New A2 Core 2.10 for Bad 1 core WU [Archive]

HayesK

09-02-09, 08:38 PM

Got new A2 core 2.10 after deleting the old 2.08 core. The new core errors on the bad 1 core A2 WU. The same bad WU is downloaded and errors a few times, then a good wu is received. Did not have to delete the old work, queue or machine.dat files, only the core.

Only a few frames completed on the 2.10 core, so do not know how performance compares to 2.08.

Edit:
Appears the new core is not compatable with the work files completed by the old core. I deleted the old core on a partially complete wu and lost the work in progress due to error on the new core, and the same wu was downloaded and started at 0%.

GeneralMac

09-02-09, 08:57 PM

ChasR

09-02-09, 09:40 PM

It shouldn't affect -bigadv WUs This problem appears to me to be the result of a core released for testing that doesn't auto update (2.01 comes to mind). Anyway, WUs completed by some older core aren't compatible with core 2.07 or 2.08. I doubt the bigadv WU will even run on the older core, so they shouldn't be affected. Core 2.10 is supposed to be compatible with -bigadv Wus, but you can't change on a current WU without losing it as Hayes described.

harlam357

09-04-09, 10:13 AM

Is anyone else noticing an increase in frame times with this new core?

PPD on my 65nm C2Qs is down ~500ppd per VM and my 45nm Q9550 is down ~1000ppd per VM. :(

Voidn

09-04-09, 10:47 AM

Yes big time. My i7 is down 25% on 2 vm's that got the new 2.10 core yesterday. 4-4.4k now 3-3.4k per vm. Any way to switch back?

harlam357

09-04-09, 10:52 AM

Not sure... I've had some clients auto-update themselves, and others that refused to run new work until I downloaded 2.10. So I think the likely answer is no... or not for very long.

I didn't think this was a fluke since I've got 4 machines (8 VMs) doing this exact same thing.

I'm going to check over on FF.org to see what the word is...

Voidn

09-04-09, 11:16 AM

Mine auto updated. I had to kill all the files because of bad wu's and then it downloaded the new core on its own :(. So no going back....

While you're over there tell them I want my 2k PPD back!

HayesK

09-04-09, 11:53 AM

I just posted the info below on my poor core 2.10 performance over at Stanford forum.

My Q9550 rigs running single client in native ubuntu 8.04 have normal frame rates.

My X3220 rig running two 2-core ubuntu 8.04 vm are ~10-20% longer frame rate. Shutting down one 2-core VM on this rig allowed the other 2-core VM to return to normal frame rate.

My Q9450 and X3210 rigs running two 4-core ubuntu 8.04 vm were ~30% longer frame rate. Shutting down one 4-core VM on these rigs allowed the other 4-core VM to return to normal frame rate.

My E6600, E6420 and E6300 rigs running one 2-core Ubuntu 8.04 VM are ~10-20% longer frame time.

All the rigs running VMs have 4GB ram, XP32 host, VMs set to 1056MB ram, and at least one GPU client.

harlam357

09-04-09, 11:53 AM

I posted over there... couldn't find a thread dedicated to the 2.10 core... but someone else started a thread regarding the now much longer frame times. I can't believe there isn't more talk about it.

The one thing I can see possibly being the culprit is available memory... maybe the 2.10 needs more? Which is causing our VMs to get into the swap and slow things down?

Edit: Opps... didn't see HayesK's post. Ok, I'm running ~700mb per VM... so if it needs 300+mb more and still doesn't come into line... then something else appears to be awry.

Where's the thread you posted in Hayes? Here's the one where I posted...

http://foldingforum.org/viewtopic.php?f=44&t=11331

HayesK

09-04-09, 12:12 PM

Where's the thread you posted in Hayes? Here's the one where I posted...

http://foldingforum.org/viewtopic.php?f=44&t=11331

Located in the problem WU one core thread

http://foldingforum.org/viewtopic.php?f=19&t=11098&p=110709#p110709

HayesK

09-04-09, 02:17 PM

Are rigs with non XP32 host having similar longer frame rates in their 2-core VMs with core 2.10?

Jolly-Swagman

09-04-09, 03:40 PM

My Notfreds have Auto updated to 2.10 core and yes they are too running at 12-16TPF whereas they were before 8-10TPF

'Cuda340

09-04-09, 04:00 PM

The last of mine auto updated today.

I'm seeing a drop across the board regardless of setup, processor or available memory.

Would have to think it's the new core.........

harlam357

09-04-09, 05:00 PM

Are rigs with non XP32 host having similar longer frame rates in their 2-core VMs with core 2.10?

All of my rigs running 2-core SMP VMs are either XP x64 or Win7 x64. All are experiencing longer frame rates.

Edit: Still see nothing of merit from FF.org. I have a hunch that maybe an updated kernel might do the trick... I'm running basically out-of-the-box Ubuntu 8.04 installs as I don't bother updating them since all they're used for is Folding.

HayesK

09-05-09, 01:08 PM

I have come to the conclusion that core 2.10 does not run efficiently in Linux on two cores, reguardless of whether the 2 cores are in a VM or native Linux.

See my recent post http://foldingforum.org/viewtopic.php?f=19&t=11098&p=110855#p110855

HayesK

09-07-09, 10:49 AM

I created new VM yesterday using Ubuntu 6.06LTS desktop, applied updates, and deployed to seven rigs with 2-core CPUs. Using Ubuntu 6.06LTS kernel 2.6.15-54 with core 2.10, these 2-core CPUs are now getting frame times similar using prior VMs with Ubuntu 8.04 kernel 2.6.24-23 on core 2.08. These VMs are running on XP32 host using VMWare server 1.06/1.09 with 1056MB ram allocated to the VM. The hosts are running at least one nVidia GPU client in XP32, the CPUs are Intel 65nm 2M and 4M cache (E6300, E6320, E6600), and the boards have Intel P965, Intel P35 and nForce 680i chipsets.

Also posted over at FF.
http://foldingforum.org/viewtopic.php?f=44&t=11367&p=111130#p111130

Bobnova

09-07-09, 12:32 PM

As it happens i made a chart of my ppd at various cpu speeds, and now at 3.7ghz i get the same ppd as i used to get at 2.7ghz. That's brutal.

ChasR

09-08-09, 12:41 PM

Watch A2 Wus closely on core a2 v2.10 closely. Today I found 2 instances running on 2 or 3 cores at about 1 hour per frame. I found another that downloaded a bad WU and killed it several times, then downloaded a good one and killed it too. It then downloaded the same good work unit and proceeded to fold, but when I noticed it wasn't making progress, I found 8 active FahCore_a2 processes on a 2 core VM. Core v 2.10 is not only slow, it appears to be screwy too.

Bobnova

09-08-09, 04:17 PM

Interestingly, the PhII 810(at 2.85ghz) i just fired up is turning out more ppd then my e1200@2.3 and e5200@3.7 combined, by a fair bit.
The e120 makes ~1600ppd, the e5200 makes ~2400-2700, and the PhII810 makes 5020.
Is this new core only an issue for duals and not quads?

Wait, i see this reported a dozen posts up. File me in the quads-are-fine pile then i guess, one more data chunk :P

Voidn

09-09-09, 11:35 AM

I haven't had much time to test this issue. But what I have seen so far is the core 2.10 does seem use 100% on each core, so it would make sense that anything else using the same core(s) could interrupt it. But I run with my affinity set so one vm gets 2 dedicated cores and the other vm shares with the gpu's and system processes. The vm with 2 dedicated cores is running WAY slower than the vm sharing cpu time with 2 gpus.

I'm down to 5k ppd with my i7 instead of 8k-8.8k

I'll do some affinity/priority testing soon.

I'd hate to roll back to an older version of linux/ older kernel to fix this. Might try some newer kernels to see if that improves things.

ChasR

09-09-09, 01:26 PM

2.6.30 patched with Con Kolivas patchset is reported to be faster than 2.6.24. At least it's going forward and not backward.

http://foldingforum.org/viewtopic.php?f=44&t=11336

Bobnova

09-09-09, 01:59 PM

I'm down to 5k ppd with my i7 instead of 8k-8.8k

That's brutal. I'd try testing a 4 core client and see what happens.
The PhII 810 i've been playing with cranks out 5.4kppd@3.09ghz using all four cores in native(non-VM) ubuntu.

Voidn

09-09-09, 03:59 PM

I have come to the conclusion that core 2.10 does not run efficiently in Linux on two cores, reguardless of whether the 2 cores are in a VM or native Linux.

toTOW seems to agree. This is pretty good read on the issue. (http://en.fah-addict.net/news/news-0-20+a-focus-on-the-smp-a2-2-10-2-11-core.php)

2.6.30 patched with Con Kolivas patchset is reported to be faster than 2.6.24.

Interesting, I might give that a try before I redo my setup. Also a very informative thread on linux affinity :).

That's brutal. I'd try testing a 4 core client and see what happens.

Yeah, I've been meaning to redo my os/raid setup so I can go naitive or run 4 core vms. But its going to be a lot of work.

All this posses interesting questions like, if the new core uses 100% CPU and slows dramatically when knocked out of sync will having hyper threading enabled, without fully utilizing it, hurt performance.

Bobnova

10-02-09, 04:18 PM

Interestingly, i've gotten a couple single core WUs even though i'm on the 2.10 core. Strange.

In other news i tested Xubuntu 9.10 (beta) for folding performance (new OS kernal version!), no change from 9.04 in dual core SMP. Oh well.

HayesK

10-02-09, 07:21 PM

I changed all my 2-core VMs over to Ubuntu 6.06LTS to avoid the core 2.10 performance issue with the 8.xx and 9.xx releases. My 2-core native Linux builds were old enough to have several older kernels available in the grub boot menu, so I run them using the kernel which came with 606LTS. The kernel with Ubuntu 7.xx also runs core 2.10 without the performance issue.

Bobnova

10-02-09, 10:58 PM

That is very interesting, good to know for VM people.

I'm out of luck as it's my main OS, but still good to know.
Think it'd run faster in a 6.06LTS in a VM on 9.04? :beer:

ChasR

10-03-09, 07:12 AM

VM overhead is usually around 5% so you probably would be faster running on an older kernel in a VM.

Bobnova

10-03-09, 11:42 AM

I'll give it a whirl if i get a chance, hopefully the lack of VT in my e5200 won't put a damper on things.

EDIT:
Downloading 6.06LTS server image, the desktop can't be found anymore it seems. Server should be fine though, right?
It'll be done in about fifteen years (ok so 3.5 hours, feels like years), then i can test it.

HayesK

10-03-09, 09:55 PM

here is link to ubuntu 606 download at UT, which still has 6.06 AMD64 desktop available.
http://ftp.utexas.edu/ubuntu-iso/CDs/dapper/

Bobnova

10-03-09, 10:48 PM

I finally got the 606 server 64, but my copy of virtualbox won't let me use more then one CPU or use AMD64, donno what is up with that.

HayesK

10-04-09, 08:01 AM

Do not know what features VB requires, but the spec for the E5200 in your sig does not show support for virtualization.

http://ark.intel.com/Product.aspx?id=37212