G34 PPD?

FLECOM

FLECOM

Modder(ator) & [H]ardest Folder Evar
Staff member
Joined
Jun 27, 2001
Messages
15,814
So I setup a 4P G34 with 4x 6128 and 32gb ram, on an 8101 I am getting about 150kppd, does that sound right? I used the install script so its running kraken ramdisk etc etc

if so why does anyone bother with this? you can get one of those C6100's with the 4x dual L5520's for less money and the 4 SMP instances will put out >110kppd
 
What MB do you have? Have you done all the optimization? 8-core 6128 are the complete bottom of the barrel, so do not expect too much. With higher end chips your PPD increase dramatically.
 
Hopefully you got one of the OC capable SM boards. If so, you'll get more out of it than that.
 
The 4P 8439's run faster than that, and that's 24 cores x 2.8ghz with only DDR2 ram.

Your PPD doesn't seem correct. Restart the job and see if it changes.

Doh! Edit.
 
My 6128 4p @ 2.47GHz just finished an 8101 for 247K with a TPF of ~18min. That's for a little over 30hr of folding.

For S&G I tried an 8104 setup for testing on my 6128 2p @ stock over the weekend, it averaged a TPF of 24min. There for it can finish an 8104 before the preferred deadline but there's no guarantee that that's all it will get so it's doing vanilla SMP.

The only guarantee we get in folding is that it will be "soon" ;)
 
Flecom, you are using octo core cpus.

#33 ballers do not.
 
FLECOM said:
....if so why does anyone bother with this?.....
Click to expand...
It's like buying the Ferrari with a VW engine installed until you can upgrade. :eek:
That other system you mentioned most likely tops out at that PPD and can't be upgraded much.
The G34 with 6128s, not overcolcked and churning an 8101 WU is about the lowest PPD you can expect with that system.
Get the OC going when when you can and look at upping to some 6166HEs if possible.
And by all means [H]ang in there!
 
FLECOM said:
So I setup a 4P G34 with 4x 6128 and 32gb ram, on an 8101 I am getting about 150kppd, does that sound right? I used the install script so its running kraken ramdisk etc etc

if so why does anyone bother with this? you can get one of those C6100's with the 4x dual L5520's for less money and the 4 SMP instances will put out >110kppd
Click to expand...

150K is believable for 32 cores at stock 2GHz.

1. 150K ppd is better than 110K ppd
2. The G34 4p is probably pulling less total power (although the C6100 is pretty efficient as well)
3. 8101s are by far the worst bigadv units - all of the others will be roughly 1/3 more ppd
4. SM boards are overclockable
5. You have upgrade options - more that you will with that C6100
 
bigted said:
how many frames has it ran?
Click to expand...

[11:42:24] + Processing work unit
[11:42:24] Core required: FahCore_a5.exe
[11:42:24] Core found.
[11:42:24] Working on queue slot 02 [April 15 11:42:24 UTC]
[11:42:24] + Working ...
[11:42:24] - Calling './FahCore_a5.exe -dir work/ -nice 19 -suffix 02 -np 32 -checkpoint 15 -verbose -lifeline 909 -version 634'

[11:42:24]
[11:42:24] *------------------------------*
[11:42:24] Folding@Home Gromacs SMP Core
[11:42:24] Version 2.27 (Thu Feb 10 09:46:40 PST 2011)
[11:42:24]
[11:42:24] Preparing to commence simulation
[11:42:24] - Looking at optimizations...
[11:42:24] - Created dyn
[11:42:24] - Files status OK
[11:42:29] - Expanded 30315831 -> 33158020 (decompressed 109.3 percent)
[11:42:29] Called DecompressByteArray: compressed_data_size=30315831 data_size=33158020, decompressed_data_size=33158020 diff=0
[11:42:29] - Digital signature verified
[11:42:29]
[11:42:29] Project: 8101 (Run 10, Clone 3, Gen 223)
[11:42:29]
[11:42:29] Assembly optimizations on if available.
[11:42:29] Entering M.D.
[11:42:37] Mapping NT from 32 to 32
[11:42:45] Completed 0 out of 250000 steps (0%)
[12:06:24] Completed 2500 out of 250000 steps (1%)
[12:29:35] Completed 5000 out of 250000 steps (2%)
[12:46:28] - Autosending finished units... [April 15 12:46:28 UTC]
[12:46:28] Trying to send all finished work units
[12:46:28] + No unsent completed units remaining.
[12:46:28] - Autosend completed
[12:52:41] Completed 7500 out of 250000 steps (3%)
[13:15:52] Completed 10000 out of 250000 steps (4%)
[13:38:59] Completed 12500 out of 250000 steps (5%)
[14:02:10] Completed 15000 out of 250000 steps (6%)
[14:25:18] Completed 17500 out of 250000 steps (7%)
[14:48:30] Completed 20000 out of 250000 steps (8%)
[15:11:41] Completed 22500 out of 250000 steps (9%)
[15:34:55] Completed 25000 out of 250000 steps (10%)
[15:58:08] Completed 27500 out of 250000 steps (11%)
[16:21:19] Completed 30000 out of 250000 steps (12%)
[16:44:34] Completed 32500 out of 250000 steps (13%)
[17:07:45] Completed 35000 out of 250000 steps (14%)
[17:30:59] Completed 37500 out of 250000 steps (15%)
[17:54:12] Completed 40000 out of 250000 steps (16%)
[18:17:26] Completed 42500 out of 250000 steps (17%)
[18:40:36] Completed 45000 out of 250000 steps (18%)
[18:46:28] - Autosending finished units... [April 15 18:46:28 UTC]
[18:46:28] Trying to send all finished work units
[18:46:28] + No unsent completed units remaining.
[18:46:28] - Autosend completed
[19:03:50] Completed 47500 out of 250000 steps (19%)
[19:27:01] Completed 50000 out of 250000 steps (20%)
[19:50:16] Completed 52500 out of 250000 steps (21%)
[20:13:32] Completed 55000 out of 250000 steps (22%)
[20:36:44] Completed 57500 out of 250000 steps (23%)
[20:59:59] Completed 60000 out of 250000 steps (24%)
[21:23:12] Completed 62500 out of 250000 steps (25%)
[21:46:28] Completed 65000 out of 250000 steps (26%)
[22:09:40] Completed 67500 out of 250000 steps (27%)
[22:32:55] Completed 70000 out of 250000 steps (28%)
[22:56:06] Completed 72500 out of 250000 steps (29%)
[23:19:20] Completed 75000 out of 250000 steps (30%)
[23:42:32] Completed 77500 out of 250000 steps (31%)
[00:05:47] Completed 80000 out of 250000 steps (32%)
[00:29:01] Completed 82500 out of 250000 steps (33%)
[00:46:28] - Autosending finished units... [April 16 00:46:28 UTC]
[00:46:28] Trying to send all finished work units
[00:46:28] + No unsent completed units remaining.
[00:46:28] - Autosend completed
[00:52:15] Completed 85000 out of 250000 steps (34%)

sc0tty8 said:
Flecom, you are using octo core cpus.

#33 ballers do not.
Click to expand...

I am not a baller

Core32 said:
It's like buying the Ferrari with a VW engine installed until you can upgrade. :eek:
That other system you mentioned most likely tops out at that PPD and can't be upgraded much.
The G34 with 6128s, not overcolcked and churning an 8101 WU is about the lowest PPD you can expect with that system.
Get the OC going when when you can and look at upping to some 6166HEs if possible.
And by all means [H]ang in there!
Click to expand...

guess I am going to have to look into overclocking, which sucks because then I have to install Ubuntu straight onto this thing bare metal instead of via Xen like my 6100's

if you have $1200 I can borrow indefinitely maybe I can get some 6166HE CPUs, until then this is the best I can do
 
6166HEs will be much less. More on the order of $500 - 600 for 4. But I don't have it to give either :)
 
I didn't see The Kraken running?

That will speed it up a bit.

Make sure the BIOS settings are right. There is one BIOS setting that will drop performance a bunch. Something about memories of intercourse or something.
 
thekraken will never show up in the log. You would need to look at the console for "NOTE: DYNAMIC LOAD BALANCING ENABLED".
 
Side observation -- if it's a dedicated folder I can't see any reason for using virtualization [?]

The easiest way to check for Kraken is running 'top'. DLB may engage w/o Kraken, too, so it's not a very good metric.

FLECOM, your TPFs look in-line with 4x6128 performance, next step is probably OC...
 
its in a rack in a datacenter, xen lets me get to it easily if linux takes a dump... plus I was planning on putting a windows VM so I could pause F@H and use it for transcoding

you cant OC from windows right? because I could just load windows 7 on it bare metal and let it run F@H and just use VidCoder when I need to
 
FLECOM said:
its in a rack in a datacenter, xen lets me get to it easily if linux takes a dump... plus I was planning on putting a windows VM so I could pause F@H and use it for transcoding
Click to expand...

Linux rarely takes a dump. When it does is caused by some gross user action. If that is the route you want to take, I would install linux natively and run windoze in VirtualBox. If you OC the things in linux( assuming you have a Supermicro MB ), the windoze instance will use and/or take advantage of the OC.
 
Flecom, you would need to boot from linux to oc and need to power off, then power back up to take effect. If it is in a data center, I would not recommend you OC it, especially remotely.

The big numbers come from these running 12/16 core cpu at 3+ ghz.
 
I don't see a problem w/ hosting OC'd machine in a datacenter. Couple guys (rhavern?) have done that.

The downside is that you'd need to spend some time in the DC (or temporarily pull the
machine out) to set the OC up.

One would (most likely) need to boot to native Linux (DVD/USB) configure the OC -- not sure if it can be
done via Xen guest.
Once OC is set up, boot OS doesn't matter but I probably wouldn't choose Windows as primary/host OS.
 
I should have been more clear, I would not recommending overclocking the machine while in DC. Perhaps my OCD, but I want to do my testing and tweaking hands on, not remotely.

Y U MAD BRO?
 
I may not have been clear. Brain doesn't always connect to hands....

No, not mad...u?
 
well I have 24/7 access to the DC so I can do it here its not a big deal...

what I don't get is the first WU it got it was doing >250kppd, now I keep getting these 8101's that do ~150kppd... should I turn off bigadv?
 
Unfortunately 8101 is undervalued.
8102, 8103 and 8104 BA projects don't suffer from this issue.

From what I can tell, 8101s do pop up every now and then; just need to grit your teeth...

I'd keep BA flag on -- SMP units will yield even less output.

20% of OC [which _should_ be possible with these chips] should put your 8101 output at 200k PPD
and 8102/8103/8104 output at 350k PPD -- much nicer, no? :D
 
I've gotten 3 in a row now :cry:
 
well I picked up the set of 4x 6166HE's that poped up in FS/FT so lets see how those do... don't think I'm going to bother OC'ing this box till I swap the chips
 
FLECOM said:
well I picked up the set of 4x 6166HE's that poped up in FS/FT so lets see how those do... don't think I'm going to bother OC'ing this box till I swap the chips
Click to expand...

For reference: 4x 6168's @ stock, 8101~330k PPD, 8102-8104~550k PPD.
 
FLECOM said:
... don't think I'm going to bother OC'ing this box till I swap the chips....
Click to expand...

The only benefit to doing it now would be for the knowledge gained and being ready out of the box with the 6166HE chips.
The exact OC setup will hold over to the new ones. You will just be able to enhance it with a vcore boost as well (HE chips allow that) and that will most likely let you OC much higher than the 6128s.
 
I know myself, so the whole thing is probably going to be frustrating and/or tragic... so I'll hold off a bit :)
 
FLECOM said:
I know myself, so the whole thing is probably going to be frustrating and/or tragic... so I'll hold off a bit :)
Click to expand...

When you are ready to proceed, catch one of us on irc - we'll walk you through the process. I know it seems a bit daunting, but it really isn't too bad.
 
Those new chips with OC should get you into the 300k-ish ppd game on 8101.
 
musky said:
When you are ready to proceed, catch one of us on irc - we'll walk you through the process. I know it seems a bit daunting, but it really isn't too bad.
Click to expand...

it hasn't been a great two weeks or so... need a mental health break
 
You must log in or register to reply here.
Back
Top