8800GTS: What to overclock? Shaders/Memory/Core?

corban

n00b
Joined
May 13, 2006
Messages
22
I have a new 8800 GTS 512MB, and have looked at a lot of posts regarding scores of Core/Memory/Shader over-clock values. In Riva, the Shaders and Core values are linked together and the memory is separate. So I am tossing this out to see what values should I look at when trying to push the over-clocking more on this card?

I have ATITool, and have used that to see how stable things are, but I was wondering how I should proceed to get things over clocked?

What is better to over-clock? Memory, Shaders, Core or some combination. Is it better to have a higher core value, higher Shader values, or higher memory values? Where do you push this card for better results? Should you keep the Core and Shaders linked together?

I got it up to 750 Core, 1970 Shader, and 1024 Memeory just with the default cooler and the fan running in auto mode. (Via Rive Tuner.) I get about 45C idle and about 65C at full load. I looks fairly stable (not fully tested) but after a while ATITool starts to find errors and starts the timer over. (After abut 12-15 mins)

Just curious if I should push more Memory or push the Core faster.

My system is and Opty 165 running at 2.8 and cooling does not look to be bad. I get 10446 in 3Dmark06 with the above numbers.

thoughts? Just trying to get conversation going about how to play with these different values on these new cards. So far so good.

thanks for the help!

cheers

corban
 
The card is a little bit memory bandwidth hungry (actually, maybe a lot) so your best improvement is probably memory speed. Once you get that pushed up, you can push the core and shader. The real reason to un-link the two is that you can often push the shader higher than it would go linked - if the core tops out kind of low an unlinked shader could still go higher and give you some benefit. As to which is more important between core and shader, I don't know - I'd guess shader in new shader heavy games like Crysis, maybe core in older stuff. Maybe someone has done some testing to see. Those clocks sound pretty decent - something around 800/2000/2100 is about the highest I've seen reported regularly.

For what it's worth AtiTool gives me artifacts at anything over 1850 shader - I haven't yet decided how much of a factor that is in games.
 
Hey Forceman...

Thanks for your reply. I will try try playing with the speed of the memory. I will start the others low and crank that up and see where I can get.

Any other suggestions welcomed!

thanks again.

cheers

corban
 
Generally you want to shoot for the highest possible stable clocks for your card for core, shader, and memory if you want the most bang for your buck. As forceman said, the card is memory bandwidth starved, so that's what you want to maximize most, but any clock increase will show improvement. The RAM is decent though, and should hit ~1100MHz+. It usually craps out at 1125MHz though. Also, unlink the core and shaders so you can maximize both. Try to max each singularly (although you might have to eventually increase your core to raise your shader clock more and vice versa, they can only be unlinked so far).

It's also helpful to use Rivatuner hardware monitoring to monitor your clocks in real-time. This has a major benefit because your clock speeds increase in steps. For example, shader clocks increases in 54MHz intervals only and nothing else. For example, it goes from 1944 to 1998 to 2052, nothing in between. No matter what MHz you set it at in that range, it'll only be one of those numbers. When you get close enough to the next, it'll switch, and you can see this in Rivatuner. Same goes for core and memory. For core, it'll go 783, 792, 800MHz. Check it out, and you'll see what I mean :D.
 
Back
Top