New "many-core" work units for server-class Folding machines, available NOW

Leonardo · Feb 1, 2015

New "many-core" work units for server-class Folding machines, available NOW

We have a few new many-core projects released and upcoming.
Project 8106 is going straight to full FAH, as it's similar to some work units we've run before and is a control for the new projects.
Project 8107 simulates membrane fusion using tethers analogous to work by our experimental collaborators.
Project 8108 simulates what happens to membrane fusion when the fusion proteins are defective.

Announcement HERE

I'm Folding an 8108 'many-core' work unit one each of my 4P rigs. I'm trying not to get too excited. It's that roller coaster thing that I want to avoid.
But still, :thup:

, Dr. Kasson.

torin3 · Feb 1, 2015

Just got a 8106, time per frame is about 9 minutes.

orion456 · Feb 1, 2015

torin3 said:
Just got a 8106, time per frame is about 9 minutes.

Ah.....what's the points assigned I wonder?

Project 8106 is as follows:
5856.0 points
Preferred deadline: 4 days
Final deadline: 43 days
k-factor: 39.2

Compare that to 8101
9829 points
Preferred deadline 2.4 days
Final deadline 4 days.
k-factor 49.43

Hmmm.....

orion456 · Feb 1, 2015

My machines picked up the P8106 project over night and an updated HFM.net predicts 700,000 on my 64 core machine and 590,000 on my 48 core. That's the first time I have seen the 48 core do significantly less than the 64 core machines. Normally I would expect 500,000 ppd on each machine (not doing a 8101). Time will tell.

Leonardo · Feb 2, 2015

I'm Folding 8106 units now. My 48-core Magny Cours machines have never seen such a high PPD. I'm a bit dizzy right now. Yes, the 8108 was something a bit faster than an old BA 8101, but was slower than a BA 8102, 03, or 04. 8106 though, holy moly!

orion456 · Feb 2, 2015

P8601 is not being credited correctly. The bonus points are missing. Apparently this has now been fixed going forward. Missed bonuses will be credited later.

Credits are still not being assigned properly to some, others are receiving bonuses but much lower than HFM indicates..

Apparently there has been a hardware error as well as bonus miss calculations that are causing problems. The end result is that bonuses will be substantially lower than listed under HFM. It appears the new WUs are not being issued right now. When the errors are corrected the large WUs will start again.

orion456 · Feb 4, 2015

Big point updates in the last two rounds - PG is catching up on missed bonuses.

orion456 · Feb 7, 2015

The new cores will start again soon.

The new k factor is 3.92, substantially lower than first announced.

Leonardo · Feb 7, 2015

When the many-core work units were first released a few days ago, I was amazed at generous the PPD awards were. Guess my reaction wasn't unwarranted.
Does anyone have know exactly which flags we'll need in the client? I've thrown them all in: -verbosity 9, -bigadv, -bigbeta, -betateam, -smp.

torin3 · Feb 7, 2015

orion456 said:
The new cores will start again soon.

The new k factor is 3.92, substantially lower than first announced.

Exactly 1/10th of the factor first announced.

orion456 · Feb 7, 2015

torin3 said:
Exactly 1/10th of the factor first announced.

The deadlines are longer now, so how that will all play out in points we will have to wait and see.

For me, if the points are the same as for SMP, I will be switching over my BAs to Rosetta.

Currently, on the p7504s I am getting about 130k ppd which is what my single AMD video card is getting. It's not worth while to run a big board with 4 cpus and 32gb of memory when I can slap in a $250 video card and get the same result.

Leonardo · Feb 7, 2015

The deadlines are longer now, so how that will all play out in points we will have to wait and see.

Exactly. That should prove to be interesting. There's a mathematical solution of course, but I don't know the formula and don't know the processing time for these units with my computers.

For me, if the points are the same as for SMP, I will be switching over my BAs to Rosetta.

Yup. I'm currently getting about 180K - 240K with each 4P with standard SMP work units. 685 Watts per machine at 200K PPD would indicate to me I need to find a more efficient charity. I am not complaining at all concerning Pande Group. If they tell me the relative scientific value of a work unit and I know how much power (therefore money) that work is costing me, I will look for greater value for my money. Maybe that would be another DC project, maybe it would cash donations to charities of my choice, maybe it would be selling my current hardware and building Nvidia GPU folding systems. (Hmm, but then there's the Maxwell/Core 18/Nvidia drivers issue.)

We'll just have to wait and see. Fingers are crossed. I've been with Folding@Home since 2000 or 2001 (can't remember) and would really enjoy staying.

torin3 · Feb 8, 2015

Still getting nothing but A3s at the moment. Working out to about 150K ppd.

Leonardo · Feb 8, 2015

Same here, just an endless stream of SMP units. I'm using the V6 command line client. What about you, Torin? I'm also running the -bigbeta flag. Interesting, if I remove the -smp flag, I get this message in the Terminal window: "Connecting...Initial: 0000; + could not authenticate Assignment Server response...+ couldn't get work instructions." I'm thinking the assignment server is not too keen on the old V6 client. Both users I've seen posting, concerning a running 8106, were on the graphical v7 client.

torin3 · Feb 8, 2015

I'm also using the V6 command line client on [H]'s folding appliance. I've got -bigadv set, but I don't think -bigbeta. I did get a couple of the 8106's when they first came out, but that was it.

Leonardo · Feb 9, 2015

I've posted at Foldingforum about the issue. I think it's a server misconfiguration, or something like that. I've shut down my machines until this is resolved. And, if the resolution is that 'many-core' work units have the same science value as stock SMP, then I'm probably done with Folding. 675 Watts per machine is just too darn expensive if the relative science value is low. Such is life.

Macaholic · Feb 9, 2015

Leonardo said:
I've posted at Foldingforum about the issue. I think it's a server misconfiguration, or something like that. I've shut down my machines until this is resolved. And, if the resolution is that 'many-core' work units have the same science value as stock SMP, then I'm probably done with Folding. 675 Watts per machine is just too darn expensive if the relative science value is low. Such is life.

Using v6 with Ubuntu here and I am not seeing the issue. Double check firewalls, wireless connections, and Langouste (if used). I've seen flaky things happen with those at times. p8106 is now the same value as SMP. Hope you get it resolved.

WhitehawkEQ · Feb 9, 2015

Leonardo said:
I've posted at Foldingforum about the issue. I think it's a server misconfiguration, or something like that. I've shut down my machines until this is resolved. And, if the resolution is that 'many-core' work units have the same science value as stock SMP, then I'm probably done with Folding. 675 Watts per machine is just too darn expensive if the relative science value is low. Such is life.

I'll hold judgement till after I see how many of these I turn in per day, if 2-3 are returned per day it may make it worth wild or not

If worse comes to worse, I'll rent both 4P's out as servers

orion456 · Feb 9, 2015

Picked up an 8106 with the -bigbeta switch over night on a v6 command line. Using the points calculator and the new 3.9 k-factor. It works out to 175k ppd on a 48 core p4. Almost exactly what a p7505 is making on another p4. A p7504 is making 131k ppd. I'll let a few more download and see what they produce.

A quick look at my regular 4 core SMPs doing p7504 shows an expected ppd for 64 cores at 128k ppd which is in line with what I am seeing on my p4 (adjusted for clock speed differences). If the new cores will do 175k ppd that's a 35% premium which isn't bad; assuming the numbers hold up over time. But its a far cry from 500k ppd I was making.

Leonardo · Feb 10, 2015

I turned one machine back on this evening to try again. Finally, the assignment server gave an 8106. It's producing at about one-third the rate of what it was at first release of these new units and at about one-third the rate of the former Bigadv units. I'll let this one complete, I'll shut down the computer, then that may be it. Sigh. I've been crossing my fingers for a year, knowing this might turn out the way it did.

New "many-core" work units for server-class Folding machines, available NOW

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Premium Member

Member

Member

Similar threads