• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

New kid on teh block!

T

tedkj

[H]ard|DCer of the Month - July 2008
Joined
Jun 25, 2008
Messages
138
Hi there,

Old avid [H] reader all the way back to the times of GeForce 256 and earlier. Been doing my share of cycletossing for distributed.net back in the day, but thought it would be uhm, smart.. to move over to something a lil' more worthwhile. Hence, I am as of the 24th, a member of the [H]orde!

I just grabbed myself a GTX 280 a week back (yaya, I know.. 4870 is teh bomb!), and I'm currently running at around 7600 PPD's on a T6850 C2D @ 3.5 with the GPU2 client. Got a 8800GT now that it replaced which I'm gonna fire up in my 2nd testboxen.


Now onto the questions;

I also have a rack of 9 Dell PowerEdge 2850 servers with random setups, some 2x 3.0 GHz Quad-Core Xenons, some 2x 3.0GHz Dual-Core Xenons.

I tried to setup the SMB client on one of these 2850's with 2x 3.0GHz Dual-Core Xenons (4 cores total, running WinServer2k8), but the work unit hardly moves in progress. Next I fired up 4x of the consoled versions of the CPU clients, and it's with running 100% CPU usage, however FahMon says I'm doing a wooping total of 285 PPD's (about 73-75 each core).

Are 4x Xenon cores 'teh sux' or is it because of WinServer2k8? Or are these normal numbers, even though it seems people are dropping -alot- higher PPD's from their low-end C2D's etc?

I can potentially run folding on all these servers (absolutley no cpu-intensive tasks run on them anymore), except 1, which would add to a total of 38x 3.0GHz Xenon cores, but if the PPD output is this low compared to other setups, it would be more trouble than I see as beneficial..

Final question;

I'm bitten by the bug and I'm going shopping large! What is the best bang for the buck setup ATM, and would it be smart to wait for F@H getting their GPU2 ATI client more optimized and see how the 4800-series fare?

Preemptive thanks for any help! :)


 
I actually have 2850's with Dual Core Xeon's running at 3.0GHz and I get around 4,000+ PPD on the 2653 WU's, and only around 2,000+ PPD on the 2665 WU's. Now that is running two SMP clients on Windows Server 2003.

The best setup is to run VM's and set up a VM per every two core's. Run Linux on them, and get them to crank out some serious PPD. My one box that has 2 Quad Core 2.66GHz Xeon's does almost 9,000PPD with this method.
 
Also, it seems like everyone has really good luck with the Q6600's and the new Q9450's. Putting them in a P35 motherboard and overclocking them seems to be the main practice around here, then you set up two VM's with Linux & the Linux SMP client, and you're good to go!
 
Damn dude! Looks like a lot of people are gonna get mowed over with that setup once you get it going. :eek:

Welcome to the [H]orde btw! ;)

 
Welcome to the [H]ardest team in the world, where madness reign with people playing with fire :p

Time to stock the [H]Mowed|Over club fridge with more beer for the new members.

 
tedkj my man, IMO you joined the #1 best foldin' team on the Internet. (me bias, nahh...., just tellin' tha' truth :p)

By the sounds of your gear it sounds like another one of those "satellite warm front setups" :eek: (when foldin' your gear shows up as a warm front on the weather satellites of weather forcasters) We've even got a guy that when all his gear is foldin' it shows up as a "volcanic eruption" on the weather satellites (Sunin I'm only kiddin") :rolleyes:

You're in luck because this team (team #33) has several members that have a large array of gear. Just off the wall, check with Tigerbiten, Sunin, Xilikon to name a few (there are several more and with my memory unlike an elephants I can't even name them all). We also have a couple of pretty sharp "cookies" in the lady department. :D

We have a gentleman by the name of relic that knows so much about IT that he seems to talk a foreign language sometimes (to those of use that aren't all that knowledgeable about IT, check out his last post, I can honestly say I didn't understand much except he unforked it) :)

Good Luck and Welcome to the [H]orde :p


 
Welcome indeed man!!

Nice farm.... should see you in my rear view here pretty quick:)

/beers and tamales for you!!

Fold On!!!

 
lelliott731 said:
I actually have 2850's with Dual Core Xeon's running at 3.0GHz and I get around 4,000+ PPD on the 2653 WU's, and only around 2,000+ PPD on the 2665 WU's. Now that is running two SMP clients on Windows Server 2003.

The best setup is to run VM's and set up a VM per every two core's. Run Linux on them, and get them to crank out some serious PPD. My one box that has 2 Quad Core 2.66GHz Xeon's does almost 9,000PPD with this method.
Click to expand...

This seems to be the route I have to go, but a problem might be W2k8 - which VM would you suggest running on this OS? Or rather, which one do you use for W2k3? VM2005?

Seems to be something very wrong with this picture, right?

Screenshot!

Thanks for the warm welcomes, much appreciated :D

 
tedkj said:
This seems to be the route I have to go, but a problem might be W2k8 - which VM would you suggest running on this OS? Or rather, which one do you use for W2k3? VM2005?

Seems to be something very wrong with this picture, right?

Screenshot!

Thanks for the warm welcomes, much appreciated :D

Click to expand...

Load them up with linux, and for the duals just run a single SMP client but on the quads run linux with smp then a vm running linux with smp.
 
jtg1993 said:
Load them up with linux, and run 1 vm per 2 cores, with a SMP client running in each vm.
Click to expand...

Idealy I'd boot em with linux, but these boxes need to run W2k8. Even tho it's lightly used, it's part of a storage cluster and used for some simple tasks from time to time. I -might- be able to do what I want with most of them tho as it's more or less non-critical work which they do now, but I'll have to clear up on that first.

What VM would you suggest for W2k8 tho?
 
Welcome.gif


Here comes more trouble in my rear view. :D


216
 
I use VMWare Server.
It's free to download, no time limit on how long you can run it for, you just need to register for a free starting serial number.
Which version of Linux64 you run inside the VM is up to you.

Luck .......... :D
 
Welcome...... and if you add anything more I'll see you in my rear view soon enough!

Fold on, Fold [H]ard!

 
Thanks for the info, guys!

I've been messing around abit trying to figure out a couple of things, but I still have som questions;

1) As I said before, I'm running a GTX280 and getting 7600-7700 PPD's on the GPU2 client. I got a C2D @ 3.5 on this box and I tried to fire up a CPU client at the same time. This didn't touch the GPU2 PPD's, but even at constant 50% load (across both cores, but overall some cycles left unused), the PPD performance was not above 350 PPD's. This seems very little to me.. so the question is, is this normal? Is it because of the CPU client delivering different WUs to be computed?

2) To further investiage I stopped the CPU client and set up a SMP client instead, letting it hog all the left over cycles. Now I got a constant 100% load on the cores. The GPU2 PPD's fell to 7200-7300, but I still havn't got a PPD read on the SMP from FahMon.. seems slow tho.

3) And finally, check out this picture please. It's running on a Dell 2850 with 4x 3.0GHz Xenon Cores (2x Dual Xenons) and only outputting 280 PPD's total. Again, why is this? Due to the CPU clients and their different WU's?

I will be setting up a testbox with 2k8 server with VMware this weekend and give it a roll with linux to see how it goes, but for now I'm scratching my head trying to see some sense in these low PPD scores.

If it causes trouble with 2k8, I'll switch atleast 2 or 3 of the 2850's to 2003 early next week. :D
 
Vmware Beta 2 works on both versions of 2K8, although it's easier on the 32bit version. With the 64 bit version you have to enable unsigned drivers, but I didn't notice any performance or stability difference between the two. I was running on 2k8 32bit but switched over to 64 bit to test out Hyper-V eventually.

 
anything is going to seem slow compared to the GPU2, but the points will make up for it. A small drop in the GPU to pick up another 1-2k on the SMP is well worth it.

Congrats.... your hooked :p

 
Kendrak said:
anything is going to seem slow compared to the GPU2, but the points will make up for it. A small drop in the GPU to pick up another 1-2k on the SMP is well worth it.

Congrats.... your hooked :p

Click to expand...

Yep, I dropped about 500-700 PPD on the GPU2 client, but the SMP is running on one core on the E6850 and doing 1380 PPD, so a gain of 700-900 PPD. Not too shabby for left over cycles.

I -still- don't get why I get 280 PPD on my 4 Xenon 3.0GHz cores tho.. It's really bugging me as I see people reporting great PPD's on similar platforms (albeit some with Linux, but not all).

*grmpf!*

Either way, I'm gonna shop for some more general boxens right about soon, and I'm thinking either one these as the initial multi-boxen buys (5 pcs to begin the farm with);

C2Q6600 CPU
8800GT 256MB GPU
Generic Intel P35 MB
Running 1x (or 2x) SMB + GPU2

or

C2Q6600 CPU
9800 GTX (About twice the cost of a 8800GT 256MB)
Generic Intel P35 MB
Running 1x (or 2x) SMB + GPU2

or

C2Q6600 CPU
3x 8800 GT 256MB in SLi
Generic nForce-750i MB
Running 1x SMB + GPU2
 
The Standard singe core client is benchmarked at 110 PpD on a P4 @2.8 Ghz.
So this works out at ~40 PpDpGhz.
The client range is 40 -> 150 PpDpGhz.
But getting anywhere near the top of the single client range, takes the newest hardware and a bit of luck with the protiens.
More normal range is 40 -> 75 PpDpGhz.

The SMP client is benchmarked at 1,760 PpD on a Dual Dual Xeon @2.2 Ghz.
So this works out at 200 PpDpGhz.
So it starts out 5x faster then the Standard client.
The client range is 200 -> 335 PpDpGhz.
Getting near the top of the range is easy with the correct configuration.
Running 2x clients on a Quad with the right protien gets you there.
Most of my Quads are running at 332 PpDpGhz.

The GPU client is benchmarked at 1,500 PpD on a 3850 running on a 4000+.
Exactly what PpD you get depends on which card you are running.
The ATI client needs a more powerfull CPU to keep it running at full power than a nVidia one does.
The range is 1,000 -> 7,500 PpD.

Hope that helps.

Luck ............ :D
 
Ah ok, that explains alot, Tigerbiten. Thanks :)

So the way to go then would be to go for the C2Q6600, place the unused cycles on SMB and focus on the GPU2 with either a 8800GT 256 or 9800GTX. Which one of those will give the most back for the buck, considering 9800GTX in Norway is at approx 2x the price?
 
Get a 8800GT, you will lose about 1000-2000 ppd only but for half the price. You can add a 2nd 8800GT later (no need for a sli motherboard, just have enough physical x16 slots).

 
Oh,

So I do not have to run the in SLI, just add cards? I.e 3 x16 slots, regardless of what chipset will allow me to run 3 GPU2?
 
tedkj said:
Oh,

So I do not have to run the in SLI, just add cards? I.e 2 or 3 x16 slots, regardless of what chipset will allow me to run 2 GPU2, or 3 GPU2?
Click to expand...

if the board has the slots it will let you run the cards...... it might not let you run them in sli mode, but for folding, sli doesn't matter

ninja edit: make sure you have on heck of a PSU to run all that goodness 24/7. 100% on 3 vid cards will make most PSU cry.

 
Great! If that's the case, I will go for this setup..(mm this is getting fun!);

MSI P7N Zilent nForce750i MB (3 PCI-Ex16 slots)
1x C2Q6600 @ 3.0GHz
3x 8800GT 256MB

Any suggestions on amount of RAM? It's dirt cheap these days anyho.. so picking up 4GB of DD2 won't hurt much.

I'll start with 2 or 3 of these and see how it goes, and expand as the wallet allows. Good thing I have a cold basement (and live in Norway) :p
 
Be warned.
The CUDA part of the client can only see one vid card at the moment.
So trying to running two clients on two vid cards does not work as they both end up on the first vid card.

Hopefully the next major upgrade of the client will fix this.

Luck .......... :D
 
tedkj said:
Great! If that's the case, I will go for this setup..(mm this is getting fun!);

MSI P7N Zilent nForce750i MB (3 PCI-Ex16 slots)
1x C2Q6600 @ 3.0GHz
3x 8800GT 256MB

I'll start with 2 or 3 of these and see how it goes, and expand as the wallet allows. Good thing I have a cold basement (and live in Norway) :p
Click to expand...

one note..... the multi GPU client for nvidia cards is not here yet.... I'd get one of the 8800GT's and get the other 2 in a few weeks when the client comes. (get the mobo that will support all of them though) IMO

anyone else ?

 
Ah ok, thanks.

Now over to figuring out how to max out these 2850's :)

 
You must log in or register to reply here.
Back
Top