Linux SMP passkeys and bonus points

Come join the... uh... er... fold.

Moderators: just brew it!, farmpuma

Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 2:08 pm

After getting "had" by a string of unstable A2-core SMP WUs crashing out on me, about a week ago I decided that it was finally time to upgrade all of my Linux SMP systems to the latest version of the client. A few things that people need to be aware of on the latest Linux client:

- The new client will give you a lot of A3-core (a.k.a. "GRO-A3") WUs.

- Unlike older SMP WUs, the GRO-A3 WUs do not launch multiple processes to do the crunching; it is a true multi-threaded implementation. The upshot of this is that you will see only a single FahCore_a3 process running, but it will peg all of your CPU cores.

- You will want to register a passkey with F@h, and configure your clients to use it. Use of a passkey makes you eligible for bonus points on the GRO-A3 WUs, and it is the bonus points which make them decent PPD producers.

- The bonus apparently does not kick in until you've successfully turned in 10 SMP WUs using your passkey, and you need to maintain at least an 80% successful (and on-time) turn-in rate thereafter to continue receiving bonuses.

- The project summary page shows the raw (no bonus) point values for the GRO-A3 WUs. This makes it difficult to guesstimate how many PPD a GRO-A3 WU is producing while it is running, since the bonus is calculated based on how far ahead of the deadline you turn in the finished WU. The HFM monitoring tool (mentioned elsewhere on these forums) supposedly has the ability to estimate the bonus points (I have not independently verified this).

- Raw point values for most of the GRO-A3 WUs (so far) appear to be in the 300-500 point range, with raw PPD in the 250-750 range depending on the system (I have a mix of mostly older dual- and quad-core AMD systems). With the bonus factored in, I am typically seeing credits of around 1500-1800 points per WU, which makes them pretty good producers -- roughly on par with the old A2s (when they weren't crashing out).
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 38141
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 3:11 pm

Here is a link to the SMP2 bonus calculator.

To further provide examples of what you can expect from the new core in terms of ppd, my results are listed below -

Q9450 @ 2.8GHz ~ 6600ppd w/ 4 cores (native Mac OS 10.6)

Q9450 @ 3GHz ~ 4400ppd w/ 3 cores (Ubuntu 9.xx vm)
Fold! And I don't mean your clothes!

Do you have a favorite gerbil recipe? Please share with the TR community!
flybywire
Gerbil Jedi
 
Posts: 1883
Joined: Wed Jun 16, 2004 2:28 pm
Location: Springfield, VA - USA

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 3:19 pm

Thanks for the link! According to that calculator, I should effectively be doing ~3K PPD on my Phenom 9550 system, which seems consistent with what I've seen so far.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 38141
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 3:26 pm

I'm folding in Windows 7 using the new SMP client. It works very well when I get a3 cores.

About once a day, though, I get a1 cores, which don't fully utilize my CPU, take twice as long, and give me half the points. ARRGGHHH!!!
Image
If you're not folding with your idle computer time you're not part of the solution.
Gerbil Jedidiah
Grand Gerbil Poohbah
Gold subscriber
 
 
Posts: 3292
Joined: Thu Jul 27, 2006 5:57 pm
Location: The Smoky Back Room

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 3:33 pm

I think the WinSMP client users are getting the shaft.
Fold! And I don't mean your clothes!

Do you have a favorite gerbil recipe? Please share with the TR community!
flybywire
Gerbil Jedi
 
Posts: 1883
Joined: Wed Jun 16, 2004 2:28 pm
Location: Springfield, VA - USA

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 6:09 pm

All of my dual core machines are worse off than they were folding a2 core WU's. The C2D is losing the most, at least 500-600+ PPD, though it's probably the least as a proportion of the total PPD. It's odd that Intel's SSE advantage doesn't seem to be doing anything with these WU's.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5355
Joined: Sun Jan 20, 2002 7:00 pm

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 6:22 pm

Ragnar Dan wrote:All of my dual core machines are worse off than they were folding a2 core WU's. The C2D is losing the most, at least 500-600+ PPD, though it's probably the least as a proportion of the total PPD.
WinSMP or LinuxSMP under VMs?

Ragnar Dan wrote:It's odd that Intel's SSE advantage doesn't seem to be doing anything with these WU's.
My experience.
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24584
Joined: Mon May 24, 2004 2:19 am

Re: Linux SMP passkeys and bonus points

Postposted on Wed Feb 24, 2010 7:57 pm

Flying Fox wrote:
Ragnar Dan wrote:All of my dual core machines are worse off than they were folding a2 core WU's. The C2D is losing the most, at least 500-600+ PPD, though it's probably the least as a proportion of the total PPD.
WinSMP or LinuxSMP under VMs?

All of my machines except my desktop are native Linux. notfred's with manual download of the new client and configuration of the passkey (a PITA). The main desktop is claiming ~325 PPD before the bonus.

Flying Fox wrote:
Ragnar Dan wrote:It's odd that Intel's SSE advantage doesn't seem to be doing anything with these WU's.
My experience.

I read that, which is why I remarked upon it. My C2D running at 3136 MHz should be doing a lot better than your X2, I'd think, but it's giving 19xx PPD according to the calculator in this (and at least one other) thread. I don't know what HFM would say since I have yet to install my copy. I usually let things wait a week after downloading before installing them, and I often forget about 'em, too.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5355
Joined: Sun Jan 20, 2002 7:00 pm

Re: Linux SMP passkeys and bonus points

Postposted on Sat Feb 27, 2010 10:29 am

Update: I think I can now confidently say that getting everything switched over to use the new client and a passkey has brought my PPD back up to approximately where it was before:
Image
You can see on the graph where the WU troubles with the old client started (around the beginning on the month), and where the bonuses on the new client kicked in (on 2/18).
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 38141
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Linux SMP passkeys and bonus points

Postposted on Wed Mar 03, 2010 10:44 pm

After going through some trouble updating the BIOS on my Gigabyte GA-965P-DS3 (Rev 3.3) motherboard to get it to work with a newer processor, it's got the Q8400 running and it seems to be doing reasonably well at 3.2 GHz. It's presently crunching a P 6025 at ~6:49 per frame, which works out to ~5264 PPD, which is much better than what it had been doing. I also added another stick of RAM, though, after a shortened period of memtesting while dining, so that may have helped. Hopefully it won't crash. :)

Edit: For the record, I'm using the old 6400's stock heatsink on the "new" processor, because it's bigger than the newer one's. Its stock is something like .05v lower, too, though I raised it for the OC from 2.66 GHz to 3.2 GHz. Anyone need a nice 6400? :wink:
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5355
Joined: Sun Jan 20, 2002 7:00 pm

Re: Linux SMP passkeys and bonus points

Postposted on Thu Mar 04, 2010 9:42 am

Ragnar Dan wrote:After going through some trouble updating the BIOS on my Gigabyte GA-965P-DS3 (Rev 3.3) motherboard to get it to work with a newer processor, it's got the Q8400 running and it seems to be doing reasonably well at 3.2 GHz. It's presently crunching a P 6025 at ~6:49 per frame, which works out to ~5264 PPD, which is much better than what it had been doing.
Same 6025 project -
Pentium Dual-Core E6300 @2.8GHz: avg 14:48 per frame, 1645.4 ppd
Athlon II X2 250 @3.0GHz: avg 14:53 per frame, 1631.6 ppd

Compared to ~1400ppd on the E6300 and ~1200 on the X2 250 when they are working on A1/A2's, it is an improvement. It is not much though. Looks like with a quad this is going to be a bigger boost?

Ragnar Dan wrote:Edit: For the record, I'm using the old 6400's stock heatsink on the "new" processor, because it's bigger than the newer one's. Its stock is something like .05v lower, too, though I raised it for the OC from 2.66 GHz to 3.2 GHz. Anyone need a nice 6400? :wink:
The old HS has the copper core too, right? ;) BTW, for sale stuff should go to the BB? ;)
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24584
Joined: Mon May 24, 2004 2:19 am

Re: Linux SMP passkeys and bonus points

Postposted on Thu Mar 04, 2010 12:47 pm

Flying Fox wrote:
Ragnar Dan wrote:[...] Gigabyte GA-965P-DS3 (Rev 3.3) motherboard [...]'s got the Q8400 running and it seems to be doing reasonably well at 3.2 GHz. It's presently crunching a P 6025 at ~6:49 per frame, which works out to ~5264 PPD, which is much better than what it had been doing.
Same 6025 project -
Pentium Dual-Core E6300 @2.8GHz: avg 14:48 per frame, 1645.4 ppd
Athlon II X2 250 @3.0GHz: avg 14:53 per frame, 1631.6 ppd

Compared to ~1400ppd on the E6300 and ~1200 on the X2 250 when they are working on A1/A2's, it is an improvement. It is not much though. Looks like with a quad this is going to be a bigger boost?

The speed definitely makes a big difference, and it being nonlinear. Oddly, the machine slowed a bit after several more frames, and because I killed and restarted it several times (using notfred's CD one has to kill it quickly before it goes on using the old version of fah6) it seems to have mostly run at something around 6:55 per frame.

Edit: BTW I think I goofed up my passkey on the system at first, so I may not get much credit for the WU. But I stopped the client and ran -configonly to set it, so it should work from now on. And it has its own CD from notfred with the passkey and -smp 4 set, so now I only have to kill the client, download the 6.29 client, untar it and run it again and all should be well whenever it restarts.

Flying Fox wrote:
Ragnar Dan wrote:Edit: For the record, I'm using the old 6400's stock heatsink on the "new" processor, because it's bigger than the newer one's. Its stock is something like .05v lower, too, though I raised it for the OC from 2.66 GHz to 3.2 GHz. Anyone need a nice 6400? :wink:
The old HS has the copper core too, right? ;) BTW, for sale stuff should go to the BB? ;)

To tell the truth I forget, now, but I think it does. Hope so, anyway.

Yeah, I may put it up there if I don't use it as a replacement for the old X2-3800 I've still got running. It's running on an old Epox AGP board, unfortunately, so not likely to sell the whole thing.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5355
Joined: Sun Jan 20, 2002 7:00 pm


Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: No registered users and 0 guests