ATI Core16

[SOA_RRU]

So 1st time folding with an ATI card, i know they take a core or a % of one but damn my SMP ppd goes

So SMP ppd by itself for my X6 1055T on a 10721 tops out @ 9.6k ppd

My 5870 running Core16 on a 11294 tops out @ 8.5k ppd

Now if i run both my SMP drops to like 3.3k

is that normal for SMP & ATI folding?

So either one or the other but not both? for max ppd

 

[dugdiamond]

the amd core will use all of the gpu and atleast 1 cpu core

on my 1055t it uses 2 cores and the cpu

just use SMP only, until the new amd cores arrive

 

[SOA_RRU]

I thought Core16 was supposed to be the core than sorted it all out? or was the new core only released to give ATI cards a PPD boost?

Suppose i could try some environment variables for the card to see if that releases any more PPD to SMP but i'm not counting on it

Ok just read http://www.hardwarecanucks.com/foru...ore-16-taking-amd-gpu-folding-next-level.html & like you said not worth doing both @ the mo

I wasn't aware of how early on in the development stage OpenCL GPU Folding still is

Now i just need to buy another cheapo rig to put this 5870 in then the spare 8800gt can go in here

 

[dugdiamond]

you are better off getting a cheap GTX460

than a 8800

 

[yuri04]

Yes, Core 16 is the one which will solve all problems, or at least that's what they hope.

The WU's that are now running on the core are far from optimized. Because the ATI cards have much more raw processing power the WU's need to be adapted to that. At the moment the shaders are starving for work. Which causes a lot of CPU being used up getting them work fast enough.

In the future the WU's will change and will use the shaders more effectivly. Core 16 is the basis for this work, so as time progresses it should become better.

The SMP Cores however can't stand anything taking their CPU time. So and will basically wait for the CPU time to be allotted to them before running on al CPU cores. This causes the rather large drop in PPD.

It can be solved though, to a certain extend, by just running the SMP core on n-1 cores (n being the total CPU cores). This will mean there is no slowdown due to the SMP core waiting because there is simpy one CPU core assigned to the GPU. If this will improve your overal PPD? you will have to try this.

Just a sidenote, when running more than 8 CPU Cores/Threads the SMP/Bigadv core will have to run on even numbers, so SMP 11 will run on 10 CPU Cores/Threads.

 

[SOA_RRU]

you are better off getting a cheap GTX460

than a 8800

Nah ive already got a gtx460 folding in my Q6600 rig, the 8800gt ive have spare is what came out of that rig.

 

[dugdiamond]

Just a sidenote, when running more than 8 CPU Cores/Threads the SMP/Bigadv core will have to run on even numbers, so SMP 11 will run on 10 CPU Cores/Threads.

it's a full amd rig

6cores max

never gonna get bigadv (unless in linux sandbox - but then the gpu won't work along side the WUs)

 

[dugdiamond]

...8800gt is what came out of that rig.

coolio

 

[yuri04]

it's a full amd rig

6cores max

never gonna get bigadv (unless in linux sandbox - but then the gpu won't work along side the WUs)

hence the sidenote

because this anwser is not only for this situation.