Folding@Home on a 4870

[Assimilator1]

Originally posted by: Cheex
Thanks. I'll have to try doing that after I'm finished benchmarking for a new article (check my sig).

I have a 4870 btw....and the PPD (according to FahMon) varies between 2000-3000.
Is that good??

Opps yea I knew that originally, it's because that's what I was going to get .

I'm very inexperienced when it comes to F@H so I don't know how good a score it is, however I can tell you my Q6600 @3.35 GHz manages to get between 3400-4200 ppd depending upon the WU type.
I believe that F@H have had alot of help from Nvidia with this GPU client (what a turn around from the last GPU client!) so it is generally faster on Nvidia cards of similiar power.

Btw I bought a 4830 in the end (due to the price hikes), I'll get that crunching sometime soon, probably Saturday when I install it .Oh & I look forward to your cat 8.11 vs 8.12 review:thumbsup:


Thanks Mark for the help .

ICXRa
Got your 4850 crunching yet?

 

[theAnimal]

I'm getting 2000-2600 on a 3870.

Are you using the new core (1.22)?

 

[Assimilator1]

Cheex
F@H have just released a new core for ATI GPUs , check out this thread.

 

[Cheex]

Originally posted by: Assimilator1

Btw I bought a 4830 in the end (due to the price hikes), I'll get that crunching sometime soon, probably Saturday when I install it .Oh & I look forward to your cat 8.11 vs 8.12 review:thumbsup:

Well the review is up....check my sig for link.

 

[Assimilator1]

Well I got my HD 4830 crunching over the few WUs its done it scores 2300-2500 ppd with 1 SMP client on my quad that's a total score of 3900-4500 ppd. This draws about 270w from the wall.

2xSMP gets 3400-4200 ppd, this draws ~233w.

I've stopped running the GPU client as the extra ~40w isn't worth the small increase in ppd.

 

[Cheex]

You know what is interesting. I've never done any folding on my CPU, only crunching SETI.

Any idea how much PPD I'd get from my E7200 @ 3.8GHz???

 

[Assimilator1]

S@H or F@H?

Re F@H GPU crunching, on 2nd thoughts I need every last point to beat FMC!

 

[Cheex]

I was referring to F@H.

 

[Assimilator1]

Well I'm no expert, but my Q6600 @3.35 GHz gets mostly between 3500-4000 ppd with the SMP client.
If you 1/2 that then add 13.5% for your faster clock speed you get 2000-2300 ppd, however that doesn't factor in that my CPU has more L2 cache than yours (mine's 4MB/2cores vs your 3MB) & I don't know how much of an effect that has, if any on F@H.

Maybe someone else does?

 

[Drsignguy]

Cheex, My E7200 @ 3.6 is now getting 2100 - 2300PPD with Client 2653, which is 1760 points. The higher the point production results in a bit lower PPD production....

 

[Cheex]

Originally posted by: Drsignguy
Cheex, My E7200 @ 3.6 is now getting 2100 - 2300PPD with Client 2653, which is 1760 points. The higher the point production results in a bit lower PPD production....

I'm not sure if I understand that last part....:roll:

 

[Drsignguy]

Originally posted by: Cheex

Originally posted by: Drsignguy
Cheex, My E7200 @ 3.6 is now getting 2100 - 2300PPD with Client 2653, which is 1760 points. The higher the point production results in a bit lower PPD production....

I'm not sure if I understand that last part....:roll:

When the clients have a higher point value, the lower the PPD you will be getting.

Example: if the point value of the client is 1760, my PPD is around 2100 - 2300
if the point value of the client is 1920, my PPD is around 1700 - 1900

 

[Assimilator1]

Maybe the higher scored WUs take disproportionately longer??

Looks like I was fairly close with my guesstimate .

 

[Drsignguy]

Originally posted by: Assimilator1
Maybe the higher scored WUs take disproportionately longer??

Looks like I was fairly close with my guesstimate .

Pretty much on the mark!

 

[Assimilator1]

Cool , any idea how much difference L2 cache makes?

Btw I'm even more confused now about the points & ppd , when you say client you don;t mean WU type?

 

[Drsignguy]

Originally posted by: Assimilator1
Cool , any idea how much difference L2 cache makes?

Btw I'm even more confused now about the points & ppd , when you say client you don;t mean WU type?

L2 cache does make a differance. My Q6600's do a little better PPD than my E7200 just because of the cache. As for client, I am just going by my Famon info.

Sid could give you/us more details on the L2 cache information. And just to add, my Q9450 completes a WU faster than all my rigs no matter what client/Wu it has.

 

[Cheex]

I think I get it now. I noticed that.

Thanks.

 

[Drsignguy]

Originally posted by: Cheex
I think I get it now. I noticed that.

Thanks.

:thumbsup: :beer:

 

[Cheex]

I'm just not getting both GPUs to fold....so I just stopped trying....

If I run F@H on my CPU....What do I do about SETI?

 

[Assimilator1]

What's stopped you running GPU folding?

Re SETI, upto you, you could set SETI to use just 1 core & see if F@H will limit itself to the spare core, of course you couldn't use the SMP client then.

Just trying out some overclocking on my HD 4830, default GPU clock is 575 MHz, so far it's upto 650 MHz .

 

[Cheex]

Well, I have to sell one of the 4870 1GB cards anyway....there was a mix-up as I said and I ended up with 2....would love to keep both but i can't afford to do so.

Anyway, I removed the one I'm gonna sell and I'm running a single card again.
I've overclocked the card to 825 on the core and 4000 on the memory.


My PPD went through the roof.....according to FahMon.
I'm now getting between 3400 to 4600 PPD.


However there is a problem...

FahCore_11.exe has stopped working
FahCore_11.exe keeps crashing...and when it does....it downloads a new work unit and starts over.

So it is not completing the units and therefore I (and we) are not getting any points.



HELP ME PLEASE!!

 

[Insidious]

Sounds like you tried to set the clocks too high.

-Sid

 

[Cheex]

The core or the memory?

 

[Insidious]

Take them both back to stock

pick one of them and raise it a little. If it stays stable for 24 hours, raise it a little more.

continue until your temps get too high or you see EUEs during the 24 hours. (back off the setting if you see EUEs or high temperatures)

Once you find the sweet spot (below the speed that causes EUEs) leave it there and start inching the other clock up the same way.

It takes time (for testing) and patience (small steps) to achieve a msximized stable overclock

 

[Cheex]

Well that I know but its stable for gaming.
I thought it would be fine also for Folding.

I just thought that this had to do with the fact that I was messing with the whole CrossFire thing.