Change in BA requirements

[bcavnaugh]

...Should I run ONE SMP Client on all 64 Cores or break them down to 4 Clients 16 Cores each (1 CPU) or down to 8 Cores and 8 Clients?...

One SMP Client should work. If you want, you can break it down into:
1) 24 CPUs
2) 16 CPUs
3) 12 CPUs
4) 8 CPUs
5) 4 CPUs

The reason for the above breakdown is that some SMP Projects are limited to 12 CPUs or 16 CPUs, etc. The reason is that they can't scale up to bigger values since the atom count is too small or some other factors. With the above breakdown, you cover a huge range of SMP Projects.

Thanks, would this work?
1) 24 CPUs client-type=bigadv / max-packet-size=big
2) 16 CPUs client-type=bigadv / max-packet-size=big OR only client-type=advanced
3) 12 CPUs client-type=advanced
4) 8 CPUs client-type=advanced
5) 4 CPUs client-type=advanced

Or should I keep them all only client-type=advanced or put nothing in?
Thanks

[PantherX]

With #1 and #2, you will still be assigned bigadv WUs until the changes are implemented. However, if you want to exclusively fold SMP WUs, just leave it at the default settings, i.e. no need to add client-type advanced or anything.

[bcavnaugh]

With #1 and #2, you will still be assigned bigadv WUs until the changes are implemented. However, if you want to exclusively fold SMP WUs, just leave it at the default settings, i.e. no need to add client-type advanced or anything.

Great, Thank you

1) 24 CPUs client-type=bigadv / max-packet-size=big 1st one is the longest one and slowest P8101. 20 Hours to complete, should be OK. 311807.7 for Credit
2) 16 CPUs for SMP 1st one P7645 for Credit 18488.1
3) 12 CPUs for SMP 1st one P7646 for Credit 15952.5
4) 8 CPUs for SMP 1st one P7647 for Credit 13723.4
5) 4 CPUs for SMP 1st one P7645 for Credit

My HFM Site: http://weather.mfc-cs.com/haf/ _FolditAMD64Cores4P Slots 00 to 04
I see how this goes for the next day or two. All Failed, needed 2 or more days to complete any so I removed all and reset as shown below

New setup No BigAdv
Name Client Type TPF PPD ETA Core Core ID Project (Run, Clone, Gen) Credit
16 Cores _FolditAMD64Cores4P Slot 00 Uniprocessor 00:08:13 24185.1 (1.753 WUs) 01:46:49 GRO_A3 A3 P8582 (R0, C2, G364) 13800.1
12 Cores _FolditAMD64Cores4P Slot 01 Uniprocessor 00:10:39 16503.5 (1.352 WUs) 05:51:27 GRO_A3 A3 P8574 (R1, C3, G151) 12205.7
12 Cores _FolditAMD64Cores4P Slot 02 Uniprocessor 00:10:33 16944.1 (1.365 WUs) 05:48:09 GRO_A3 A3 P8571 (R1, C1, G157) 12413.9
12 Cores _FolditAMD64Cores4P Slot 03 Uniprocessor 00:10:43 16140 (1.344 WUs) 05:53:39 GRO_A3 A3 P8577 (R1, C1, G105) 12011.6
12 Cores _FolditAMD64Cores4P Slot 04 Uniprocessor 00:10:44 16384.4 (1.342 WUs) 05:54:12 GRO_A3 A3 P8576 (R1, C0, G151) 12212.5
All at 67% Post Date 25 Dec 22:33 MST
4P 64 Core Rigs I do not think run SMP very well.

[orion]

Plus 16-24 thread, depending on hardware, may not make BA deadlines.

[NookieBandit]

In fact, "please" does work, at least with me.

I've got a 32 core AMD 6274 box currently folding BA, and glad to configure it to optimally run SMP. Is it best to simply change the Slot-Option to client-type "normal" and leave it at that, or is there a better configuration to utilize for this rig to efficiently fold SMP?

[PantherX]

...Is it best to simply change the Slot-Option to client-type "normal" and leave it at that, or is there a better configuration to utilize for this rig to efficiently fold SMP?

Changing the client-type to normal is fine. It is the default setting of the CPU Slot so if nothing is present, it means the same and may not be printed in the log (depends on the setting).

While a single SMP Slot of 16/24/32/64 CPUs might work, a possible limitation that you may encounter is limited number of SMP Projects. The reason is that due to different atom counts and other factors, there are some limits to the number of CPUs that a particular SMP Project can fold on. Thus, you can break down a single CPU Slot into the following combination to add-up to your total CPU count:
1) 24 CPus
2) 16 CPUs
3) 12 CPUs
4) 8 CPUs
5) 4 CPUs

Unfortunately, there isn't a way to figure out what projects have what limits so if you see yourself getting a steady stream of Project X, you may want to change the CPU values to different ones as shown above which may shuffle the Projects being assigned.

[cokeman54]

OK.

Please aim your resources at SMP rather than BA.

(I've got to admit I feel pretty stupid doing this.)

Will change over soon to regular SMP work. All you had to do was ask kindly. Merry Christmas and nothing stupid about this, all we can do is try.

[Horvat]

OK.

Please aim your resources at SMP rather than BA.

(I've got to admit I feel pretty stupid doing this.)

Why would you feel stupid asking for help? You have just earned a lot of respect from me. I posted this on the EVGA site and I will say it here; if you need help ask instead of forcing it down our throats. Old saying, you catch more flies with honey than with vinegar.

I would rather turn my resources to complete an -smp backlog than have my entire rack of 2P Xeon servers become useless for -bigadv. I have also posted at EVGA that I can understand the 24 core requirement but to follow it up 2 months later with a 32 core requirement is not right and not well thought out. Not to mention it will run a lot of folders off just like it did last time.

In conclusion, ask and thou shalt receive. I will unleash my server rack on -smp's when the current -bigadv WU's are complete.

Fold On!

[Bill1024]

Don't forget the WCG Christmas challenge is still going on.
And it may take a day or two spread the word. Plus with the Holiday.
Lets see how this shakes out over the next several days.
As of right now I am on holiday.
Merry Christmas

[Mike_Shaffer]

I agree with the latest posts from Bill1024 and the post on another forum. Points don't buy coffee. I added a few 8/12 core 1P tonight and will work on switching over a 40 core 2P to 16/16/8 and see if that works okay in V7.

Thanks PantherX for the smp core explanations.



Merry Christmas

[Punchy]

The real solution is so obvious - PG created too much demand for bigadv by making the reward high. All they need to do is create more demand for SMP by making its reward appropriately high - whether on a temporary or permanent basis. Begging seems like an acknowledgement that they really don't know how to manage their own points system.

Perhaps PG should also hire an economist that can help structure a reward system to manage supply and demand.

Back to the original topic, I offer 3 equally helpful bits of guidance which you may use to help with your future choices:
Geologist: there will be a major earthquake on the San Andreas fault in the future
Doctor: you will die in the future
PG: core requirements will change in the future

[mflanaga]

All you guys had to do was ask nicely in the first place. Just communicate the need next time rather than come up with artificial machinations to achieve your objectives.
Switching over next wu's!

[ChristianVirtual]

Geologist: there will be a major earthquake on the San Andreas fault in the future
Doctor: you will die in the future
PG: core requirements will change in the future

With those choices we all hope PG is faster

[Skripka]

The real solution is so obvious - PG created too much demand for bigadv by making the reward high. All they need to do is create more demand for SMP by making its reward appropriately high - whether on a temporary or permanent basis. Begging seems like an acknowledgement that they really don't know how to manage their own points system.

Perhaps PG should also hire an economist that can help structure a reward system to manage supply and demand.

I believe I can speak for lots of folks on my team in mostly agreeing with this.

I would add that GPU QRB put the nail in the coffin of SMP's worthwhileness for the power. The SMP projects only get more and more huge (i.e. 4p bigadv TPF size), and the returns get worse and worse....until finally for the handful of points per day, donors ask "why bother, when I can GPU or BA and actually have something respectable to show for my efforts?". Hell all BA donors have idle 1P desktop machines they would gladly fold on, if it was even *remotely* worth the electricity for the points and science. Crank the multiplier for 1p SMP to something similar to GPUs, and suddenly a ton more clients would come online and that backlog would vanish.

To quote the philosopher,"equal points for equal work".

[Macaholic]

And it may take a day or two spread the word. Plus with the Holiday.
Lets see how this shakes out over the next several days.

Thank you. Merry Christmas.