• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

nVidia shader OC

APOLLO

[H]ard|DCer of the Month - March 2009
Joined
Sep 17, 2000
Messages
9,089
I have two EVGA Superclocked 8800GT cards and have managed to OC their shaders up to 1890MHz. The next step for shaders is 1944MHz but I cannot hold this frequency stable for more than 3-4 steps of a WU. These cards are designed for high clocks having stock shaders at 1782MHz. Has anyone OC their 8800s over 1900MHz? If so, what did you do to keep them stable?
 
I've read around that they have sweet spots, and I think 1890 is one if those

I'm not sure how they find these sweetspots but I don't recal anyone hitting over 1900 stable
I couldn't get any of my 4 over 17 something so I guess your lucky or I just suck LOL
Posted via [H] Mobile Device
 
Try lowering your Core and memory clocks.
I got higher OCs on my 9800gt that way.

I thought the next step was 1936?

Also where did you find those stepping charts? I cant seem to find the ones i used to use.
 
I didnt think it was a chart so much as it was some formula
That said, please do post a chart if one was made
 
I had a chart, sorry lost it somewhere. :eek:
It showed 54 per step on the shader.
 
Code:
Nvidia GTX 2xx Series GPU Overclocking Step Chart
   by: Razor_FX_II
  Linked      Not Linked
Core  Shader    Memory
+9    +54       +9
576   1242      801
585   1242      810
594   1296      819
603   1296      828
612   1350      837
621   1350      846
630   1404      855
639   1404      864
648   1458      873
657   1458      882
666   1512      891
675   1512      900
684   1566      909
693   1566      918
702   1620      927
711   1620      936
720   1674      945
729   1674      954
738   1728      963
747   1728      972
756   1782      981
765   1782      990
774   1836      999
783   1836      1008
792   1890      1017
801   1890      1026
810   1944      1035
819   1944      1044
828   1998      1053
837   1998      1062
846   2052      1071
855   2052      1080
864   2106      1089
873   2106      1098
882   2160      1107
891   2160      1116
900   2214      1125
909   2214      1134
918   2268      1143
927   2268      1152
936   2322      1161
947   2322      1170
956   2376      1179
965   2376      1188
974   2430      1197
983   2430      1206
992   2484      1215

Code:
Nvidia 8800/9600/9800 Series
 GPU Overclocking Step Chart
   by: Razor_FX_II
 
  Linked      Not Linked
Core  Shader    Memory
+9    +54       +9
513   1188      792
522   1242      801
531   1242      810
540   1296      819
549   1296      828
558   1350      837
567   1350      846
576   1404      855
585   1404      864
594   1458      873
603   1458      882
612   1512      891
621   1512      900
630   1566      909
639   1566      918
648   1620      927
657   1620      936
666   1674      945
675   1674      954
684   1728      963
693   1728      972
702   1782      981
711   1782      990
720   1836      999
729   1836      1008
738   1890      1017
747   1890      1026
756   1944      1035
765   1944      1044
774   1998      1053
783   1998      1062
792   2052      1071
801   2052      1080
810   2106      1089
819   2106      1098
828   2160      1107
837   2160      1116
846   2214      1125
855   2214      1134
864   2268      1143
873   2268      1152
882   2322      1161
891   2322      1170
900   2376      1179
909   2376      1188
918   2430      1197
927   2430      1206
936   2484      1215
 
I stand corrected. Hopefully I can get my rig back up and try this on my flock of 88's lol

Thanks!

Can someone explain why it is these numbers are better for us to use and how much it helps by? I know its better, but I'm curious as to why/how much

Also, since folding only uses shaders, can we hit a number on this chart then drop core/mem speeds to help with heat issues?
 
Last edited:
I stand corrected. Hopefully I can get my rig back up and try this on my flock of 88's lol

Thanks!

Can someone explain why it is these numbers are better for us to use and how much it helps by? I know its better, but I'm curious as to why/how much

Also, since folding only uses shaders, can we hit a number on this chart then drop core/mem speeds to help with heat issues?

Thats what I do.

Increasing the core helps a little bit but were talking minimal impact.
I increase by those steppings and then try to drop the core/mem back to stock as long as its still stable.
This helps with the heat.
 
Can someone explain why it is these numbers are better for us to use and how much it helps by? I know its better, but I'm curious as to why/how much?
Every Video card has a Oscillator Crystal that sets the frequency of the entire video card. (usually little silver tower that can be found on the PCB and is 27mhz on the top of it)
This means that a video card only moves in 27mhz increments or a multiple/division of the 27mhz like for the core which is usually in steps of 9mhz and shader usually jumps by 54mhz.
So if you set the core,shader or memory to an in between mhz the board will either round up or down to the nearest step.
Most video card testing software that report clock speeds can not read these low level base speeds that the card is actually running at but the card is in fact running at multiple/division of 27mhz increments.

Most video cards are created at the factory with a stable ratio of core to shader speed like 2:1, the entire card is built with this in mind for optimal stability, if you deviate from the ratio then you may introduce instability into parts of the card.
(its like having 2 larger tires on the right side of your car and smaller ones on the left side, the car will still drive but semi out of control and the further you deviate the size of the tires the risk of crashing increases)

For optimal and stable video cards it is best to overclock the linked core/shader in steps that preserve the ratio that the card was designed to run at and the memory also moves in these steps as well but is not linked to the core/shader so may be independently overclocked from the core/shader ratio.

I hope that helps explain it and I didn't just confuse everyone.:)
 
Last edited:
So, once we hit those shaded clocks, should we drop the core or leave it in sync to keep stability?

And thanks for the info, great explanation
Posted via [H] Mobile Device
 
Personally I run my core/shader linked according to that chart and if I need to reduce heat then I raise fans and/or lower memory clocks.
If heat is still an issue after increase fan speed and lower memory then my last choice would be to lower core speed but this could lead to some instability of the shaders which then may have to be lowered.
 
I've read around that they have sweet spots, and I think 1890 is one if those

I'm not sure how they find these sweetspots but I don't recal anyone hitting over 1900 stable
There might have been one individual, but believe he accomplished it through overvolting the card.

I couldn't get any of my 4 over 17 something so I guess your lucky or I just suck LOL
No, not lucky. These cards are EVGA 8800GT Akimbo SC edition. The version with a small HS on the backside of the cards plus main dual-slot cooler. They have very high stock clocks and was wondering how high they can actually go. 1890MHz is only one step higher than my other dual-slot 8800GTs, so it's a marginal increase at best but it's very stable. At that shader clockspeed, they do 5000 PPD with the 1888pt WUs and around 6200 PPD with the 353 pointers.

Personally I run my core/shader linked according to that chart and if I need to reduce heat then I raise fans and/or lower memory clocks.
If heat is still an issue after increase fan speed and lower memory then my last choice would be to lower core speed but this could lead to some instability of the shaders which then may have to be lowered.
I don't believe heat to be an issue with these cards. The stock coolers are quite good and I have a 120mm fan blowing on them on the intake with an open case, plus a ghetto-rigged 80mm fan on the exhaust side exterior to the case within a PVC pipe blowing the air out.

Anyway, you're recommending that I link the shader clock to the core clock when I OC these cards? BTW, thanks for providing the charts for the various clock steps. They're the most comprehensive charts I have seen. :cool:
 
lol ghetto rigged
if they are still getting over 5k PPD I really need to get my 4 way back up then, thats a few points right there!
 
lol ghetto rigged
Yes, but it doesn't look that way. More of a purpose-built, industrial look. It's dual-socket server, so it fits with the overall 'heavy iron' theme, LOL. :p

if they are still getting over 5k PPD
Yes, I haven't seen a major reduction in PPD with the 8800s in the year they've been running except for the 511 WUs.

I really need to get my 4 way back up then, thats a few points right there!
No resolution on that problem yet? Did you post a thread on the motherboard forum?
 
I havent had a chance to try what they suggested here, I don't have a floppy in the computer that works or the one thats broke lol. I just got one today so now I have to try to make one of those discs and see how that works. I'm not expecting much tho.

Hopefully I can up my production by another 20k soon with a project Ive been working on, should come online in 2 weeks if I can get a good PSU that has enough 6 and 8 pin connectors for cheap. I've got about 45-50k in production just sitting on the shelves waiting for me to get time/money to build up the rigs. Also have to worry about my electric bill, thats one to keep an eye on, plus Ive got to juggle 2 jobs, one of which is the Army, and a woman, and I'm still getting used to my new leadership position so Ive got 8 more people to keep track of, and I moved a few weeks back and this house wont be done for another year or so and I still have shit in boxes and..... ug... life can realy suck sometimes...
 
I havent had a chance to try what they suggested here, I don't have a floppy in the computer that works or the one thats broke lol. I just got one today so now I have to try to make one of those discs and see how that works. I'm not expecting much tho.

Hopefully I can up my production by another 20k soon with a project Ive been working on, should come online in 2 weeks if I can get a good PSU that has enough 6 and 8 pin connectors for cheap. I've got about 45-50k in production just sitting on the shelves waiting for me to get time/money to build up the rigs. Also have to worry about my electric bill, thats one to keep an eye on, plus Ive got to juggle 2 jobs, one of which is the Army, and a woman, and I'm still getting used to my new leadership position so Ive got 8 more people to keep track of, and I moved a few weeks back and this house wont be done for another year or so and I still have shit in boxes and..... ug... life can realy suck sometimes...
I'm sorry to hear that.

Have you gotten around to trying the BIOS recovery yet?
 
Anyway, you're recommending that I link the shader clock to the core clock when I OC these cards?
Yes I have found for DC projects as well as some picky games it is best to keep the core and shader clock ratios or at least close for stable operation.

What ever core clock you are using then use the shader clock to the right listed in its row of your core clock in those charts for setting up a stable card, the memory part of the charts do not need to be linked in a row with the core and shaders.

Some OC version cards start out with there core a step or two higher than what they should be stock from the factories but this does not mean that they are any different in the charts, treat them all the same and use the chart.
 
I use unlink and only OC the shaders on my 285s for folding. For gaming, I run them stock.
 
Yes I have found for DC projects as well as some picky games it is best to keep the core and shader clock ratios or at least close for stable operation.

What ever core clock you are using then use the shader clock to the right listed in its row of your core clock in those charts for setting up a stable card, the memory part of the charts do not need to be linked in a row with the core and shaders.

Some OC version cards start out with there core a step or two higher than what they should be stock from the factories but this does not mean that they are any different in the charts, treat them all the same and use the chart.
I used the frequencies in the chart and linked the core+shader but had no better luck, unfortunately. The cards are probably at their limits without a voltage increase. However, only one client is throwing errors which might indicate one card is better than the other. I want both cards in that system with the same frequencies and will stick with 1890MHz. Funny thing is, the higher shader clock at 1944MHz wasn't producing any better PPD. Actually, it was a little lower by about 100 PPD. Strange, but I'm glad in a way.
 
Hmmm might have been the WU. All 3 of mine are the same vendor same model same clock, buy t every now and again 1 makes like 300 PPd less on all the same WU's
Posted via [H] Mobile Device
 
I havent had a chance to try what they suggested here, I don't have a floppy in the computer that works or the one thats broke lol. I just got one today so now I have to try to make one of those discs and see how that works. I'm not expecting much tho.
Is the board posting? All the boards I had the misfortune to experience not posting, there wasn't anything I could do without a BIOS chip replacement.

Hopefully I can up my production by another 20k soon with a project Ive been working on, should come online in 2 weeks if I can get a good PSU that has enough 6 and 8 pin connectors for cheap. I've got about 45-50k in production just sitting on the shelves waiting for me to get time/money to build up the rigs.
Man! We got to get your rigs up and running!! That's way too much latent production collecting dust and accruing obsolescence to be worth leaving around like that...! :eek:

Hmmm might have been the WU. All 3 of mine are the same vendor same model same clock, buy t every now and again 1 makes like 300 PPd less on all the same WU's
I interrupted the WUs in the middle and restarted them with the higher linked clocks. So, they are identical WUs. It's the same thing I did yesterday when I lowered the core+memory clocks and increased only the shaders. I also tried raising just the shader clocks and leaving everything else stock.

All three different procedures yielded the same or slightly less performance than the 1890MHz step. Something about 1944MHz that my cards don't like. Even if it was stable, I wouldn't run it that high because the performance seems to be the same or lower. :confused:
 
Is the board posting? All the boards I had the misfortune to experience not posting, there wasn't anything I could do without a BIOS chip replacement.

eh... He's got a help thread already posted, and another I kinda jacked for that same purpose. (Error #88/AGP Folding)

Some good info in there.
 
I know!!! Im sorry guys, I know im not being [h]ard right now, but once I get offf work (OT, woo woo!!) I'll try the bios recovery. If that shot tho, I'll need to trade my Q66 for a K9A2 and a cheap AMD so I can run my 4 88's, and then once I get a decent PSU I can get my other project up, both rigs should put out between 20-24 on the cards alone, but I'll probly end up with P4's or cheap AMD's so SMP wont help much

Ill get right soon, I promise, FOR THE [H]ORDE!!

EDIT - Ug, just got slammed with 3 slow WUs, all 3 260's are backed up now.. :(
 
I thought I'd post some odd observations with the shader clocks. I managed to get one EVGA card at a sustained 1944MHz shader without crashing the client. The strange thing is, it did not result in any better performance. Weird.

I mentioned that earlier in the thread but wasn't certain because having both cards running @1944MHz crashes the clients beyond several steps. Now, after observing the performance of the higher clockspeed for over half an hour, I'm certain what I was seeing was in fact true. Shaders @1890MHz is the highest these cards perform - anything higher doesn't improve performance. Why? :confused:
 
Is there some sort of error correction in the client?

Check with a decent artifact test tool to see if the data being spewed out is garbage.
 
Is there some sort of error correction in the client?
No idea.

Check with a decent artifact test tool to see if the data being spewed out is garbage.
Unfortunately, I can't do that right now. The computer is furthest from my displays and I can only access it though VNC. Any further testing will have to wait until I turn it off for maintenance.
 
Back
Top