F @ H questions.

[XJustMeX21]

I haven't folded in over a year, but I would like to start again. I notice you can now fold with your video card. Which would provide the best results? Amd 965 @ 3.2ghz or an ati 4890?

Also when I used to fold to use a quad core properly you had to set up four different instances and specifically set up the affinity. Is this still the case? If so does anyone have a walk though for setting it up correctly?

Thanks.

 

[Cook1]

No, set up the SMP with your quad, get much more PPD vs setting up four instances. Also you can fold GPU2 on your 4890 along with SMP. I'm getting about 5k ppd on my 955 @ 3.8GHz (bonus points of course) and when I had a 4890 it was rather low...2kish. Hopefully GPU3 will bring some ATI card love.

If you need some guides on how to set these clients up let me know via PM and I can link you some guides.

 

[Assimilator1]

You've been able to fold with vid cards for years, since the X1900 days anyway

As said you'd be best to run the SMP client (the most recent one is best with 1 client apparently) & the GPU client.

Welcome back to folding, I've only just re-started myself after 6mths away.
Hmm I could fold on my HD 4830 again I suppose, I think I stopped because the CPU doesn't like the extra heat & I'd have to down clock it, which I'm loathed to do (even though it would only be down to ~3.2 GHz).
......Ah no, now I remember why I stopped, my room was getting too hot in the summer! (gets pretty hot in here anyway without a PC).

 

[imaheadcase]

No, set up the SMP with your quad, get much more PPD vs setting up four instances. Also you can fold GPU2 on your 4890 along with SMP. I'm getting about 5k ppd on my 955 @ 3.8GHz (bonus points of course) and when I had a 4890 it was rather low...2kish. Hopefully GPU3 will bring some ATI card love.

If you need some guides on how to set these clients up let me know via PM and I can link you some guides.

I've yet to get SMP and GPU folding to work together since they started having GPU clients.

No matter what I try, one or the other has to be running, or the other is slow its not worth running.

I always get in the top 20 list for anandtech folding team though so not a big deal, just very annoying.

 

[Assimilator1]

Last I did it the GPU client was only (mostly) taking 5-10% of 1 core IIRC, but I had to tweak the GPU client to not hit 1 core too hard.

What CPU have you got?

 

[Cook1]

Use flags when you're folding with the SMP to get your SMP and GPU client working together. For example I have a quad core so I use the flag -smp 3 to leave one core for my GPUs to use.

 

[Assimilator1]

Yea you can, but you don't have to leave a whole core to the GPU client, I found with my HD 4830 I could have it using 5-10% (out of 25%) of 1 core & their was minimal if any loss in GPU output. Does a 4890 need that much more CPU power?

 

[Cook1]

Depends on what GPU's you're housing. And if you're having issues getting SMP and GPU2 client working together, this is a sure way to do it.

 

[Mr. Pedantic]

Yea you can but don't have to leave a whole core to the GPU client, I found with my HD 4830 I could have it using 5-10% (out of 25%) of 1 core & their was minimal if any loss in GPU output. Does a 4890 need that much more CPU power?

I've found with AMD GPUs and the A3 WUs even 5-10% of 1 core is too much. Since the program runs on threads, each thread has to be synchronized and that means that the entire WU runs at the speed of the slowest thread, which is the one on the core running GPU too. This drastically reduces A3 ppd because of the bonus, and for AMD GPUs the increase in ppd barely offsets that, for another 200W of power and lots of noise from the fans. So at the moment I'm just running -smp 8 on my Core i7, folding on my 4870 as well is just not desirable nor feasible.

 

[Assimilator1]

(Corrected typos in previous post )

Going back through an old thread http://forums.anandtech.com/showthread.php?t=277895 ,I found that I was actually running F@H GPU alongside DPAD, hence my different results.
Running with DPAD was a win-win scenario, I could give just a small amount of 1 core's CPU power to the GPU client, & DPAD had only a minimal loss of output.

Btw when folding on my 4830 it would increase system power useage by about 50w, now I know your more powerful 4870 will draw more power, but not 4x as much surely?! .Unless you're running a 4870 X2 or 2 4870s? I guesstimate that a 4870 could use upto ~50% more power than a 4830.

PS lol at your sig.
<--- crosses fingers that their are no errors!

 

[SemperFi]

I have an ATI HD4670 video card that I am currently running the GPU client. This same box is running smp client on a quad core amd. If I am reading this right, the gpu is slowing down the smp client and I would be better off turning off the gpu client to maximize my smp bonus points. Is this correct?

 

[Mr. Pedantic]

Btw when folding on my 4830 it would increase system power useage by about 50w, now I know your more powerful 4870 will draw more power, but not 4x as much surely?! .Unless you're running a 4870 X2 or 2 4870s? I guesstimate that a 4870 could use upto ~50&#37; more power than a 4830.

TDP of a 4870 is 188W. Folding is nearly as bad as Furmark in terms of power draw. So I'd expect the 4870 to pull about 200W. However, I may have exaggerated a little, not taking idle power into account. So probably about 150W. It's still a heck of a lot of noise and heat.

 

[Assimilator1]

Yea the GPU client certainly generates more heat & noise.

Re power draw differences between the 4830 & 4870 I've been looking over a few articles. Hexus shows the 4850 draws about 27w more than the 4830 under load http://www.hexus.net/content/item.php?item=15990&page=9, & another review shows the 4870 draws about 26w more than the 4850 (neither are 1GB versions) http://www.hexus.net/content/item.php?item=13972&page=21.

So going by Hexus the 4870 512MB would draw about 53w more than a 4830 under load.
Although it may well be a little bit wider for F@H.

AnandTech's review http://www.anandtech.com/show/2647/10 ,shows a load power difference of ~76w 4830 vs 4870.
So based on my 4830 readings & adding the highest difference gives a figure of ~126w extra power draw running F@H GPU on a 4870 vs 4830 (plus a little bit for F@H). So both our 1st guesses were out by ~75w lol :biggrin: , your 2nd one was much closer though .
Although at the end of the day you'll only really know by having a plug in the wall meter or a clamp meter.

SemperFi
If the GPU client is taking up a whole core then probably yes, give it a shot & see what ppd HFM shows, let us know before & after scores .

 

[SemperFi]

I had all 4 cores assigned to the smp so the video card had to interfere with smp processing.

My gpu just shut down which it has been doing on a regular basis over the past couple of weeks. So I will just leave it off for now.

 

[Mr. Pedantic]

If your AMD GPU client is taking up a whole core, try this:

http://www.overclock.net/overclock-...-environment-variables-lower.html#post7362078

It should lower the CPU usage dramatically. Whether or not that makes a positive difference on the SMP2 client is questionable because of the threaded nature of the SMP2 client, but if nothing else it improves system responsiveness.

 

[Pokey]

I'm not well versed in this but back when I first started using the GPU clients, the recommendation was to create an environment variable:

Variable: NV_FAH_CPU_AFFINITY Value: 0 (of course this was for Nvidia card)

Then install a little program called FAH SMP Affinity Changer Link This automatically distributes the load between CPU cores.

There is also a setting in the GPU configuration dialogue that needs to be checked do NOT lock cores to a specific cpu.

Again, I use Nvidia cards so this may or may not be helpfull.

 

[Mr. Pedantic]

Then install a little program called FAH SMP Affinity Changer Link This automatically distributes the load between CPU cores.

There is also a setting in the GPU configuration dialogue that needs to be checked do NOT lock cores to a specific cpu.

Again, I use Nvidia cards so this may or may not be helpfull.

No, for AMD cards they don't work very well. I already tried.

 

[SemperFi]

Well I am just finishing up the week with the gpu shut down. Definitely a good move, I have 7000 more points this week so far and folding monitor shows I will be sending about 2700 more points up before midnight.

Thanks for the info. Can't believe it made that much a difference.

 

[Assimilator1]

Nice improvement , but pitty you can't use your grx card. Well at least their'll be less heat output!