Question?... Don't Be Ridiculous!

[MMc2K3]

Is there a divider between graphics card memory and GPU as there is between system RAM and CPU?

Is there a ratio (1:1) between both, where the components will complement each other perfectly*, optimized if you like?

And how would you work out the ratio and set the speed of each component accordingly?

Or is it just a case of crank the shit out of it with ATI Tool until it fits*

*ATI ToolX1900XT

Although I do get a lot of ammusment from OverClocking the card (W1zzard ) it would be good to know.

 

[xman2007]

there is no such divider on graphics cards between the gpu and memory afaik the only slightly similar thing we see is with shader clocks being tied to memory speed on newer graphics card, so you overclock the memory you overclock the shaders.

please feel free to correct me if im wrong

 

[pbmaster]

I don't know if it's exactly a ration sort of thing, but some clocks work better together than others. I know my card performs better at 650 core/900 mem than at 650 core/950 mem.

 

[xman2007]

I don't know if it's exactly a ration sort of thing, but some clocks work better together than others. I know my card performs better at 650 core/900 mem than at 650 core/950 mem.

thats cause you took your memory too high obviousley it has nothing to do with dividers etc

 

[MMc2K3]

I don't know if it's exactly a ration sort of thing, but some clocks work better together than others. I know my card performs better at 650 core/900 mem than at 650 core/950 mem.

Interesting..

The reason I asked is similar. I noticed a significant hit in performance during testing as I went a little higher with GPU speed... went higher still on another run and performance hit gone.

 

[MMc2K3]

thats cause you took your memory too high obviousley it has nothing to do with dividers etc

You deserve that.

 

[pbmaster]

No I didn't take it too high, when I put it at 1000 MHz I get a great performance increase.

 

[MMc2K3]

No I didn't take it too high, when I put it at 1000 MHz I get a great performance increase.

I'm with you dude,

Unless Xman would care to explain?

 

[xman2007]

sure ill explain he didnt mention that he can get his memory to 1000 so going by his first post one would assume with the gpu clock being the same in both examples he's taken his memory too high, now if he would of added that he could also take his memory to 1000 with no problems then you could say well thats strange for some reason it performs worse @ 950 than 900. there still is no divider between the gpu and the memory on graphics cards afaik, they work in a completely different way to cpu/motherboards and more efficiently because of it. theres no bottleneck between the memory speed and gpu

I don't know if it's exactly a ration sort of thing, but some clocks work better together than others. I know my card performs better at 650 core/900 mem than at 650 core/950 mem.

btw this:

You deserve that.

quite immature as i was simply trying to answer a question you posted and it was your lack of understanding my post in the first place

 

[pbmaster]

Hey now just calm down...no one is going to get helped if we keep this up. I may not have included all the information to begin with but that's no reason to go about telling people what they do and don't understand

 

[MMc2K3]

sure ill explain he didnt mention that he can get his memory to 1000 so going by his first post one would assume with the gpu clock being the same in both examples he's taken his memory too high, now if he would of added that he could also take his memory to 1000 with no problems then you could say well thats strange for some reason it performs worse @ 950 than 900. there still is no divider between the gpu and the memory on graphics cards afaik, they work in a completely different way to cpu/motherboards and more efficiently because of it. theres no bottleneck between the memory speed and gpu



btw this:

quite immature as i was simply trying to answer a question you posted and it was your lack of understanding my post in the first place

I was simply objecting to your response to a legitimate post, quite confrontational I thought.


Now we've establisehed there's no divider, can you answer the real question here?

For best performance do you need to run at the max overclock on both core and memory or is it possible to achive better results by having the two working better together at a lower speed?

 

[MMc2K3]

Hey now just calm down...no one is going to get helped if we keep this up. I may not have included all the information to begin with but that's no reason to go about telling people what they do and don't understand

Makes thing interesting though...

I've pretty thick skin

 

[pbmaster]

It's never a good idea to run anything at it's maximum. Eventually there will be no performance increase no matter how much you increase the clocks and it's simply putting more stress on the part. There is something called a "delta" that has to do with an optimal amount you increase the clocks. Like 10 MHz versus 14 MHz...14 could be the better one to increase by. I think...

 

[MMc2K3]

Which is why when finding max using ATI tool it increases in 6Mhz increments?

Can this delta be changed on ATI Tool?

Sorry if I'm bugging, just in the process of OCing using ATI tool. My current method is affecting my 3DMark results and I can't get my head round it.

http://forums.techpowerup.com/showthread.php?p=578512#post578512

 

[Monkeywoman]

ATI shaders are locked with the core speed. i.e. 3870 has 775Mhz core and 320 shaders at 775Mhz . Nvidea has core and shader split. i.e. 8800GT has 600Mhz core and 112 shaders at 1500Mhz. this is the MAIN difference between ATI and Nvidea