Dual Xeon -bigadv worklog

Seriously? That soon? :eek: :cool:

I don't know when it will be out, I would think that most of the hard work has been done with the A3 core, just need to package WU.
 
Last edited:
I broke down and put it in a case last night, might help with the fans.... they are from foxcon, but I swear they are sourced from Delta :eek:

delta.jpg


nexus4.JPG


nexus5.JPG


nexus6.JPG
 
I recommend a fan controller or adjust them in the BIOS if there's anyway you can do it. Love the case, BTW! :cool:
 
I recommend a fan controller or adjust them in the BIOS if there's anyway you can do it. Love the case, BTW! :cool:

4 settings for the HSF

Workstation
Generic
High Density
Full speed

I wish there was something in between Generic and High Density.

Generic does a fine job cooling the CPUs with HT off..... not so much with HT on.
 
I like my P182 as well. It is in my living room so it's nice that it's just a solid silver/black with no windows and bling. The noise damping capability is pretty effective on internal parts.
 
This weekends test: -bigadv EVGA VM on 8 threads and Windows SMP2 soaking up the rest.

Lets see how much this messes things up :eek:
 
Well.... if someone can figure out how to get the windows client to lock just on HT or the VM to lock to just LC this was a bust.

The SMP2 client added about 9k. It however moved the -bigadv unit to a TPF of 53 min.

I tried Bills PM but it wont let me assign the VM to just logical cores.
 
Well.... if someone can figure out how to get the windows client to lock just on HT .

Are you asking how to keep the Windows cores on cetain CPU cores after a new WU has started?

If so, set the Client to the cpu cores you want. it will carry over the settting to each wu that is started.
You will also have to set the first WU affenity manaully since it has already started prior to setting the above affenity.
 
Kendrak I forget...is this PC only for folding or is it for general use as well? Are you going to have a GPU folding in this box as well?

If it is dedicated for folding with no GPU, you might as well think about just running native linux, which with that setup would be good for 22-25K PPD with just regular A3 cores.
 
Iyou might as well think about just running native linux, which with that setup would be good for 22-25K PPD with just regular A3 cores.

A3 gets the same points per day in Windows & Linux.
(Actually the Windows A3 seems just slightly more stable then it linux counterpart)
Dual E5530s will get between 17,500k - 20,000 points per day with current A3 projects.

You should not need any 3rd party application to set and keep affinity.
 
Is there a tag that I would start the client with?

If there is, what would it be?
 
Well.... if someone can figure out how to get the windows client to lock just on HT or the VM to lock to just LC this was a bust.

The SMP2 client added about 9k. It however moved the -bigadv unit to a TPF of 53 min.

I tried Bills PM but it wont let me assign the VM to just logical cores.
HT doesn't work like that. You don't have logical cores and physical cores. All you have are logical cores, and each pair of logical cores runs on one of the CPU's physical cores. The core that a set of logical cores run on will divide its resources between the two logical cores depending on load. As far as the OS is concerned, there is no distinction between a logical core and a physical core.
 
HT doesn't work like that. You don't have logical cores and physical cores. All you have are logical cores, and each pair of logical cores runs on one of the CPU's physical cores. The core that a set of logical cores run on will divide its resources between the two logical cores depending on load. As far as the OS is concerned, there is no distinction between a logical core and a physical core.

So i need to make sure that priorities are set correctly or I'll steal from the -bigadv correct?
 
OK with a bit of help from Evil (informing me that I'm an idiot), and you too Zero ;)

I think I've got this figured out. I'll let it run the night and see what we get.
 
OK I don't have a TPF update on the -bigadv.... they take a while. But the SMP2 is running at just under 9 min, or 3500 ppd. Worth it if my TPF on the big adv stay around 33 min.

If the TPF on the big ADV does stay the same that would mean that there is a good big of CPU power that is not getting used. I'm wondering with all the stuff running though memory if I will lose 1-2k ppd on the big adv.

Tonight will tell.

Edit, TPF went up to 38:48 from ~33:00. That is a loss of 5k ppd. Not worth running the 2nd client.
 
Edit, TPF went up to 38:48 from ~33:00. That is a loss of 5k ppd. Not worth running the 2nd client.
I hear that, I stuck three GPUs in my -bigadv boxen and my TPF went from 30:00 to 33:40. Thankfully the GPUs are doing around 22K PpD so the loss of points from bigadv doesn't hurt that much. I think it may be time to put a system together just for my GPUs. :rolleyes:
 
OK I don't have a TPF update on the -bigadv.... they take a while. But the SMP2 is running at just under 9 min, or 3500 ppd. Worth it if my TPF on the big adv stay around 33 min.

If the TPF on the big ADV does stay the same that would mean that there is a good big of CPU power that is not getting used. I'm wondering with all the stuff running though memory if I will lose 1-2k ppd on the big adv.

Tonight will tell.

Edit, TPF went up to 38:48 from ~33:00. That is a loss of 5k ppd. Not worth running the 2nd client.
How did you assign the affinities? Can you try setting the VM on the first 8 logical cores and SMP2 on the next 8? Or is that what you did already?
 
I hear that, I stuck three GPUs in my -bigadv boxen and my TPF went from 30:00 to 33:40. Thankfully the GPUs are doing around 22K PpD so the loss of points from bigadv doesn't hurt that much. I think it may be time to put a system together just for my GPUs. :rolleyes:
Yeah, I know exactly where you're coming from. Only one of my -bigadv systems is running GPU clients. They draw a significant amount of CPU cycles. I see an improvement just from the difference between the 353s and 1888s WUs. I could imagine how much better that system would perform without a single GPU client and running native Linux instead of Win 7.

I have been thinking of shutting down my system with the Ati client (good for only 2000 PPD max) and building another with enough PCI-E slots to accommodate many of my single-slot cards around my farm. My cards are mid-range cards at best and starting to show their age. I can't really sell them for anything decent and I'm trying to see what the best solution is in deploying them. I would need an affordable board with as many PCI-E slots as possible for starters.
 
I will have 1x of my new systems offline for a week.Starting getting windows errors left and right.
Turns out I have 1 or more faulty memory sticks.
Running memtest+ now to determine which ones are defective.
 
Bummer Evil, hopefully you can identify the bad stick(s) quickly.
 
Well it seems jumping to CentOS we a bit extreme for me.

Ubuntu on the other hand seem to get along with me better.

I'm running -smp 15 on V9.10. I'm getting TPF from 29:05 to 29:33. Good for just over 30k ppd on the bonus calc. Don't know what HFM says yet, because to set up sharing I have to do a reboot and I want to let this run for a while.

It also seems that the RAM usage isn't linear. 15 threads and the system is using 6GB even. Looks like 8 GB is plenty on this rig... won't stop me from upping it to 12 ;)
 
Well it seems that with Ubuntu 9.04 and 9.10 with -smp 15 and -smp 16 I can't get the turbo boost to kick in. Worse off with 9.04 is seems that it speed steps back down to 1.6ghz :(

I'm going to try and bios settings (kill off speedstep) and see what happens.
 
Nope... I'm back in windows.
running just the smp -8 with no HT I'm guessing?

I wonder if you ran one bigadv and one SMP2, but played around with alternating the threads on your logical CPUs.

I actually haven't fired up SMP2 yet, so I'm not sure how it works, but if it's like SMP1, where it launches multiple worker threads, could you maybe run

one bigadv on CPUs 0, 2, 4, 6, 8, 10, 12, and 14,
with the SMP2 on CPUs 1, 3, 5, 7, 9, 11, and 13
 
I wish a manufacturer would release an affordable (relatively) dual S1366 board that can OC, like a modern Asus PC-DL equivalent. Or at least, someone would come up with a way to OC this architecture like it was possible with Xeons in the past.
 
Well i noticed in the BIOS of my SM X8DAL-i the Multiplier is only 18, where as on the X58s you can set it up to 21.

So i wonder if you could possibly create a custom BIOS?

What tools would be needed to do this? I'd try it on my box.
 
I wish a manufacturer would release an affordable (relatively) dual S1366 board that can OC, like a modern Asus PC-DL equivalent. Or at least, someone would come up with a way to OC this architecture like it was possible with Xeons in the past.

I don't know about affordable but evga should be releasing their dual 1366 mobo soon. I think I read something along the lines of $500. Given the cost of single socket enthusiast boards that doesn't sound too unreasonable. Now if only they could work voodoo so i could go get 2 microcenter i7 920s instead of spending $400 a piece on slower cpus I would be all over it.
 
I don't know about affordable but evga should be releasing their dual 1366 mobo soon. I think I read something along the lines of $500. Given the cost of single socket enthusiast boards that doesn't sound too unreasonable.
Yes, were it to be released at $500. The EVGA board makes the Skulltrail look run-of-the-mill by comparison, yet the latter's release price was stratospheric. I'm hoping it will be reasonably priced, but thinking in more realistic terms and looking at prices EVGA has on all its apex products, I wouldn't bet the farm on it.

Now if only they could work voodoo so i could go get 2 microcenter i7 920s instead of spending $400 a piece on slower cpus I would be all over it.
True indeed, but I have not seen this happen since the BP6 days, and that was only possible because regular processors could run in dual sockets at that time. Shortly after, Intel prevented it to ensure their enterprise level products retained enterprise level pricing and exclusivity. If EVGA can pull a rabbit out of the hat, kudos to them and they gained a life-long customer.
 
I don't think your going to get i7 class procs working on a dual board. From what I've read dual proc systems need chips that have 2x QPI links (I think that is the term... on my phone) All the i7 chips only have one while all the 55xx xeons have 2.

It simply just won't work.

Custom bios would be great. But we would all have to decide on getting the same mobo or some one has a bunch of work to do.
 
I don't think your going to get i7 class procs working on a dual board. From what I've read dual proc systems need chips that have 2x QPI links (I think that is the term... on my phone) All the i7 chips only have one while all the 55xx xeons have 2.

It simply just won't work.

Custom bios would be great. But we would all have to decide on getting the same mobo or some one has a bunch of work to do.

Thats correct the i7's wont work in Duallys because of the QPI.

But the custom bios would be for "OC" type options for us using the Dual Boards.

Kendrak

What type of BIOS does the ASUS use?
 
I don't think your going to get i7 class procs working on a dual board. From what I've read dual proc systems need chips that have 2x QPI links (I think that is the term... on my phone) All the i7 chips only have one while all the 55xx xeons have 2.

It simply just won't work.
I know it wont work. Thats why i said work voodoo. If they used magic I'm sure they could make it work. :D

I had picked out 2 E5520s until I found out you can not OC with current boards. Which means they really wouldn't fold much better than my i7 at 4.0.

If some one makes a custom bios for one of the cheap $250 dual socket asus mobos ill be all over it.

Ram prices still hurt though.
 
I'm back in Ubuntu 9.04

I noticed something... It does NOT like Firefox when I'm using -smp 15. Firefox seems to pull down about 4-5 threads to 1/2 speed. Not sure why, but once I closed it they all shot back up to 100%

This might explain last time I ran -bigadv in Ubuntu I was getting 40+ min TPF, got fed up and then switched back to windows A3.

I'm running it now, with no firefox opend and AM getting about 28:30 TPF. or 30k ppd with -smp 15. I do wish turbo boost would work. I would probably see 27-26 Min TPF.

Anyone know how to force turbo boost in Linux? Or do I need to switch distros?
 
Anyone know how to force turbo boost in Linux? Or do I need to switch distros?
It could very well be Ubuntu causing your problems. I've read several bug reports stating turbo boost doesn't always work with 9.04 and 9.10 on some systems.
 
Back
Top