• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Question on config for max PPD.

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
I have two dual quad core i7 E5540s (Dell r710) running ESX 4.0. I can give each VM anywhere between 1 and 8 vCPU. Right now I am running 1 VM with 8 vCPU on each server. Should I be running more to take advantage of the HT that these CPUs have? Maybe another 4 vCPU VM per host? Or even another 8 vCPU machine?

What about the big WUs and bonus points? How does that all work? Can 2 VMs per box finish them in time?

EDIT: Currently the machines are running ~2:40 min per frame on an A2 core WU.
 
Last edited:

Shelnutt2

Overclockers Team Content Editor
Joined
Jun 17, 2005
Location
/home/
I've never run the big work units, but from what I understand you can only run a single one per i7. So with a dual quad, you probably could run two, assuming you have 8 gigs plus of memory. So setup each vm for 8 cores.

Finish the current workunit (stop the client and restart with -oneunit flag). After it's done, download http://www.stanford.edu/~kasson/folding/linux/fah6 and replace the current fah6.

Then run fah with the -bigadv flag. Also you need to setup a passkey, http://folding.stanford.edu/Spanish/FAQ-passkey .
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
I've never run the big work units, but from what I understand you can only run a single one per i7. So with a dual quad, you probably could run two, assuming you have 8 gigs plus of memory. So setup each vm for 8 cores.

Finish the current workunit (stop the client and restart with -oneunit flag). After it's done, download http://www.stanford.edu/~kasson/folding/linux/fah6 and replace the current fah6.

Then run fah with the -bigadv flag. Also you need to setup a passkey, http://folding.stanford.edu/Spanish/FAQ-passkey .

How do bonuses factor in. I know the faster I finish the larger the bonus. Where is that break even point? 1 WU will clearly finish faster on a box with only 1 VM, but am I wasting processor power by not utilizing the HT with a second VM?
 

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
A single instance run at -smp 16 is much more productive than two at -smp 8. The multiplier is calculated as follows:

Bonus multiplier = If WU_Time < Preferred_Deadline then SQRT(Final_Deadline x Kfactor / WU_Time) else 1.

So the multiplier on a WU that takes 31 minutes/frame, 2.15 days, to fold calculates like this:

multiplier = SQRT(6x2/2.15)= 2.361

WU_Credit = Base_Value x Multiplier
WU_Credit = 59,976 = 25403 x 2.361

PPD = WU_Credit / WU_Time
PPD= 27,896 = 59,976 / 2.15


With your configuration, being limited to 8 vcpus per VM, even though you have 16 logical processors, I'm not sure how you'd fare with two -bigadv instances. You still only have 8 FPUs so adding a 2nd client will probably slow both and give you less ppd.
 

Shelnutt2

Overclockers Team Content Editor
Joined
Jun 17, 2005
Location
/home/
How do bonuses factor in. I know the faster I finish the larger the bonus. Where is that break even point? 1 WU will clearly finish faster on a box with only 1 VM, but am I wasting processor power by not utilizing the HT with a second VM?

The bonuses factor in only if you get it in before the preferred deadline.

Project 2681 is the initial project in this series. It is benchmarked at 25403 points, preferred deadline 4 days, final deadline 6 days.
We recommend running on systems that can complete this work unit in <= 3 days.

However the faster you complete it the higher the bonus.
At current time, the maximum bonus factor is 10x (completion of a 2681 work unit in 1/50th the deadline = approx 2 hrs 50 min). If you're in a position to exceed that, talk to us.

The formula is:
If WU_time <= timeout time, bonus factor = sqrt(deadline_time * k / WU_time)


http://foldingforum.org/viewtopic.php?t=10697


I think you are just going to have to experiment. Since you have two dual rigs. I'd run one with 2 bigadv, and one with 4 bigadv. See which one give you more ppd, and then you'll know.
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
Alright, I just replaced fah6 and set the -bigadv flag on one of the VMs. I'm going to let it finish the other unit and see what happens. I'll post up frame times later once it gets going.

ChasR the actual CPU load on the physical boxes is 100% so maybe I'll skip trying to overload them with another 2 vCPU or 4 vCPU machine. Let me know what your thoughts are.
 

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
Based on your 2:40/frame times on A2, I figure you ought to hit about 27:30/frame on a single -bigadv. That'll earn you 33,300 ppd. I would let the big dog run with the exponential bonus scheme.
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
Based on your 2:40/frame times on A2, I figure you ought to hit about 27:30/frame on a single -bigadv. That'll earn you 33,300 ppd. I would let the big dog run with the exponential bonus scheme.

~29:30min/frame. How does that compare to a single socket i7 w/HT? What should I expect for PPD out of this?
 

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
TPF varies from 31:00 to 33:00 depending on the WU on a i7 @ 3.8 GHz, when run in a VM. 32:00.0/frame = 26,560 ppd

29:30/frame = 30,000 ppd.

Numbers are ignoring upload and download time.
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
TPF varies from 31:00 to 33:00 depending on the WU on a i7 @ 3.8 GHz, when run in a VM. 32:00.0/frame = 26,560 ppd

29:30/frame = 30,000 ppd.

Numbers are ignoring upload and download time.

Is that before or after bonus? Is there a list of all these WUs and their point values somewhere?
 

jonspd

Super Speedy Senior
Joined
Jan 18, 2003
I would use HT and see what 2 clients on bigadv- units get as for PPD before I just decide to run only 1.
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
I would use HT and see what 2 clients on bigadv- units get as for PPD before I just decide to run only 1.

I might do that tomorrow. How fast are the i7 quads that people are running? I'd guess I'd get about a 50% hit on times if I ran two 8 vCPU machines. So figure in the 45ish min per frame. How does that factor into the bonus points? I think I could get 2 VMs running and meeting the deadline, but the bonus heavily factors in time.
 

jonspd

Super Speedy Senior
Joined
Jan 18, 2003
Not sure I want be able to test bigadv until I figure out what 6+ gb set to get.

I don't see it taking a 50% hit. Maybe like 35-40 is my guess... (could be wrong)

if its a dual quad thats 1 8 core smp big adv per cpu really you have 16 cores = 2x8 core smp per rig right?
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
Not sure I want be able to test bigadv until I figure out what 6+ gb set to get.

I don't see it taking a 50% hit. Maybe like 35-40 is my guess... (could be wrong)

if its a dual quad thats 1 8 core smp big adv per cpu really you have 16 cores = 2x8 core smp per rig right?

My VM is only using 2GB of ram. I have assigned it 6GB.

The physical servers are dual quads w/HT. I'm running 1 VM on each server with 8 vCPU right now and I am not taking advantage of HT.
 

jonspd

Super Speedy Senior
Joined
Jan 18, 2003
understood but I would go ahead and take advantage of that HT.

if it weren't for the ht the i7 920's wouldn't be able to do the bigadv- units I thinks.
 

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
First off the main purpose of the bonus scheme is to discourage folks from running two instances. The result is huge bonuses for early completion.

There are only 8 FP units available. So while it may not double times, running 2 x 8 core WUs will probably lengthen the time of each by 70% or so. Assuming that's close, your frame times would increase to 50 minutes. your ppd would be about 13,600 per instance, less than a single instance.
 

Adak

Senior Member
Joined
Jan 9, 2006
PMSfishy, all this advice to increase your production sounds very dodgy all around. I think you should wait a couple of weeks and consider all the options, etc.

Whatever you decide, I think you should phase it in, over a month or so - after I've passed you sounds perfect! :p :p :p
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
PMSfishy, all this advice to increase your production sounds very dodgy all around. I think you should wait a couple of weeks and consider all the options, etc.

Whatever you decide, I think you should phase it in, over a month or so - after I've passed you sounds perfect! :p :p :p

I'll take that into consideration. Too bad I saw that you were catching up and decided that I had to do something.
 

ChasR

Senior Member
Joined
Apr 12, 2004
Location
Atlanta
If you could run a 16 core VM, you would maximize production. Production figures for the dual 5540 posted on FF are all over the place from 23 to 29 minutes per frame with HT enabled. It appears the Linux kernel makes a significant difference. 2.6.31-14 and up are the ones to have. If you produced WUs 3 minutes faster in a 16 core VM, production @ 26:30/frame would be 35,250 ppd.
 
OP
{PMS}fishy

{PMS}fishy

Senior Member
Joined
Dec 20, 2001
Sweet, just turned in my first big WU. Not sure how many points I got for it though. There were 2 other WUs turned in at the same time. So 3 WU = 62,206 points. I have 1 GPU2 client and 1 other SMP client running. I'd guess worst case that I ended up with around 58,000ish points with bonus. It took ~51 hours to complete. That works out to ~27294ppd with my fuzzy math.

ChasR how does that sound?

Also I have some quad Core2 Quad Xeon boxes that I wanted to get going again. I think they are 3ghz. How do you think that will do? I'm guessing much better with the bonus scheme. Also how fast do you think it would finish a frame on a regular SMP WU with -smp 16? If I could do 20 in a day that is 38,000 points on standard WUs.