Core 2.20 for Fermi

If you think it might be a driver problem, see viewforum.php?f=79

Moderators: slegrand, Site Moderators, PandeGroup

Re: Reports on my first Project: 7621 (Run 445, Clone 0, Ge

Postby 7im » Fri Aug 19, 2011 4:57 pm

GreyWhiskers wrote:...

5. :shock: These WUs seemed to have a beneficial effect on the SMP wu going on. I had been getting ~ 30:50 TPF for the p6900s, running the "normal" p6802 - p6804 type GPU Wus in parallel. Now, I'm looking at 30:00 TPF for the p6900, still in -SMP 7.

I think the effect in #5 may be due to exceedingly low CPU usage by Core 15. After chewing on the p7621 for about 2.3 hours, windows Taskmon is showing only 8 seconds of CPU usage.



I don't think the new fahcore uses less CPU cycles. I think in the smaller WUs, the GPU took the WU data as fast as it could, using CPU time to keep feeding constant chunks of data to the GPU. In the larger WUs, the GPU takes a few cycles to chew the larger amounts of data, so the GPU isn't calling on the CPU to feed it with data as often, allowing more CPU cycles to flow to the SMP client. We've seen the same thing in previous generations of GPU clients. Same for the high resource usage and temps, etc. ;)
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: Core 2.20 for Fermi

Postby dschief » Fri Aug 19, 2011 5:51 pm

for what it's worth:
I have an XP-Pro SP3 box with twin Geforce GTX460SE's with 266.58 driver.

Picked up 7621 run 333 Clone 0 Gen 1 & 7620 run 69 Clone 0 gen1 this morning, each grabbed the 2.20 ver. with no trouble

both are chugging away with frame times in the vicinity of 9min 20 sec .all clocks are factory stock
User avatar
dschief
 
Posts: 246
Joined: Tue Dec 04, 2007 5:56 am

Re: Core 2.20 for Fermi

Postby tjlane » Fri Aug 19, 2011 7:12 pm

Update RE: error "CoreStatus = 63 (99)"

We do not believe this is an error due to any specific driver, and does not appear to be due to the card. I've gotten over 100 WUs back off of GTX 480s, so I don't think it's the hardware. @filu, do you get a log file generated in your FAH directory? Have you modified any driver/library settings for your card locally? Any more information you can give will help!

TJ
User avatar
tjlane
Pande Group Member
 
Posts: 175
Joined: Wed Jun 01, 2011 11:19 pm
Location: Stanford, CA

Re: Core 2.20 for Fermi

Postby 7im » Fri Aug 19, 2011 8:05 pm

CoreStatus = 63 (99) is the result of the fah client failing to start.

Either a missing file, missing file permission, etc. Some requirement of the client has not been met.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: Core 2.20 for Fermi

Postby filu » Fri Aug 19, 2011 9:15 pm

I changed nothing. This is what I get in the log have shown in an earlier post. With an earlier version of the core client works properly without any errors.

7im wrote:Either a missing file, missing file permission, etc. Some requirement of the client has not been met.

This is why the client is working properly with the core version 2.15?
Code: Select all
                       Folding@Home Client Version 6.41r2

                          http://folding.stanford.edu

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

Launch directory: C:\GPU
Service: C:\GPU\Folding@home-Win32-gpu.exe
Arguments: -svcstart -verbosity 9 -d C:\GPU

Launched as a service.
Entered C:\GPU to do work.

[07:05:39] - Ask before connecting: No
[07:05:39] - User name: filu (Team 276)
[07:05:39] - User ID: 1637D1286AE493C0
[07:05:39] - Machine ID: 2
[07:05:39]
[07:05:39] Gpu type=3 species=20.
[07:05:39] Work directory not found. Creating...
[07:05:39] Could not open work queue, generating new queue...
[07:05:39] - Preparing to get new work unit...
[07:05:39] Cleaning up work directory
[07:05:39] - Autosending finished units... [August 19 07:05:39 UTC]
[07:05:39] Trying to send all finished work units
[07:05:39] + No unsent completed units remaining.
[07:05:39] - Autosend completed
[07:05:39] + Attempting to get work packet
[07:05:39] - Will indicate memory of 511 MB
[07:05:39] Gpu type=3 species=20.
[07:05:39] - Detect CPU. Vendor: GenuineIntel, Family: 15, Model: 4, Stepping: 1
[07:05:39] - Connecting to assignment server
[07:05:39] Connecting to http://assign-GPU.stanford.edu:8080/
[07:05:40] Posted data.
[07:05:40] Initial: 40AB; - Successful: assigned to (171.64.65.64).
[07:05:40] + News From Folding@Home: Welcome to Folding@Home
[07:05:40] Loaded queue successfully.
[07:05:40] Gpu type=3 species=20.
[07:05:40] Empty passkey
[07:05:40] Connecting to http://171.64.65.64:8080/
[07:05:41] Posted data.
[07:05:41] Initial: 0000; - Receiving payload (expected size: 44901)
[07:05:42] - Downloaded at ~43 kB/s
[07:05:42] - Averaged speed for that direction ~43 kB/s
[07:05:42] + Received work.
[07:05:42] + Closed connections
[07:05:42]
[07:05:42] + Processing work unit
[07:05:42] Core required: FahCore_15.exe
[07:05:42] Core found.
[07:05:42] Working on queue slot 01 [August 19 07:05:42 UTC]
[07:05:42] + Working ...
[07:05:42] - Calling '.\FahCore_15.exe -dir work/ -suffix 01 -nice 19 -priority 96 -checkpoint 15 -service -verbose -lifeline 272 -version 641'

[07:05:42]
[07:05:42] *------------------------------*
[07:05:42] Folding@Home GPU Core
[07:05:42] Version 2.15 (Tue Nov 16 09:05:18 PST 2010)
[07:05:42]
[07:05:42] Build host: SimbiosNvdWin7
[07:05:42] Board Type: NVIDIA/CUDA
[07:05:42] Core      : x=15
[07:05:42]  Window's signal control handler registered.
[07:05:42] Preparing to commence simulation
[07:05:42] - Looking at optimizations...
[07:05:42] DeleteFrameFiles: successfully deleted file=work/wudata_01.ckp
[07:05:42] - Created dyn
[07:05:42] - Files status OK
[07:05:42] sizeof(CORE_PACKET_HDR) = 512 file=<>
[07:05:42] - Expanded 44389 -> 169787 (decompressed 382.4 percent)
[07:05:42] Called DecompressByteArray: compressed_data_size=44389 data_size=169787, decompressed_data_size=169787 diff=0
[07:05:42] - Digital signature verified
[07:05:42]
[07:05:42] Project: 6800 (Run 17932, Clone 0, Gen 317)
[07:05:42]
[07:05:42] Assembly optimizations on if available.
[07:05:42] Entering M.D.
[07:05:44] Tpr hash work/wudata_01.tpr:  4227011903 3897715976 4028041445 2073059638 1820643720
[07:05:44] Working on PEPTIDE (1-42)
[07:05:44] Client config found, loading data.
[07:05:44] Starting GUI Server
[07:05:46] Setting checkpoint frequency: 500000
[07:05:46] Setting checkpoint frequency: 500000
[07:07:00] Completed    500000 out of 50000000 steps (1%).
[07:08:13] Completed   1000000 out of 50000000 steps (2%).

Tomorrow I will try to set up a new OS.
Image
i7-2600K@4.8 Asus P8P67 EVO 2x2GB GTX480
i7-920@4.0 GA-EX58-UD5 3x2GB 2xGTX560Ti
2x Xeon 5620 6x 2GB
2x Xeon 5645 6x 2GB
filu
 
Posts: 76
Joined: Mon Aug 03, 2009 9:33 am
Location: Krzeszyce, Poland

Re: Core 2.20 for Fermi

Postby MtM » Fri Aug 19, 2011 9:53 pm

@filu Do you have an anti virus using a file hash ( like md5 or something ) which verifies which files are allowed? If your old core starts without problems, and the new one doesn't ( since you don't have a wudata log I'm assuming it's blocked from execution ) that seems the only plausible explanation?
MtM
 
Posts: 3054
Joined: Fri Jun 27, 2008 2:20 pm
Location: The Netherlands

Re: Core 2.20 for Fermi

Postby filu » Fri Aug 19, 2011 10:06 pm

MtM wrote:@filu Do you have an anti virus using a file hash ( like md5 or something ) which verifies which files are allowed?

On this computer I have no antivirus. This is a dedicated machine only to fold on the GTX480.
I turned off the system firewall, but sfc showed a lot of the changed system libraries, so there is nothing left for me to reinstall system.
Many thanks for your help. :)
filu
 
Posts: 76
Joined: Mon Aug 03, 2009 9:33 am
Location: Krzeszyce, Poland

Re: Core 2.20 for Fermi

Postby GreyWhiskers » Fri Aug 19, 2011 10:26 pm

quick update. I have almost 2.5 of the new WUs under my belt. With what I reported about the excellent coexistance with the SMP Core a5 WUs, I put the SMP flag back to SMP 8 (actually, no number). Lo and behold, the SMP performance got even better than it was with -SMP7, with the p762x running as the GPU WU. :arrow: The interference from these GPU WUs is very little, indeed.

SMP7 - P6900 SMP - 30:00 TFP - 29894 ppd - p7621 GPU - 5:00 TPF/14889 ppd - 44783 ppd total - 360-369 watts -- 121 ppd/watt

Stopped the GPU for a few frames of SMP only - reset to SMP 8
smp8 - P6900 SMP - 27:25 TFP - ~34,000 ppd -- 180-189 watts -- 180 ppd/watt

Added the GPU back - leaving SMP 8.
SMP8 - P6900 SMP - 28:14 TFP - ~32,600 ppd - p7621 GPU - 5:00 TPF/14889 ppd - 48489 ppd total - 360-369 watts -- 129 ppd/watt
Almost as good as the SMP with no GPU.

--> the p7621, while being very productive, consumes 180 watts for 14,889 ppd or 83 ppd/watt :cry:

I just got a note from our electrical utility, PG&E, today saying my electrical consumption was VERY HIGH compared to other comperable houses. Their #1 suggestion was to reveiew computer usage in light of parasitic power use in pugged-in but not in use computers. :!:

I'll continue to fold this way - but the power consumption for the GPUs with the new projects is really an eye-opener.
User avatar
GreyWhiskers
 
Posts: 767
Joined: Mon Oct 25, 2010 5:57 am
Location: Saratoga, California USA

Re: Core 2.20 for Fermi

Postby Dennis Ernst » Fri Aug 19, 2011 10:44 pm

2.20 fixed the problem I was having running Project: 6800 (Run 17935, Clone 0, Gen 353) on my GTS 450 with current Nvidia drivers. I hadn't been GPU folding for weeks because 2.15 kept doing its "CoreStatus = 63 (99)" on 6800 and 6801. It just finished the 6800 and switched to Project: 6801 (Run 5925, Clone 0, Gen 288) and is running that one as well. So for me, 2.20 is working great.

Ernie
Dennis Ernst
 
Posts: 7
Joined: Thu Jun 30, 2011 11:40 pm
Location: San Jose, Ca

Re: Core 2.20 for Fermi

Postby Leonardo » Sat Aug 20, 2011 3:36 am

Dennis, which Nvidia drivers are you using? Which Fermi chip does your GTS 450 have? (GPU-Z will show the chip model.)
Last edited by Leonardo on Sat Aug 20, 2011 9:20 pm, edited 1 time in total.
User avatar
Leonardo
 
Posts: 597
Joined: Tue Dec 04, 2007 5:09 am
Location: Eagle River, Alaska

Re: Core 2.20 for Fermi

Postby DocJonz » Sat Aug 20, 2011 6:57 am

I have three GTX 560Ti's all running Win7x64 with nVidia 275.33 and with -advmethods using the new core - all have completed about 4 WU's happily so far. I note the following observations for information (rig: GPU and CPU (i7 930 @ stock with -smp) are both running at 100%);
1. power draw has increased from 310 to 345W, i.e. 35W or 11% increase
2. GPU temperature increase of 5<T<10DegC
3. lag has started on the desktop
User avatar
DocJonz
 
Posts: 354
Joined: Thu Dec 06, 2007 6:31 pm
Location: United Kingdom

Re: Core 2.20 for Fermi

Postby filu » Sat Aug 20, 2011 11:26 am

The problem was solved. I reinstalled the operating system and core v.2.20 works without a problem.
Thank you all for your help. :)
filu
 
Posts: 76
Joined: Mon Aug 03, 2009 9:33 am
Location: Krzeszyce, Poland

Re: Core 2.20 for Fermi

Postby Waldeusz » Sat Aug 20, 2011 5:01 pm

@tjlane - OK it's a system problem.
I have 2 machines dedicated for 2x gpu GTX460
WinXp x64 - not work new core 2.2 (nvidia drivers 280.26) :(
Win7 x64 - All OK. I get 2 work unit p7620 and p7621 (nvidia drivers 267.24)

Have a nice day
Image
Waldeusz
 
Posts: 4
Joined: Tue Oct 19, 2010 11:04 pm

Re: Core 2.20 for Fermi

Postby kc2lrc » Sat Aug 20, 2011 5:52 pm

Picked up a couple of these so far on a GTX 470 yesterday. Here's my observations:

First, system setup:

- AMD X6 1090T running smp -5
- Asus GTX 470 running GPU3 -advmethods
- Sapphire HD5770 in the second slot (put in because a second 470 failed) running ATI core 11 with flags, FLUSH_INTERVAL = 256

Observations:
1. GPU temperature and fan speed are up. This unit obviously is using the entire CUDA array on the GPU! The temp, which would typically be high 70s/low 80s, was now hovering at 92C. The fan stays around 77% and is quite audible.
2. The inductor whine from the GPU seems to be gone, or at least reduced. Hard to tell because I have another box full of old ball-bearing HDDs running nearby.
3. PPD on the 470 is about the same.
4. PPD on the X6 has increased due to very low CPU usage (0-1%) from core 15.
5. There is some lag on the desktop. Interestingly, it only occurs on the primary monitor. The second monitor driven by the 470 does not experience it, nor does the additional monitor driven by the 5770.

The temp on the 470, it turns out, was being exacerbated by the 5770 sitting in the lower slot, which vents its heat right into the case. For now I disabled FAH on that card, as I'll soon be swapping another 470 back in. I'm not sure I should be concerned with temps until they exceed 95C, as this is just what NVIDIA designed the cards to do. The thermal throttle point is set to 105C on 470/480 cards. A 215W GPU is only going to run so cool!

Sounds like this is what us Fermi owners have been waiting for, and this will be the project that shows what our GPUs are capable of! I bought the 470s after all so the big projects wouldn't degrade my PPD after watching how lesser cards reacted to project 6811.
kc2lrc
 
Posts: 9
Joined: Sun Jan 04, 2009 11:10 pm

Re: Core 2.20 for Fermi

Postby Waldeusz » Sat Aug 20, 2011 6:44 pm

System setup :
Win7 x64 -> i7 875k @4.0
smp -8 bigadv /p6900/
2x GTX 460 @820 - working p7620 and p7621
Temp. gpu > 10 st C.... but ppd < ~ 400-500 for one gpu :e?:
I expected that it will be some bonus for tasks in extreme :lol:
Waldeusz
 
Posts: 4
Joined: Tue Oct 19, 2010 11:04 pm

PreviousNext

Return to V6 GPU3 beta (including Fermi) OpenMM

Who is online

Users browsing this forum: No registered users and 1 guest

cron