getting childish...

Still stock core/mem @ 550/800

Shader is up to 1836 mhz.... Project 5013 showing 5120 PPD!!!!!!

It's still ticking...

1890 mhz.... shows 5184 PPD.

I am more than a little surprised it is still holding together. I think I am going to quietly walk away and let it fold overnight.

This $65 card (after rebate) is now officially outfolding the rest of my computers in my house combined, 4 (budget) SMP & one ATI 3870.

[dabaerman]

Originally Posted by SteveI

Sorry guys, I know this is getting childish...

Still stock core/mem @ 550/800

Shader is up to 1836 mhz.... Project 5013 showing 5120 PPD!!!!!!

It's still ticking...

1890 mhz.... shows 5184 PPD.

I am more than a little surprised it is still holding together. I think I am going to quietly walk away and let it fold overnight.

This $65 card (after rebate) is now officially outfolding the rest of my computers in my house combined, 4 (budget) SMP & one ATI 3870.

WTG SteveI, just like TR said, it's the shaders. trust him. while a mild oc on gpu/mem clock may yield some improvement, it's all in pushing the shaders. I have found the limit using the EVGA Performance Tools. the MSI 9600GT tops out @1800 on the shaders and gets me 4,000 to 4,400 ppd. I will try Riva Tuner when the newest 9600 arrives.

Ron

[ThunderRd]

The thing is, if we find the maximum on the shaders, and then we raise the core, will the shaders get a bit more headroom?

I have to take a look at that on mine. I didn't push the core because of the heat, but these 9600s run so cool [with a single slot solution, too] that I don't think that I have to worry

[SteveI]

I got the shaders up to 1890mhz (40% OC) and has run overnight... (5250 ppd) The next notch up is 1944mhz, seriously tempted to try.

For everyone who has overclocked their shaders, what should I be looking for? So far, it seems too good to be true. I have not tried benchmarks like 3dmark.... just GPU folding. Should I be checking anything else?

[PorPorMe]

I had the 9600Gt and the highest I got the shaders was 1745. Start running 3Dmark06 after each change. You might even want to go back a couple. When your mark starts going backwards, you've gone to far.

[SteveI]

@porporme,
Thanks, I am really taking shortcuts with my testing... F@H is stable so far, and I don't want to pinch myself.... Plus running 3dmark will slow me down, lol. I'm planning on continuing my effort with just F@H as my stability test..... What do you want from me, I'm greedy.

@TR,

Originally Posted by SteveI

....while increasing shaders only yields diminishing returns.....

I may need to correct myself. Given that FahMon time estimates may not be accurate until given time to compensate, some of my numbers may have been off a bit... Looking at my numbers in aggregate, it looks like increasing shaders DOES have a linear impact to folding performance. The direct translation appears to be 60-65%. In other words, increasing the shader speed by 10% yields 6% in PPD. Increasing shaders by 40% gets 24% more PPD. This still means there is about a 35-40% drag. Effectively, this could be caused by everything else being locked, much from core, memory, and pcix speeds. Honestly, TR's original assessment may be the best balance. To paraphrase, most of the gain is from shaders, and core and memory should help, but not worth the extra heat.

Final thoughts... the gravy train is over.... I finally got F@H to crap out at 1944mhz... I guess I will have to settle for 1890mhz. Very happy there, and will wait til the weekend to try other angles, like core/memory impact, as well as more serious stability testing.

[ThunderRd]

Originally Posted by SteveI

For everyone who has overclocked their shaders, what should I be looking for? So far, it seems too good to be true. I have not tried benchmarks like 3dmark.... just GPU folding. Should I be checking anything else?

I have one card at 1904, and one at 1912. They seem to like it there for folding. However, I can set the clocks much, much higher for 3dMark. My best run was 741/995/2034 for a score of 17851-with only one card in the machine. Screenshots here if you want to try my settings:
Squeeze your computer AOA EXIBITION

Since the machine folds 24/7 I just leave the settings at the best I can for folding. If I am ever inspired to benchmark again I know where I can start.

The only thing to remember is that from what I have seen, the shaders don't increase heat much at all, but the core and memory clocks shoot it up significantly and do little for folding speed.

EDIT: You posted about a minute before me and beat me to the punch...

[SteveI]

The only benchmark I am interested in ATM is PPD.

As long as it is stable, in games (as well), I have no desire to mess with a good thing. The card replaced an old ATI x700, so I don't need 3dmark to tell me how much it improved. Stock core and memory will do me fine. I'm happy, and the 384meg version is on sale again.... $50 AR from newegg......

[gedon]

benchmarked my "old" folding rig, too (2GHz A64X2, 8800GT)

1min 7sec per % p5013

more than 6000ppd
I want somebody to beat this

[PorPorMe]

Has any one taken a look at the 9800gt? I haven't had a chance to and thought i'd ask you guys.

[SteveI]

Originally Posted by gedon

benchmarked my "old" folding rig, too (2GHz A64X2, 8800GT)

1min 7sec per % p5013

more than 6000ppd
I want somebody to beat this

Sweet. How many shaders does your card have? what are your settings?

EDIT: Folding benchmarks don't count unless you are actively folding.

[dabaerman]

I have the MSI 9600GT OC edition. viewing GPU-Z, shows stock clock at 750 with current speed of 950. mem speed is 1900. the real supprise to me is the stock shader clock is 1680 with 2280 showing as the current shader clock. I haven't figured out how to get a screen shot of GPU-Z to share. also don't know the ppd out put at these settings however, 480 every 3 hours = 3840ppd. that works for now! another 9600GT on the way with a 9800GT to follow next week (the OT netted me $500). I am hooked on GPU Folding and I will pair one of the three on the sli board as soon as I can.

Ron

[dabaerman]

I think GPU-Z might not be reading my card correctly. I am using the EVGA tools to set clocks and don't have any of those settings applied (mem is stock @ 1900).

Ron

[SteveI]

Originally Posted by dabaerman

I think GPU-Z might not be reading my card correctly. I am using the EVGA tools to set clocks and don't have any of those settings applied (mem is stock @ 1900).

Ron

I never took it out of the box.. my card is evga as well.
Does it have fan control, and separate shader clock control? If it does, I may try it out.

[gedon]

Originally Posted by SteveI

Sweet. How many shaders does your card have? what are your settings?

it's a zotac amp edition, factory overclocked to 700/1700/1000, 112 Shaders
max oc: 756/1836/1095 [core/shader/mem]

Originally Posted by SteveI

EDIT: Folding benchmarks don't count unless you are actively folding.

this card folded from end of june until sep 30th. it did a great job, 500k in three month
It wasn't easy for me to turn it off, but we have to save some money

[dabaerman]

Originally Posted by SteveI

I never took it out of the box.. my card is evga as well.
Does it have fan control, and separate shader clock control? If it does, I may try it out.

the EVGA utility, I think it's called EVGA 'Advantage' Tools, has fan control and separate shader clock control as well. it was recommended to me becausr the MSI utility didn't support separate shader clock. my current 9600GT is an MSI OC version. I'm going to try RivaTuner for the next card.

Ron