Got a new box - approx 25,500 pts/day

Lola_Doggy

Weaksauce
Joined
Jun 19, 2002
Messages
67
Hello, haven't posted for a few years.

Got an interesting rig recently built up, 2xE5520 plus 3x Tesla C1060's, it's for a project where I work.

Each Tesla seems to be good for around 7,700 pts/day and the 8 cpu cores are each running an instance of the regular client (smp client stills seems a bit flakey).

Well, see you in another few years.

.ld
 
Don't just go!

Welcome! Glad to see ya post.

Also:

1) There is a new SMP client that we will see sometime in the near future, keep an eye out for it.

2) You should be able to run a VM (very easy setup) to run -bigadv WU. With a setup like that you could see an extra 20k+ ppd from the CPU side of that box making it just shy of a 50k box.

We will be glad to help ya with anything you need.

Thanks for folding, Fold on!

 
Ok, I'm staying.

What's with all this big daddy WU business, I like the sound of that, is there a guide floating around ?.

Tks for peaking my interest.

.ld
 
criccio got it.

The guide should have any issues hammered out of it by now.

However, if you have questions or need clarification, we would be happy to help just post!
 
Well the only thing that seems to be a sticking point is that after I change the settings (name, team etc.) they don't seem to get saved after pressing the reconfigure button and after reboot of the VM it's back to Mr notfreds settings.
 
Hi, Welcome back to the [H]orde! :cool:

Well the only thing that seems to be a sticking point is that after I change the settings (name, team etc.) they don't seem to get saved after pressing the reconfigure button and after reboot of the VM it's back to Mr notfreds settings.
Which ditro are you using for your VM? Notfreds with -bigadv client is reported not to be supported. I don't know if this has been rectified. Your system is more than capable to run these WUs and if you have a lot of RAM, there really is no other way to get the most out of it. People are running their single socket i7s with -bigadv.
 
-notfred's IS NOT supported, I have tried to run it myself off an i7 my dad purchased and its completely buggy and very non-compliant.

You may have to go with a classic linux smp distro instead of the normal VM way to go in order to get -bigadv to work the way it needs to.

Now it may have juust been me, never know, lol! :)
 
Someone estimate LD's PPD with his Teslas x 3 and -bigadv...
 
Well he will only be able to run a single 8 core bigadv unit and finish in time. so that is gonna net him around 15k PPD plus the 22k he has on the teslas.
 
No VMs can run on 16 cores
And if you run two BigAdvs neither will finish on time
 
No VMs can run on 16 cores
And if you run two BigAdvs neither will finish on time
Hmm, this system has the equivalent of two i7 processors, so it should 'theoretically' be able to run dual -bigadv VMs providing sufficient (read: boatloads) of RAM is allocated for each. I'm assuming there is also sufficient clock frequencies and realize stock frequencies aren't appreciably high. I am accustomed to OCing and have OC'd everything from small form factor PCs to servers. So, I may be assuming too much with this setup if it stays at stock frequencies. I just don't know enough about dual S-1366 setups, without having worked with them yet.

If it indeed proves unattainable, a good compromise is to run one -bigadv VM with all the supported hardware thrown at it, and one standard SMP VM with the remaining resources. That is still a huge production from one box when taking the GPUs into account.
 
Hmm, this system has the equivalent of two i7 processors, so it should 'theoretically' be able to run dual -bigadv VMs providing sufficient (read: boatloads) of RAM is allocated for each. I'm assuming there is also sufficient clock frequencies and realize stock frequencies aren't appreciably high. I am accustomed to OCing and have OC'd everything from small form factor PCs to servers. So, I may be assuming too much with this setup if it stays at stock frequencies. I just don't know enough about dual S-1366 setups, without having worked with them yet.

If it indeed proves unattainable, a good compromise is to run one -bigadv VM with all the supported hardware thrown at it, and one standard SMP VM with the remaining resources. That is still a huge production from one box when taking the GPUs into account.

The E5520's runs at 2.26GHZ each
They cannot be OCd by any means found thus far.

Since you can only run 8 core VMs it would perform worse than an i7 @ stock.
Since HT is not 100% efficient you are not essentially doubling your processing power with HT.

If you run two 8 core bigadvs it will not finish. This is also true for the E5530s.
If you want to be able to run two 8 core BigAdvs you need at least X5550's and depending on your internets upload speed a mobo with working turbo mode.

I am running the E5530s (slightly faster @2.4Ghz) and I have the same problem.

/me anxiously awaits new SMP2 :)
 
Last edited:
Nitro, What is your frame time on bigadv with HT disabled on your E5530s? I realize that you lose some performance when not running HT, but maybe the extra bonus would make up for it.
 
The other option is to run a full Linux install as your OS and run the GPU's under wine.
That way you could run the bigadv with an smp 15 or smp 16 flag depending on how much of a cpu core the gpu's need under wine.
You should get ~30k PpD from the bigadv and ~20k from the gpu's, if you can get them to play nicely under wine.
So a 50k PpD is possible with a bit of work.

Luck ............ :D
 
The E5520's runs at 2.26GHZ each
They cannot be OCd by any means found thus far.

Since you can only run 8 core VMs it would perform worse than an i7 @ stock.
Since HT is not 100% efficient you are not essentially doubling your processing power with HT.

If you run two 8 core bigadvs it will not finish. This is also true for the E5530s.
If you want to be able to run two 8 core BigAdvs you need at least X5550's and depending on your internets upload speed a mobo with working turbo mode.

I am running the E5530s (slightly faster @2.4Ghz) and I have the same problem.

/me anxiously awaits new SMP2 :)
That's extremely unfortunate. I make it a point to purchase motherboards that have been tested and proven to OC, whether it's enthusiast, workstation or server boards. I have a pair of 'old' E5310s (1.6GHz) that are running @2.5-2.55GHz and a pair of E5320s (1.8GHz) @ 2.675-2.7GHz. This necessitated a combination of OC methods including BSEL mod plus software OC tool. Can't these processors at least benefit from a BSEL mod or has that been disabled by Intel in the Nehalem series? I wish I knew more about the 5500 series but the cost doesn't justify moving my older hardware into mothballs just yet. :(
 
He'd have to run a native install of Linux, no WM, to be able to get the most out of these processors with -bigadv.
 
Hello, haven't posted for a few years.

Welcome back man.
thumbsup.gif
The team can certainly use all the processing power we can get going in order to hold on to the # 1 spot. Things are a bit different with the clients now vs the "old" days. But there's lots of good folks here with the know how to get you running full speed. :)
 
Nitro, What is your frame time on bigadv with HT disabled on your E5530s? I realize that you lose some performance when not running HT, but maybe the extra bonus would make up for it.

48min TPF
 
I would think that if staying in windows is important, just run 2 8 core normal SMP setups, 7k per card plus around 7k per VM, that puts you at 35k or so, and you wouldnt need as much RAM as bigadv and the system would probly be a bit more responsive. Hell, if system lag is an issue, just run a 8 core and a 4 core, leaving 4 cores open, dedicate one core to each GPU and youll have 1 core untouched and 3 barely being used, should be fine for a responsive system

And welcome back, good to see some of the old timers coming back, stick around, I'm sure youll like what weve done with the place. If not, toss in some ideas, were always looking to make it better
 
Thanks for all the encouragement and messages.

Everything is up and running and seems to be stable. I had to do some experimentation with what worked best and didn't have any negative impact on overall system use.

This box has to run XP and right now does not have enough physical ram installed to support the extra large WU.

The best combination (for me at least) turned out to be 3 X GPU clients plus one smp 8 running under VM Player, seems to be good for something north of 30k plus a day. I'll have a more accurate read after it's be running for a few days.

My ppd has gone from around a 3k average up to 33k.

.ld
 
Outstanding, welcome again back to the fold Brother Doggy.
 
Well I never did really leave in the sense that where possible I kept as many clients running as I could, it's more that I tailed off the drop-ins some months after genmay split off.

Interestingly enough though I still have some utilitarian servers (albeit dusty) that have remnants of v3 client installs, those of course hark back to the days of mojo, yellow ducks, and trading 1st place with ocau on a seasonal basis etc.

Kind of fun really (and tempting) since there's a pile of sixteen more Tesla's along with four more supermicro chassis which is to be their final resting place.

Probably not a reality though, we do have to actually develop cuda code here.

.ld
 
Well I never did really leave in the sense that where possible I kept as many clients running as I could, it's more that I tailed off the drop-ins some months after genmay split off.

Interestingly enough though I still have some utilitarian servers (albeit dusty) that have remnants of v3 client installs, those of course hark back to the days of mojo, yellow ducks, and trading 1st place with ocau on a seasonal basis etc.

Kind of fun really (and tempting) since there's a pile of sixteen more Tesla's along with four more supermicro chassis which is to be their final resting place.

Probably not a reality though, we do have to actually develop cuda code here.
Glad to have you folding for us again. It would certainly be great if you can bring at least some of those systems online folding. :cool:

BTW, whatever happened to the GM folding team? Are they still folding, I haven't checked their stats in well over a year?
 
The box really needs at least 6 gig in it to run 8 cores. That's my understanding anyway. :)
 
An 8 core bigadv VM will need a minimum of ~4.5Gb just to run.
Mines useing 5,091,984k at this moment out of the 6Gb set aside for it.
8Gb in system.

Luck ......... :D
 
Yea but regular SMP wont use that much even running on 8 cores
 
An 8 core bigadv VM will need a minimum of ~4.5Gb just to run. Mines useing 5,091,984k at this moment out of the 6Gb set aside for it.
Bare minimum -bigadv installations can be configured with much less memory. My sole -bigadv installation is running on only 4GB of RAM, albeit as a native Linux box. The EVGA -bigadv VM is preconfigured with 3.6GB and believe that is probably the least it will run. How well or how fast it performs with that amount is another question. My native 4GB box is doing 35.5-36m TPF depending on the OC.

Yea but regular SMP wont use that much even running on 8 cores
Yeah, it won't use anywhere near that much. I tried 8 thread standard SMP with as little as 600MB and it ran both smoothly and quickly. -bigadv is another story.
 
But he's asking about -bigadv and we know that regular SMP doesn't work those. :) (nice ramp up BTW)

Yea he doesnt have that much ram so I just wanted to make sure he knew.

I know you guys already know. :)

Thx
 
Back
Top