Introducing... jibberegg

[jibberegg]

Hi there all,

Apologies for not doing this sooner, but I only just realised there was a "distributed computing" forum on here. Been reading back through some of the old threads so a big thank you to phaxmohdem for the epic stats work and to everyone who logs in and says hi/congrats to the joiners/milestoners.

I'm currently cranking out 18k ppd from an overclocked 2600K running (usually) 24/7. I've got crossfire'd 5850s in my machine too. Should I be running GPU client(s) too, or just stick with the 1 SMP client I have now. I live in England so excess heat is actually a plus I've not had my heating on in a month now!

 

[jibberegg]

Oh, and phaxmohdem isn't the only one who can do stats http://img545.imageshack.us/i/fhprogress.png/

 

[Sunny129]

Hi there all,

Apologies for not doing this sooner, but I only just realised there was a "distributed computing" forum on here. Been reading back through some of the old threads so a big thank you to phaxmohdem for the epic stats work and to everyone who logs in and says hi/congrats to the joiners/milestoners.

I'm currently cranking out 18k ppd from an overclocked 2600K running (usually) 24/7. I've got crossfire'd 5850s in my machine too. Should I be running GPU client(s) too, or just stick with the 1 SMP client I have now. I live in England so excess heat is actually a plus I've not had my heating on in a month now!

you definitely need to be running 2 GPU clients w/ your 5850's! i don't know that their crunching will be as efficient running a single GPU client in crossfire mode as they would be if they were running 2 separate GPU clients individually. i don't know how difficult it is to switch in and out of crossfire mode on a regular basis, but if its a PITA, just leave them in crossfire - they'll still get plenty of work done. either way, you must dedicate your GPU's to TeAm AnandTech, and more importantly, to science!

btw, welcome to the DC forum...

 

[GLeeM]

Welcome jibberegg

I thought I remember reading that you can run two GPU clients even if they are in Crossfire, although maybe it two on nvidia in SLI or whatever it's called. I'll let someone chime in on this to be sure.

More importantly, if you add the -bigadv switch to your CPU client you could probably almost double your points per day! The WUs take longer, maybe two+ days on your computer. If you are already using a passkey you will get the bonus right away, otherwise you have to complete ten WUs. Those ten WUs can be the ones you are doing now.

 

[biodoc]

Welcome to TeAm Anandtech, jibberegg!

 

[Sunny129]

I thought I remember reading that you can run two GPU clients even if they are in Crossfire, although maybe it two on nvidia in SLI or whatever it's called. I'll let someone chime in on this to be sure.

i think you're right. sorry if it seemed like i was implying that you could only run 1 GPU client while the cards are in crossfire.

 

[jibberegg]

OK, so had a bit of a read of the FAQ (http://folding.stanford.edu/English/WinGPUGuide) and it has the following caveats with CF/SLI set ups...

SLI or Crossfire does not have to be disabled to run the GPU client. However, SLI and CF cannot be used to make a dual GPU card or cards perform like one super GPU. The SLI and CF link does not have enough bandwidth to support that functionality. Run one FAH GPU client for each GPU chip.
If you are running multiple GPU cards with SLI or Crossfire, you may need to extend the desktop for the client to recognize multiple GPUs.
If Windows prompts for a new driver, hit CANCEL. The default Windows driver may not be the best driver to use. Allowing Windows to install a driver automatically runs the risk of having sub-par performances and/or client installation issues.
If you run Windows Vista or Windows 7, you may need to have each GPU connected to a monitor (a KVM can also work but not all models of KVM will provide a signal when it is not the active monitor). Another alternative is to use VGA dummy plug.

Think I might just stick to the one GPU client for now. Two looks like a bit of a hassle. Sure I'll get bored on a weekend and have a go though!

 

[theAnimal]

Welcome, jibberegg!

I would recommend not folding on ATI until the new client is released and if there is any significant increase in ppd, but definitely add the -bigadv to your smp client for a nice bump in points.

 

[Sunny129]

i think maybe i should have specified that he use his ATI GPUs for projects & apps like Milkyway@Home or SETI@Home Multibeam & Astropulse, simply b/c those apps are far more efficient running on ATI GPUs than on nVidia CUDA GPUs. in the mean time, if he's set on crunching Folding@Home, he should either stick to the MSP client or pick up an nVidia CUDA card since, unlike the MW@H and S@H GPU apps, the F@H GPU app is far more efficient running on nVidia CUDA GPUs than on ATI GPUs. or he could also do as you suggested and upgrade to BOINC client v6.12.18. while it may not be a fully released client, its no longer beta software, and is available for download. so if jibberegg feels the urge, he can certainly give v6.12.18 a test drive and see if it functions properly on his platform. if it does, he can see if F@H crunches any faster on an ATI card than it does when using the current BOINC v6.10.58.

but either way, jibberegg shouldn't let the substantial compute power of not one, but two 5850 GPUs go to waste! and if F@H isn't any faster on an ATI GPU while running under BOINC v6.12.18, then he should put them to work on two projects whose GPU apps are very efficient on an ATI GPU, such as MW@H and S@H MB & AP...of course that's just my humble opinion, and jibberegg should do with his hardware whatever suits him best in the end.

i myself don't have a CUDA card, only a double-precision ATI card. and so i let it do what it does best in the names of efficiency and productivity - and those projects are MW@H & S@H.

 

[jibberegg]

Well I've started the gpu client running at ~1am GMT 24th March (Note to self here ) so I'll try for a week and see what happens to my ppd. If it isn't much of a bump then I'll switch to a better optimised client. I'll hold off on doing the bigadv to make sure the test is clean. Hopefully science will cope without that extra power for a week!

 

[VirtualLarry]

Welcome, jibberegg!

I would recommend not folding on ATI until the new client is released and if there is any significant increase in ppd, but definitely add the -bigadv to your smp client for a nice bump in points.

Ditto. The ATI GPU client takes up more CPU cycles (I think?), and doesn't provide nearly as much PPD as the NVidia client. So if you have a high-performance CPU (2600K), then just running a SMP -bigadv, would probably be best. Unless you wanted to invest in some NV GPUs.

I run the regular SMP (bigadv requires support for 8 threads, and a fast CPU, faster than my Q9300 @ 3.0), and an NV GPU client on my GTX460 (back to stock 715 clocks for now), I get around 16K PPD for both CPU and GPU.

 

[jibberegg]

Cheers Larry. I've been checking my ppd and it actually looks like it's gone DOWN since running GPU client. I'm more than a little surprised I have to say! Didn't think it would take enough cycles away from the SMP to reduce it. Very odd.

Going to switch to bigadv and see what happens there...

 

[GLeeM]

Going to switch to bigadv and see what happens there...

The reason the points went down is because the bonus points are for early return - the sooner it gets back the bigger the bonus.
You will notice a bigger difference when doing -bigadv.

For example: The GPU shows 9,000 ppd and the -bigadv shows 30,000 ppd when both are running.
If the GPU is stopped the -bigadv will go up to ~35,000 ppd!
EDIT: the above example is from my i7 920 at 3.99 and GTX275
actually the GPU has lately been at 8,000 ppd - I should check again, it might not be worth the electricity to run it!

 

[Sunny129]

Cheers Larry. I've been checking my ppd and it actually looks like it's gone DOWN since running GPU client. I'm more than a little surprised I have to say! Didn't think it would take enough cycles away from the SMP to reduce it. Very odd.

Going to switch to bigadv and see what happens there...

you should also do as theAnimal suggested and try out the new BOINC client release (v6.10.58 was updated to v6.10.60 just yesterday). v6.12.19, the BOINC "use only for testing" version, also might be worth a try in the event that v6.10.60 does nothing to improve the run times of F@H tasks on your ATI GPU. i know v6.12.19 "may be unstable" and is to be used for testing only, but if you don't get any errors while using it, and it doesn't have any adverse effects (like slowing down tasks across the board, causing GPU conflicts, etc.), then i don't see why it can't be the version you run all the time. i've been running the previous test platform, v6.12.18, for weeks now without a hiccup. i'd direct link you to a download of each individual version, but i have no idea what OS you're running, so i'll just post a link to general BOINC client download page HERE.

at any rate, if neither v6.10.60 or v6.12.19 does anything to improve the run times of F@H tasks on your ATI GPUs, then i'd say its time to put your 5850s to work on other projects (like MW@H and S@H) so that you're not wasting precious GPU cycles. i'd at least do this until the developers can come up with a solution that allows ATI GPUs to handle F@H work without being dog slow compared to nVidia CUDA GPUs, and without significantly slowing down the pace of the SMP client.

 

[bryanW1995]

I just checked in my boinc project manager and I don't see a folding@home option. Is that only with the newest boinc versions?

 

[brownstone]

Welcome to the team jibberegg!

 

[Sunny129]

holy crap...i just realized i've been talking out of my @ss for the last few posts. i completely forgot that F@H is not a BOINC application. that is, it runs on some other [non-BOINC] client. so all that crap i mentioned about various versions of BOINC, kindly disregard. its also obvious that i misunderstood which client theAnimal was talking about. in retrospect, he clearly isn't talking about BOINC. sorry if i mislead anyone into thinking that F@H is a BOINC application.

despite my monumental (if not comedic) blunder, i still think you should use your 5850s's for a more ATI-friendly project/application until the F@H application & client code are optimized for ATI GPUs.

 

[jibberegg]

I believe internet forum convention is to assert your views more strongly when you realise you made a mistake. All of this "admitting an error" stuff, not the done thing

I hadn't switched to bigadv yet and the TOTAL ppd went down, which I think is really weird i.e. SMP + GPU < SMP.

 

[jibberegg]

Logged into BIONIC and it still has my S@H stats going back to September 2004. The PC history on there is a little nostalgic...

http://img832.imageshack.us/i/pchist.jpg/

 

[GLeeM]

which I think is really weird i.e. SMP + GPU < SMP

I tried explaining about this in the post above.

SMP points bonus is based on how fast the results get back. The faster it gets done the bigger the bonus.

If you run another program that uses too many CPU cycles then the SMP will take longer to finish = less points.

How SMP works: If you assign eight threads to work on it, then the work is divided into eight pieces. If each piece takes 30 minutes to finish and they all finish together then that is 1&#37; of the WU done.
If during that time another program were to use 2% of one core (as my GPU WU does) then the SMP thread that was on the same core would take 35 minutes to finish instead of 30 minutes - that thread delays the whole process so that it now takes 35 minutes to finish 1%!

Perhaps with AMD GPU it would be better to assign only seven threads to SMP? Or run a program on the GPU that does not use CPU cycles?

 

[petrusbroder]

I believe internet forum convention is to assert your views more strongly when you realise you made a mistake. All of this "admitting an error" stuff, not the done thing

() Yeah, you are right. Maturity is over rated .... ^_^ :biggrin:

 

[jibberegg]

Sorry GLeeM, I misunderstood. Thanks for the excellent clarification!