• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Measured effects of core/shader/mem clocks

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
Lots of opinions get tossed around about the effects of core, shader and memory clocks on FAH production. I've spouted a few seat of the pants observations/opinions myself and finally decided to see if I was anywhere close to right. To test, I chose to use a 8800GTS (G92), a middle of the road GPU considering the current installed base. I chose to use a 511 point WU as they tend to use the most resources. I backed up the WU and used the same frames in each comparisson. Here's what I found:

Sorry to have to wrap this in code, but it's the only way I know of to keep things lined up.

Code:
Stock 675/1728/972  TPF: 1:55.4  Fan Auto Temp: 79

Everybody knows higher shaders = more ppd.

Two step Shader OC (108MHz)
675/1836/972		TPF: 1:50.3 Fan: 82% Temp: 62

A 6.25% shader increase results in a 4.6% ppd increase

Three step Shader OC (162MHz)
675/1890/972		TPF: 1:47.8  Fan: 82% Temp: 62

A 9.4% shader increase results in a 7.05% ppd increase

Shaders aren't quite linear

I've said core and memory clocks don't matter much.  So I declocked the core 49 MHz
to see what happens:
 626/1728/972		TPF: 1:56

Then I upped the core 27 MHz above stock to 702 MHz
702/1728/972		TPF 1:55

A 12% (76 MHz) change in the core clock makes a .9% change in ppd.
THe core clock doesn't matter much. 

I declocked memory to 900 MHz
675/1728/900		TPF: 1:57
A 9.2% decrease in the memory clock results in a 1.36% decrease in ppd, a bit more
effect than the core clock.

Maximum production is with everything clocked up:
702/1890/999		TPF: 1:47  Fan: 82% Temp: 63

Upping the core and memory cuts a little off the frame time.  I'd need to run more of
the WU to be precise about how much since it's less than 1 second per frame. I 
might be able to get a little higher core and memory, but I don't know if I could 
measure it's effect.  This is about 8% better production than stock.

In conclusion, it's not strictly true that core and memory clocks don't make any 
difference, but they certainly don't make a lot of difference.

One surprise to me was that the core and memory clocks made very little difference
in the temperature with the fan set at a 82% duty cycle.

I also checked on the effect of the fan duty cycle.
Clock			Duty Cycle		Temp
675/1890/972	       Auto (50%)               79
			60%			71
			70%			66
			82%			62
			91%			61
			100%			60

Not much reason to go above 80% or so.
 
Last edited:

ozzlo

Senior whose title will pier
Joined
Mar 17, 2002
Nice Work ChasR. This is very good information.
 

Shiggity

Member
Joined
Dec 16, 2007
Location
Chicago, IL
I wonder if the shader % changes from WU to WU. I would think it would be directly linear to the points of that WU, but that may not be true.

Still a good job, but I think you need a bigger data set :)
 
OP
ChasR

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
I have a bigger data set. I lack time to use it :) I'll do some quick and dirty calcs in a bit. It's damn near impossible to use the 1888 point WUs because the results aren't repeatable. You can run the same frames with the same config and get a different answer each time you do it. If you try to run comparisons on a rig you're actually using you'll affect the results. If I get a non 590x on one of my GTX260s I'll give see if I get similar results.
 

Shiggity

Member
Joined
Dec 16, 2007
Location
Chicago, IL
I have a bigger data set. I lack time to use it :) I'll do some quick and dirty calcs in a bit. It's damn near impossible to use the 1888 point WUs because the results aren't repeatable. You can run the same frames with the same config and get a different answer each time you do it. If you try to run comparisons on a rig you're actually using you'll affect the results. If I get a non 590x on one of my GTX260s I'll give see if I get similar results.

Ah yeah, didn't think of that. So many variables ;)
 
OP
ChasR

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
I checked results on a GTX260 (192) running secondary while leaving everything else running. I got pretty good results until a virus scan started.

On p5669, 353 points
A one step shader OC (4.2%) results in a 3.4% ppd increase
two steps (8.33%) results in a 5.9% ppd increase
three steps (12.5%) results in a 7.5 % ppd increase
four steps (16.67%) results in a 9.1% ppd increase

THe effect of shader OC on this card and WU is similar, but less linear than on the 8800GTS.

Decreasing the core clock 10% resulted in a 1% reduction in ppd. Decreasing the memory clock 10% made no difference whatsoever.
I haven't got results, due to the virus scan, on the effect of increasing the core clock to see if it has any affect on shader linearity.

The fan duty cycle testing proved that the sweetspot for this card is 70% where it ran 56 degrees. 100% fdc resulted in a temp of 53 degrees. In auto, the fdc was 40% and the temp 68 degrees.
 
OP
ChasR

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
Once I ran the shaders to max, 1512 MHz, a 10 % OC on the core, to 684 MHz, lead to a 3% improvement over the stock core (621 MHz) with max shaders. So with a 10% core OC and a 4 step shader OC (16.67%), I was able to boost production on the 353 point WU by a little over 12%.
 

Evilsizer

Senior Forum Spammer
Joined
Jun 6, 2002
yea TY chaser!! i think it is time to V-mod the 9600GSO's i got. was told they can do 1.8ghz shader but thats not enough for me. lol, though i may not have free time to do it. :(
 

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Thanks for those figures ChasR... looks like I probably have some optimizing to do. :)

Also, this is now linked from the 'Intro' Sticky. :)