Windows BIGADV is Here

38 min TPF on Skynet @ 3.33ghz, thats only 25k, that's very depressing
 
hmmm...
-smp 16 -bigadv -verbosity 9

only one frame done right now but I show 38 TPF

2x E5530s @ 3.33Ghz (185x18)

EDIT
OK, so I just finished my 2%, now I'm showing 35min TPFs or 29k ppd, thats a little better, once 7 rolls around I'll be getting my i7 back on this

EDIT x2
OK, so looking at the log file I'm pulling an exact 30:01 TPF between the last two frames, why does HFM report it otherwise?
 
Vaulter98c said:
hmmm...
-smp 16 -bigadv -verbosity 9

only one frame done right now but I show 38 TPF

2x E5530s @ 3.33Ghz (185x18)

EDIT
OK, so I just finished my 2%, now I'm showing 35min TPFs or 29k ppd, thats a little better, once 7 rolls around I'll be getting my i7 back on this

EDIT x2
OK, so looking at the log file I'm pulling an exact 30:01 TPF between the last two frames, why does HFM report it otherwise?
Click to expand...

HFM.net does not give a true TPF till 3 frames have passed.
 
Kendrak said:
HFM.net does not give a true TPF till 3 frames have passed.
Click to expand...
Depends on what your preferences are set for, but I think LastThreeFrames is the default. You can set it to "LastFrame", "AllFrames", or "EffectiveRate" as other choices.
 
gkanakis said:
52-54:ish range. My work computer is at 3.3GHz, but I forgot the number.
Click to expand...

Good deal, that jives with the 3 frames I've seen it finish so far as I'm sitting right about at 53 minutes per frame. Problem is, this is my main gaming rig too...time to get a dedicated boxen if I'm going to get serious about bigadv it looks like:)
 
Vaulter98c said:
Just grabbed one on skynet :)
Altho I will say this, is there utilization issues with this?? with A3 SMP's I had a steady power draw, maybe a 2 watt vary in each direction. Now, my draw is jumping around between 380 and 417 (417 where I was with A3's) but task manager shows me at constant 100% CPU use. ideas??
Click to expand...


usually fluctuates when its checkpointing.. at least thats what ive noticed.. especially if your running your ram underclocked or just plain slow ram..
 
mjannusch said:
Depends on what your preferences are set for, but I think LastThreeFrames is the default. You can set it to "LastFrame", "AllFrames", or "EffectiveRate" as other choices.
Click to expand...


^^ pretty much what he said vaulter... HFM by default only shows the last 3 frames.. so its not as accurate..
 
Also, does running -smp 15 instead of -smp 16


have the same effect as AMD systems of boosting PPD? (something about a free thread?)?
 
Ok, my Gulftown is currently clocked at 4.33 (PSU can't handle all the OC) and I'm getting 30:26 per frame for 36.6 PPD.
 
So, how come there are A3s units that out score a bigadv? They need to fix the scoring on these, why waste 3 days on a unit you may not even finish to get less points than a few hour unit that doesn't crash? This needs fixed now, I could understand a bit when they nixxed bigadv but if they nix it to the point where it's not worth more than a normal unit IF it finishes
Posted via [H] Mobile Device
 
So a 12 core rig at 4.3 gets the same or slighlty less than a 16 core at 3.3

atleast it was a good investment
Posted via [H] Mobile Device
 
Anyone else have memory usage numbers for these?
 
16 cores, like a hair below 1.3 gigs
Posted via [H] Mobile Device
 
Anyone else seeing higher values on standard a3s?
Posted via [H] Mobile Device
 
Ok, times are updated, 29:25 TPF, just a scratch below 40k ppd
Posted via [H] Mobile Device
 
PPD is fluctuating. Between 33.5K and 35K. I think it is the 480s causing it.
 
I've ctrl-c'd the client twice and restarted my computer mid unit (ok, eleet crashed it) without losing the progress. I haven't done 10 smp units for the bonus credit so there isn't much point to continue on this one.

[00:37:26] Loaded queue successfully.
[00:37:26]
[00:37:26] + Processing work unit
[00:37:26] - Autosending finished units... [June 30 00:37:26 UTC]
[00:37:26] Trying to send all finished work units
[00:37:26] + No unsent completed units remaining.
[00:37:26] - Autosend completed
[00:37:26] Core required: FahCore_a3.exe
[00:37:26] Core found.
[00:37:26] Working on queue slot 04 [June 30 00:37:26 UTC]
[00:37:26] + Working ...
[00:37:26] - Calling '.\FahCore_a3.exe -dir work/ -nice 19 -suffix 04 -np 8 -checkpoint 5 -verbose -lifeline 5024 -version 629'

[00:37:26]
[00:37:26] *------------------------------*
[00:37:26] Folding@Home Gromacs SMP Core
[00:37:26] Version 2.22 (Mar 12, 2010)
[00:37:26]
[00:37:26] Preparing to commence simulation
[00:37:26] - Ensuring status. Please wait.
[00:37:35] - Looking at optimizations...
[00:37:35] - Working with standard loops on this execution.
[00:37:35] - Previous termination of core was improper.
[00:37:35] - Going to use standard loops.
[00:37:35] - Files status OK
[00:37:39] - Expanded 20082381 -> 30791309 (decompressed 153.3 percent)
[00:37:39] Called DecompressByteArray: compressed_data_size=20082381 data_size=30791309, decompressed_data_size=30791309 diff=0
[00:37:39] - Digital signature verified
[00:37:39]
[00:37:39] Project: 2684 (Run 2, Clone 23, Gen 0)
[00:37:39]
[00:37:40] Entering M.D.
[00:37:46] Using Gromacs checkpoints
[00:37:51] Resuming from checkpoint
[00:37:52] Verified work/wudata_04.log
[00:37:53] Verified work/wudata_04.trr
[00:37:53] Verified work/wudata_04.xtc
[00:37:53] Verified work/wudata_04.edr
[00:37:54] Completed 4740 out of 250000 steps (1%)
 
29:00

anyone found stability issues yet? It seems the system is more responsive when using the rest of it while bigadv'ing than it dis under a Linux VM

anyone seeing issues yet?
Posted via [H] Mobile Device
 
Ok, I am getting TPF's of about 51 minutes. I am running a 3.46ghz overclock, so what Evil said a while back of 3.4ghz being the limit is definately still true for Windows SMP bigadv.
 
Vaulter98c said:
29:00

anyone found stability issues yet? It seems the system is more responsive when using the rest of it while bigadv'ing than it dis under a Linux VM

anyone seeing issues yet?
Posted via [H] Mobile Device
Click to expand...

No issues.

I turned off Hugh and ripped 2x 2gb sticks out of him and put them in Nexus. Hugh is folding win -bigadv with 2Gb of memory. And it's still pushing 20k :cool: No problem on the start up and resume.

Nexus is doing just about 39k while @ 175x19 with 12GB RAM (total overkill but it get trip channel)

I'm thinking for any future rigs we are only going to need 1GB sticks
 
Vaulter98c said:
So, how come there are A3s units that out score a bigadv? They need to fix the scoring on these, why waste 3 days on a unit you may not even finish to get less points than a few hour unit that doesn't crash?
Click to expand...
Agreed, and that essentially has been the controversy surrounding the P2684 from the get go, hence the giant thread devoted to it and all the issues people have been experiencing. There has occasionally been large point discrepancies between newly released, bigger WUs over older series on the same clients. I don't know why this is the case. I've seen it enough times to realize there must be something odd about the way Stanford assigns credit to WUs, something that is not mentioned by them in all the threads where this topic comes up.

This needs fixed now, I could understand a bit when they nixxed bigadv but if they nix it to the point where it's not worth more than a normal unit IF it finishes
Click to expand...
You guys are braver than I am, but I would suggest that one or two crashes at reasonable clocks and I would not run them until Stanford gets the issues resolved. Hopefully, the P2684s are more stable in Windows since A3 seems to be tailored for it.

Vaulter98c said:
So a 12 core rig at 4.3 gets the same or slighlty less than a 16 core at 3.3
Click to expand...
Yeah, that looks to be about right. :cool:

DooKey said:
PPD is fluctuating. Between 33.5K and 35K. I think it is the 480s causing it.
Click to expand...
Quite likely especially if you're running the GPU client.
 
Mine crashed at 2% on a 3.9ghz i860 that's been running VMware bigadv since January without a single crash. Hmmm. Guess I can try upping the volts a bit to see what happens.
 
OK, anyone who is dissatisfied with the performance and has tons of cores and RAM available, maybe try multiple clients to see if there's an overall improvement? I know Stanford frowns on this but it's just for testing purposes and I only suggest it for those who have the big iron to try it out.
 
APOLLO said:
OK, anyone who is dissatisfied with the performance and has tons of cores and RAM available, maybe try multiple clients to see if there's an overall improvement? I know Stanford frowns on this but it's just for testing purposes and I only suggest it for those who have the big iron to try it out.
Click to expand...

Here is what I'm thinking about that......

I'm running 3.3GHz on 16 threads. Now 3.3 on it's own will just maybe hit the deadline for what 14k ppd (less?)
The system now is showing 38700ppd.

Now I'm going to think that running 2x client will slow things down jsut a bit for the both of 'em so they might not even hit the bonus.

Also they way the bonus is set up, running multi clients like we did back in the day with A1 cores just wont get ya the most ppd anymore. You need as much power as you can get in one client with one bonus structure.
 
Kendrak said:
Also they way the bonus is set up, running multi clients like we did back in the day with A1 cores just wont get ya the most ppd anymore. You need as much power as you can get in one client with one bonus structure.
Click to expand...
I agree that's what the situation is on paper. We don't know how efficient the client or WU is in actual practice. I realize these new units run much slower than the A2 predecessors. It's something only those who have the horsepower to attempt without risk losing the bonus credit. It was only a suggestion.
 
Vaulter98c said:
So a 12 core rig at 4.3 gets the same or slighlty less than a 16 core at 3.3

atleast it was a good investment
Posted via [H] Mobile Device
Click to expand...

Makes sense...

12 * 4.3 = 51.6
16 * 3.3 = 52.8

Gotta push them quads higher!
 
When someone completes one of these, can you post the amount of data that was uploaded to Stanford? I can crank out a bunch of these, but I am limited on how much total data I can upload, and my upload speed is "unimpressive".
 
Project ID: 2684
Core: GRO-A3
Credit: 8955
Frames: 100


Name: BigADV - CC
Path: \\fah\fah\
Number of Frames Observed: 100

Min. Time / Frame : 00:42:24 - 22,306.8 PPD
Avg. Time / Frame : 00:44:13 - 20,946.3 PPD


Name: Daviserver
Path: \\daviserver.davis.local\folding\smp2
Number of Frames Observed: 290

Min. Time / Frame : 00:43:11 - 21,702.6 PPD
Avg. Time / Frame : 00:48:37 - 18,168.1 PPD


Name: minerva
Path: \\minerva\folding\
Number of Frames Observed: 206

Min. Time / Frame : 00:37:52 - 26,430.3 PPD
Avg. Time / Frame : 00:42:05 - 22,559.1 PPD
 
mjannusch said:
When someone completes one of these, can you post the amount of data that was uploaded to Stanford? I can crank out a bunch of these, but I am limited on how much total data I can upload, and my upload speed is "unimpressive".
Click to expand...

Bigadv work units typically end up being 94-101 meg uploads. On my 384kbps upload speeds, it literally takes forever.
 
Scaling has to be pretty decent based on our 16 core numbers and 8 cores at the same speed

hell were even kinda close on the 12 to 16 core scale
Posted via [H] Mobile Device
 
I7 @ 3.78 (21 x 180), memory at 1080 triple channel - 46.30 tpf, a little over 19K ppd.

Compared to my one Linux box still running one of these:

I7 @ 3.8 (20 * 190), memory at 1520 triple channel - 46.50 tpf, a little less over 19K ppd.

There is defintely a performance gain running these in Windows, and memory speed appears to be less important (which I also saw running these in Linux.)

It is going to take someone smarter than me to figure out how to run the Windows SMP client under Wine in Linux. I got close, but couldn't get the MPICH stuff (install.bat) working.
 
You must log in or register to reply here.
Back
Top