• 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.

Magny-Cours?

FLECOM

FLECOM

Modder(ator) & [H]ardest Folder Evar
Staff member
2FA
Joined
Jun 27, 2001
Messages
15,828
Anyone tried out any of the new Opteron CPU's by any chance?

I was thinking one computer with 48 cpu cores would get some sick points... thoughts?
 
Zero82z said:
About the same as an Intel system with half as many cores: http://www.bit-tech.net/hardware/cpus/2010/03/31/amd-opteron-6174-vs-intel-xeon-x5650-review/9
Click to expand...

True, but that is a dual Magny Cours system with 2 12 core processors. I'm very curious to see how 4 of those 12 core processors would perform with a total of 48 cores. Would even the Linux build be able to take advantage of so many cores, or would one end up having to run multiple clients in order to take advantage of it?

Even if the Intels can perform close with half the cores, you have to admit that it is sweet and a lot more reasonably priced than one would expect out of such a setup.
 
I did some initial analysis on this a couple of weeks back.

4 x AMD Opteron 6172 Magny-Cous 2.1GHz = $4,396 (assuming $1099 per CPU)
MoBo: Approx $850 (SuperMicro and the Tyan is pretty comparably priced)
Total Cost (not inclusive of RAM, HDD,...) = $5,246.00
12 x 2.1 x 4 = 100.8 GHz of Processing Power

2 x Intel E5650 2.66GHz = $1,998 (assuming $999 per CPU)
MoBo: eVGA SR-2 $600
Total Cost (not inclusive of RAM, HDD,...) =$2,598
12x 3.8 x 2 = 91.2GHz
* 3.8 is a somewhat conservative estimate, but you can hit 4.0 with the multiplier
* even if you don't get the full 91.2GHz, you'll get most of it for obviously a lot less money than you'll pay for a slightly better AMD Magny-Cours system.

I could get almost two SR-2 w/2 x E5650's for the price of one 4 x Opteron 6172 Magny-Cours.

Going the Intel route, you should be able to hit 150K PpD. The Magny-Cours may be able to hit 160 to 170K PpD.

 
how are you getting that many ppd?

I have a dual 5450 thats getting less than 3kppd?
 
Dual E5450's @ 3.0GHz

3.0GHz x 4 Cores (no Hyperthreading) x 2 CPU = 24GHz of total computing power.

@ 3.0GHz, it is kinda equivalent to a Core i7 920 @ 3.0 which won't make the deadlines for Big Adv, but should still be good for a lot more than 3K PpD.
 
capreppy said:
Dual E5450's @ 3.0GHz

3.0GHz x 4 Cores (no Hyperthreading) x 2 CPU = 24GHz of total computing power.

@ 3.0GHz, it is kinda equivalent to a Core i7 920 @ 3.0 which won't make the deadlines for Big Adv, but should still be good for a lot more than 3K PpD.
Click to expand...

so dual quad core 3ghz xeons wont make deadlines?

I think I'm starting to remember why I left
 
FLECOM said:
so dual quad core 3ghz xeons wont make deadlines?

I think I'm starting to remember why I left
Click to expand...

Big adv where built for clusters orig. A couple years later a i7 that is OC-ed over 3.2Ghz or above can do about the same amount of work.

You should be getting 12k ppd out of that rig running SMP A3 cores.

Are you using HFM.net or FAHSpy to monitor your clients?

FAHMon does not know how to calculate the bonus. (on a side note, you get the bonus after 10 WU are turned in using your passkey)
 
FLECOM said:
so dual quad core 3ghz xeons wont make deadlines?

I think I'm starting to remember why I left
Click to expand...

How would such a setup not be able to make the -bigadv deadlines? After all, that is a total of 8 actual cores versus a Core i7 which has only 4 actual cores.

Can't a stock Core i7 930 make the 4 day bigadv deadline @ 2.8ghz? I guess i'm a bit confused.
 
amdgamer said:
True, but that is a dual Magny Cours system with 2 12 core processors. I'm very curious to see how 4 of those 12 core processors would perform with a total of 48 cores.
Click to expand...
By the same token, you may as well just put together a quad Beckton system.
FLECOM said:
how are you getting that many ppd?

I have a dual 5450 thats getting less than 3kppd?
Click to expand...
Follow the guide: http://hardforum.com/showthread.php?t=1511766
FLECOM said:
dunno?

shortcut has

" -smp -bigadv -advmethods"

FAHSpy sais the core is GRO-A3 so I think so?
Click to expand...
You can't run -bigadv in Windows. You need to use a Linux VM or run native Linux. -bigadv workunits use the A2 core, which only works with the Linux SMP client. The A3 core is not -bigadv.
 
I did follow that guide
 
amdgamer said:
How would such a setup not be able to make the -bigadv deadlines? After all, that is a total of 8 actual cores versus a Core i7 which has only 4 actual cores.

Can't a stock Core i7 930 make the 4 day bigadv deadline @ 2.8ghz? I guess i'm a bit confused.
Click to expand...

Maybe, I'm not 100% sure what the cut off is (since I've never run a rig close to it)

One of the issues is the newer generation chips are just that much faster when you add HT into the mix and improvements in the chip.

I've got a 16 thread box running -bigadv and I get 29 min TPF. The killer is the speed. a 2.6 to 3.2 change is almost a 25% speed bump.

What capreppy is talking about is ramping up those chips to 3.8 and with the K factor the ppd you will get is HUGE!
 
FLECOM said:
I did follow that guide
Click to expand...
Then you should be getting more than 3000PPD. Either Fahspy is not calculating the bonus correctly for the unit you're working on, or your performance is much lower than it should be. Can you post an excerpt from your log file including the workunit you're currently folding and some frames?
 
this is a dell poweredge, cant overclock it
 
--- Opening Log file [April 22 22:15:57 UTC]


# Windows SMP Console Edition #################################################
###############################################################################

Folding@Home Client Version 6.29

http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: C:\Program Files\Folding@Home Windows SMP Client V1.01
Executable: C:\Program Files\Folding@Home Windows SMP Client V1.01\Folding@home-Win32-x86.exe
Arguments: -smp -bigadv -advmethods

[22:15:57] Configuring Folding@Home...


[22:16:48] - Ask before connecting: No
[22:16:48] - User name: FLECOM (Team 33)
[22:16:48] - User ID not found locally
[22:16:48] + Requesting User ID from server
[22:16:51] - Machine ID: 1
[22:16:51]
[22:16:51] Work directory not found. Creating...
[22:16:51] Could not open work queue, generating new queue...
[22:16:51] - Preparing to get new work unit...
[22:16:51] Cleaning up work directory
[22:16:51] + Attempting to get work packet
[22:16:51] Passkey found
[22:16:51] - Connecting to assignment server
[22:16:51] - Successful: assigned to (130.237.232.140).
[22:16:51] + News From Folding@Home: Welcome to Folding@Home
[22:16:51] Loaded queue successfully.
[22:16:55] + Closed connections
[22:16:55]
[22:16:55] + Processing work unit
[22:16:55] Core required: FahCore_a3.exe
[22:16:56] Core not found.
[22:16:56] - Core is not present or corrupted.
[22:16:56] - Attempting to download new core...
[22:16:56] + Downloading new core: FahCore_a3.exe
[22:16:58] + 10240 bytes downloaded
[22:17:01] + 2340489 bytes downloaded
[22:17:01] Verifying core Core_a3.fah...
[22:17:01] Signature is VALID
[22:17:01]
[22:17:01] Trying to unzip core FahCore_a3.exe
[22:17:02] Decompressed FahCore_a3.exe (8087552 bytes) successfully
[22:17:07] + Core successfully engaged
[22:17:12]
[22:17:12] + Processing work unit
[22:17:12] Core required: FahCore_a3.exe
[22:17:12] Core found.
[22:17:12] Working on queue slot 01 [April 22 22:17:12 UTC]
[22:17:12] + Working ...
[22:17:12]
[22:17:12] *------------------------------*
[22:17:12] Folding@Home Gromacs SMP Core
[22:17:12] Version 2.17 (Mar 12, 2010)
[22:17:12]
[22:17:12] Preparing to commence simulation
[22:17:12] - Looking at optimizations...
[22:17:12] - Created dyn
[22:17:12] - Files status OK
[22:17:13] - Expanded 1796866 -> 2078149 (decompressed 115.6 percent)
[22:17:13] Called DecompressByteArray: compressed_data_size=1796866 data_size=2078149, decompressed_data_size=2078149 diff=0
[22:17:13] - Digital signature verified
[22:17:13]
[22:17:13] Project: 6012 (Run 1, Clone 74, Gen 54)
[22:17:13]
[22:17:13] Assembly optimizations on if available.
[22:17:13] Entering M.D.
[22:17:19] Completed 0 out of 500000 steps (0%)
[22:20:23] Completed 5000 out of 500000 steps (1%)
[22:23:30] Completed 10000 out of 500000 steps (2%)
[22:26:31] Completed 15000 out of 500000 steps (3%)
[22:29:32] Completed 20000 out of 500000 steps (4%)
[22:32:33] Completed 25000 out of 500000 steps (5%)
[22:35:34] Completed 30000 out of 500000 steps (6%)
[22:38:37] Completed 35000 out of 500000 steps (7%)
[22:41:38] Completed 40000 out of 500000 steps (8%)
[22:44:40] Completed 45000 out of 500000 steps (9%)
[22:47:41] Completed 50000 out of 500000 steps (10%)
[22:50:43] Completed 55000 out of 500000 steps (11%)
[22:53:43] Completed 60000 out of 500000 steps (12%)
[22:56:44] Completed 65000 out of 500000 steps (13%)
[22:59:45] Completed 70000 out of 500000 steps (14%)
[23:02:46] Completed 75000 out of 500000 steps (15%)
[23:05:48] Completed 80000 out of 500000 steps (16%)
[23:08:49] Completed 85000 out of 500000 steps (17%)
[23:11:51] Completed 90000 out of 500000 steps (18%)
[23:14:52] Completed 95000 out of 500000 steps (19%)
[23:17:54] Completed 100000 out of 500000 steps (20%)
[23:20:56] Completed 105000 out of 500000 steps (21%)
[23:23:57] Completed 110000 out of 500000 steps (22%)
[23:26:59] Completed 115000 out of 500000 steps (23%)
[23:30:00] Completed 120000 out of 500000 steps (24%)
[23:33:02] Completed 125000 out of 500000 steps (25%)
[23:36:04] Completed 130000 out of 500000 steps (26%)
[23:39:07] Completed 135000 out of 500000 steps (27%)
[23:42:08] Completed 140000 out of 500000 steps (28%)
[23:45:11] Completed 145000 out of 500000 steps (29%)
[23:48:13] Completed 150000 out of 500000 steps (30%)
[23:51:15] Completed 155000 out of 500000 steps (31%)
[23:54:17] Completed 160000 out of 500000 steps (32%)
[23:57:19] Completed 165000 out of 500000 steps (33%)
[00:00:20] Completed 170000 out of 500000 steps (34%)
[00:03:22] Completed 175000 out of 500000 steps (35%)
[00:06:23] Completed 180000 out of 500000 steps (36%)
[00:09:25] Completed 185000 out of 500000 steps (37%)
[00:12:26] Completed 190000 out of 500000 steps (38%)
[00:15:27] Completed 195000 out of 500000 steps (39%)
[00:18:29] Completed 200000 out of 500000 steps (40%)
[00:21:30] Completed 205000 out of 500000 steps (41%)
[00:24:32] Completed 210000 out of 500000 steps (42%)
[00:27:34] Completed 215000 out of 500000 steps (43%)
[00:30:35] Completed 220000 out of 500000 steps (44%)
[00:33:37] Completed 225000 out of 500000 steps (45%)
[00:36:39] Completed 230000 out of 500000 steps (46%)
[00:39:41] Completed 235000 out of 500000 steps (47%)
[00:42:43] Completed 240000 out of 500000 steps (48%)
[00:45:45] Completed 245000 out of 500000 steps (49%)
[00:48:47] Completed 250000 out of 500000 steps (50%)
[00:51:49] Completed 255000 out of 500000 steps (51%)
[00:54:50] Completed 260000 out of 500000 steps (52%)
[00:57:52] Completed 265000 out of 500000 steps (53%)
[01:00:53] Completed 270000 out of 500000 steps (54%)
[01:03:55] Completed 275000 out of 500000 steps (55%)
[01:06:57] Completed 280000 out of 500000 steps (56%)
[01:09:59] Completed 285000 out of 500000 steps (57%)
[01:13:01] Completed 290000 out of 500000 steps (58%)
[01:16:03] Completed 295000 out of 500000 steps (59%)
[01:19:05] Completed 300000 out of 500000 steps (60%)
[01:22:06] Completed 305000 out of 500000 steps (61%)
[01:25:08] Completed 310000 out of 500000 steps (62%)
[01:28:09] Completed 315000 out of 500000 steps (63%)
[01:31:12] Completed 320000 out of 500000 steps (64%)
[01:34:13] Completed 325000 out of 500000 steps (65%)
[01:37:15] Completed 330000 out of 500000 steps (66%)
[01:42:17] Completed 335000 out of 500000 steps (67%)
[01:45:20] Completed 340000 out of 500000 steps (68%)
[01:48:23] Completed 345000 out of 500000 steps (69%)
[01:51:26] Completed 350000 out of 500000 steps (70%)
[01:54:28] Completed 355000 out of 500000 steps (71%)
 
TPF look spot on, it's an A3 too.

If you got a passkey entered you should get bonus after 10 WU.

FAHSpy is not calculating the bonus correctly.
 
What Kendrak said. You're doing ~3 minute frame times, which is very good and should give you about 16-17k PPD according to the online bonus calculator.
 
ya passkey is entered, guess I will wait and see then, FAHSpy is telling me 2230ppd is what its doing :confused:
 
If I use this calc: http://www.linuxforge.net/bonuscalc2.php
I get over 16k ppd. That tells me you should be able to run -bigadv on this if you can get a VM on it and feed it 5GB of RAM min.

Otherwise, 16k ppd is a good amount of points from a single box.

fleappd.jpg
 
pardon the possibly stupid question but why does it have to be in a VM?
 
FLECOM said:
cores are at 100%

i just started folding again a few hours ago, i have no idea what machines those other stats were
Click to expand...

Well i'm glad that you are back :)

These guys got me totally addicted in just the 2 weeks i've been folding.
 
FLECOM said:
pardon the possibly stupid question but why does it have to be in a VM?
Click to expand...

-bigadv (at this moment) runs on the A2 core. That is a linux only core.

I'm making the assumption that your running windows server. If you running linux native, have at it.
 
FLECOM said:
pardon the possibly stupid question but why does it have to be in a VM?
Click to expand...

From what I understand, you can only get assigned the -bigadv A2 work units using the LInux client. Unless you are running Linux natively, you have to run a virtual machine with a Linux image on it otherwise as the windows SMP client does not do those work units. I think the closest the windows SMP comes are the 6040 and 6041 WU's.
 
ah ok, ya its a server 2008 enterprise machine

need to get moar ram for it :(
 
FLECOM said:
ah ok, ya its a server 2008 enterprise machine

need to get moar ram for it :(
Click to expand...

You need 5GB is your running 8 threads. 7 is enough for 16.

(I'm putting in a bit of a buffer on those numbers, some people get away with a stripped down native linux install with 4GB for 8 threads, but that is on the razors edge. I just don't want to do 2 days worth of work and bork the WU for not having enough RAM)
 
amdgamer said:
From what I understand, you can only get assigned the -bigadv A2 work units using the LInux client. Unless you are running Linux natively, you have to run a virtual machine with a Linux image on it otherwise as the windows SMP client does not do those work units. I think the closest the windows SMP comes are the 6040 and 6041 WU's.
Click to expand...

your starting to sound like a old pro
 
FLECOM said:
Anyone tried out any of the new Opteron CPU's by any chance?

I was thinking one computer with 48 cpu cores would get some sick points... thoughts?
Click to expand...

No, but 48 cores of "old" istanbul chips at 2.00ghz is generating 160k PPD from a single box and is the current highest PPD box in F@H.
Max per box at the moment on just CPU folding i reckon is about 200k using 8439SE opterons - however the cost is stupid.
 
Nathan_P said:
Yes it can - even an 920 at stock can make the deadlines.
Click to expand...

Yep, my work pc will make the deadline....a Dell Vostro with an i7 860. The break even point for me is around 12-13k ppd. That makes it so the minimum TPF I need is around 52 for bigadv. If it's slower, then I'm better off with A3 and the Windows client.

I won't be making the max bonus anytime soon, but so far it seems better than what I can get with A3.
 
Nathan_P said:
No, but 48 cores of "old" istanbul chips at 2.00ghz is generating 160k PPD from a single box and is the current highest PPD box in F@H.
Max per box at the moment on just CPU folding i reckon is about 200k using 8439SE opterons - however the cost is stupid.
Click to expand...

hehe, im just thinking about the highest possible PPD box if cost is no object
 
FLECOM said:
hehe, im just thinking about the highest possible PPD box if cost is no object
Click to expand...
I honestly think you should focus on building an i7 based farm with minimal GPU clients at this time. You'll probably get more production accomplished that way, unless you already have many systems waiting with empty PCI-E slots. Also, power usage will be a lot less. Note that I mentioned 'at this time.' Who knows if there will be another paradigm shift at Stanford changing the whole scenario in the not too distant future...?

nitrobass24 said:
Beckton :)
Click to expand...
One would think so but the numbers posted recently by someone claiming to have a Beckton system with four processors left much to be desired with its performance numbers. Mind you, that was with a single client. Running multiple clients will probably be a superior setup to reduce scaling problems but this all speculative until someone tries it. So far, I aint impressed with the PPD per dollar.
 
APOLLO said:
I honestly think you should focus on building an i7 based farm with minimal GPU clients at this time. You'll probably get more production accomplished that way, unless you already have many systems waiting with empty PCI-E slots. Also, power usage will be a lot less.
Click to expand...

This is soo very true.

Nexus only pulls 200 watts from the wall and does 30k ppd.

When -bigadv comes to A3 I'm expecting a bump to 35k or more.

Now if SR-2 ever shows up, I wont take anything less then 50-60k ppd on ~300 watts.
 
You must log in or register to reply here.
Back
Top