• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

poor ppd with GPU folding

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

pik4chu

Senior Yellow Forum Rat
Joined
Jan 17, 2003
Location
Centennial, Colorado
Well I don't know what it is but I get terrible PPD with any video card Ive been trying to fold with and I dont know why. And I mean terrible like 1/2 what Im told the cards should be getting.

8800GTS @ ~1600ppd
GTX 260 @ ~2800PPD
GTX 285 @ ~3300PPD :mad:

Stuff like that. I didnt do anything wierd or have some unique setup, host OS are either XP Pro or Vista Ultimate 64bit. Processors are Q6600 and E6300/E4300 with the 8800GT's and I7 920's for the 2XX GTX's. The cpu's are doing what they should be roughly ppd wise but the video cards just aren't doing a thing despite how quickly they seem to be tearing through WU's. Settings for the clients are normal, with the 'slightly higher priority' option and 100% CPU.

Are there some kind of special settings, switches or registry tweaks I am missing here? Because with as much heat these things dump out (especially this dang 285) this almost doesn't feel worth it to fold on them if this is the kind of efficiency I will be getting.
 
The GPU priority must be set to low (console config or slightly higher in systray) and the VM priority must be set to low using task manager or idle by editing the VMware config file (.vmx). Here is the script to set affinity and priority on Vmware server:
VM1:

priority.ungrabbed = "idle"
processor0.use="TRUE"
processor1.use="TRUE"
processor2.use="FALSE"
processor3.use="FALSE"

VM2:
priority.ungrabbed = "idle"
processor0.use="FALSE"
processor1.use="FALSE"
processor2.use="TRUE"
processor3.use="TRUE"

replace priority.ungrabbed="normal" with the lines above in each VM for a quad running two VMs.
 
not using VMWare for the CPU clients, just running them as services on the windows Boxen. See image
 

Attachments

  • FAHGPU.PNG
    FAHGPU.PNG
    22.7 KB · Views: 182
The GPU priority must be set to low (console config or slightly higher in systray) and the VM priority must be set to low using task manager or idle by editing the VMware config file (.vmx). Here is the script to set affinity and priority on Vmware server:
VM1:

priority.ungrabbed = "idle"
processor0.use="TRUE"
processor1.use="TRUE"
processor2.use="FALSE"
processor3.use="FALSE"

VM2:
priority.ungrabbed = "idle"
processor0.use="FALSE"
processor1.use="FALSE"
processor2.use="TRUE"
processor3.use="TRUE"

replace priority.ungrabbed="normal" with the lines above in each VM for a quad running two VMs.

what if the VMX does not have those lines where would it go ???

Code:
#!/usr/bin/vmplayer
.encoding = "windows-1252"

# This is a Workstation 5 or 5.5 config file
# It can be used with Player
config.version = "8"
virtualHW.version = "4"

# Number of virtual CPUs. Your virtual machine will not
# work if this number is higher than the number of your physical CPUs
numvcpus = "2"

# Memory
memsize = "1024"

# Selected operating system for your virtual machine
guestOS = "other-64"

# displayName is your own name for the virtual machine
displayName = "Folding@Home"

# These fields are free text description fields
annotation = "Diskless Folding@Home"
guestinfo.vmware.product.long = "Diskless Folding by Nick Reilly"
guestinfo.vmware.product.url = "http://reilly.homeip.net"
guestinfo.vmware.product.class = "virtual machine"

# First serial port, physical COM1 is not available
serial0.present = "FALSE"

# Optional second serial port, physical COM2 is not available
serial1.present = "FALSE"

# First parallel port, physical LPT1 is not available
parallel0.present = "FALSE"

# Settings for physical floppy drive
floppy0.present = "FALSE"

# Logging
logging = "FALSE"

# These settings decides interaction between your
# computer and the virtual machine
isolation.tools.hgfs.disable = "FALSE"
isolation.tools.dnd.disable = "FALSE"
isolation.tools.copy.enable = "FALSE"
isolation.tools.paste.enabled = "FALSE"

# First network interface card
ethernet0.present = "TRUE"
ethernet0.virtualDev = "e1000"
ethernet0.connectionType = "nat"
ethernet0.addressType = "generated"
ethernet0.generatedAddressOffset = "0"

ide0:0.present = "TRUE"
ide0:0.fileName = "folding.vmdk"

ide0:0.redo = ""

extendedConfigFile = "folding.vmxf"

virtualHW.productCompatibility = "hosted"
tools.upgrade.policy = "manual"

checkpoint.vmState = ""
ethernet0.generatedAddress = "00:0c:29:f0:54:40"
uuid.location = "56 4d 46 c3 f0 86 5f 88-a3 98 59 11 e8 f0 54 40"
uuid.bios = "56 4d 46 c3 f0 86 5f 88-a3 98 59 11 e8 f0 54 40"
vmotion.checkpointFBSize = "16777216"

tools.syncTime = "FALSE"
debugStub.linuxOffsets = "0x0,0xffffffff,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0,0x0"
 
so based on the recommended VMware settings are you suggest I set affinity for the CPU clients but give the GPU client full access to all cores?
 
You must be having a priority issue. If that screenie is from a sCPU client, you've got to change it to lowest priority to avoid having it steal cpu resources the GPU needs.

CPU to idle, gpu to low.
 
VMware player may default to idle priority. I don't know. THe script can be put anywhere in the file. Make a backup first, since I haven't tried this in player.
 
You must be having a priority issue. If that screenie is from a sCPU client, you've got to change it to lowest priority to avoid having it steal cpu resources the GPU needs.

CPU to idle, gpu to low.

That is the GPU client console. Didn't know the CPU and GPU GUIs looked the same or I would have specified.
 
If you're running the cpu systray client it looks pretty much the same as the GPU client ( I don't run either of the systray clients so I'm going from memory). Anyway, no matter what, the GPU client has to be set to higher priority than any DC client running on the cpu.
 
Read the file client.cfg and make sure the GPU client has a line that says priority=96 and the cpu client says priority=0 or no priority statement at all.

The results from my rigs on one of the worst WUs and one of the best:

Code:
 Project : 5754
 Core    : GPUv2 Gromacs
 Frames  : 100
 Credit  : 511


 -- CR 260GTX 1 --

 Min. Time / Frame : 1mn 25s  - 5194.16 ppd
 Avg. Time / Frame : 1mn 30s  - 4905.60 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- CR 260GTX 2 --

 Min. Time / Frame : 1mn 22s  - 5384.20 ppd
 Avg. Time / Frame : 1mn 23s  - 5319.33 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- PR 8800GT --

 Min. Time / Frame : 2mn 05s  - 3532.03 ppd
 Avg. Time / Frame : 2mn 05s  - 3532.03 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- Val2 8800GTS 1 --

 Min. Time / Frame : 1mn 51s  - 3977.51 ppd
 Avg. Time / Frame : 1mn 51s  - 3977.51 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- HTPC 9800 GTX --

 Min. Time / Frame : 1mn 39s  - 4459.64 ppd
 Avg. Time / Frame : 1mn 39s  - 4459.64 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- Val2 8800GTS 2 --

 Min. Time / Frame : 1mn 48s  - 4088.00 ppd
 Avg. Time / Frame : 1mn 48s  - 4088.00 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame

Project : 5772
 Core    : GPUv2 Gromacs
 Frames  : 100
 Credit  : 353


 -- CR 260GTX 1 --

 Min. Time / Frame : 40s  - 7624.80 ppd
 Avg. Time / Frame : 40s  - 7624.80 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- CR 260GTX 2 --

 Min. Time / Frame : 38s  - 8026.11 ppd
 Avg. Time / Frame : 38s  - 8026.11 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- PR 8800GT --

 Min. Time / Frame : 54s  - 5648.00 ppd
 Avg. Time / Frame : 54s  - 5648.00 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- Val2 8800GTS 1 --

 Min. Time / Frame : 49s  - 6224.33 ppd
 Avg. Time / Frame : 49s  - 6224.33 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- HTPC 9800 GTX --

 Min. Time / Frame : 44s  - 6931.64 ppd
 Avg. Time / Frame : 44s  - 6931.64 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame


 -- Val2 8800GTS 2 --

 Min. Time / Frame : 48s  - 6354.00 ppd
 Avg. Time / Frame : 49s  - 6224.33 ppd
 No Cur. Time / Frame
 No R3F. Time / Frame
 No Eff. Time / Frame

Yours should fall in the same range.

What drivers are you running?
 
Try this and see what happens. In task manager, set the affinity of the CPU clients so that they avoid the processor the GPU client is running on. On a quad, the GPU client should default to 3. Set the cpu clients to avoid core 3. You don't have to stop the clients. Doing this removes priority as an issue.
 
185.5 and 185.1 (tried both, same results) on the 2XX cards and something apparently much older on this 8800.

Here is a log snippet from the 8800's current WU

Code:
[21:31:14] Project: 5751 (Run 1, Clone 224, Gen 293)
[21:31:14] 
[21:31:14] Assembly optimizations on if available.
[21:31:14] Entering M.D.
[21:31:20] Working on Protein
[21:31:24] Client config found, loading data.
[21:31:24] Starting GUI Server
[21:35:23] Completed 1%
[21:39:27] Completed 2%
[21:43:28] Completed 3%
[21:47:27] Completed 4%
[21:51:30] Completed 5%
[21:55:30] Completed 6%
[21:59:27] Completed 7%
[22:03:22] Completed 8%
[22:07:18] Completed 9%
[22:11:20] Completed 10%
[22:15:16] Completed 11%
[22:19:11] Completed 12%
[22:23:10] Completed 13%
[22:27:09] Completed 14%
[22:31:08] Completed 15%
[22:35:02] Completed 16%
[22:39:03] Completed 17%
FAHMON estimates this to be running 1855PPD based on the above.

and the priority statements for both clients are as they should be (96 for GPU and no statement for CPU)

CPU Client.cfg
Code:
[settings]
username=[OC]Pik4chu
team=32
passkey=
asknet=no
machineid=1
bigpackets=big
extra_parms=-smp -deino -verbosity 9
local=21

[http]
active=no
host=localhost
port=8080
usereg=no

[clienttype]
type=3

[core]
addr=

GPU client.cfg
Code:
[settings]
username=[OC]Pik4chu
team=32
passkey=
asknet=no
machineid=5
bigpackets=big
extra_parms=
local=76

[http]
active=no
host=localhost
port=8080
usereg=no
proxy_name=
proxy_passwd=

[core]
priority=96
cpuusage=100
disableassembly=no
nocpulock=0
checkpoint=15

[power]
battery=no
 
I don't see anything wrong in the configs. You're running 4 minute frames when you should be running 2:10 or so. If you're running any driver older than 180.60 on the 8800, you must avoid having the SMP client run on the same cpu core as the GPU client. That could explain the problem with that GPU.
On the 2xx cards, are they in the same machine?
 
no, and the 260 isnt in a machine anymore for the moment, PSU is being RMA'd

here are snippets from the I7 rig in my sig

Code:
[23:06:51] Project: 5766 (Run 6, Clone 130, Gen 94)
[23:06:51] 
[23:06:51] Assembly optimizations on if available.
[23:06:51] Entering M.D.
[23:06:57] Working on Protein
[23:06:58] Client config found, loading data.
[23:06:58] Starting GUI Server
[23:07:32] Completed 1%
[23:08:06] Completed 2%
[23:08:41] Completed 3%
[23:09:15] Completed 4%
[23:09:49] Completed 5%
[23:10:23] Completed 6%
[23:10:57] Completed 7%
[23:11:32] Completed 8%
[23:12:06] Completed 9%
[23:12:40] Completed 10%
[23:13:14] Completed 11%
[23:13:48] Completed 12%
[23:14:23] Completed 13%
[23:14:57] Completed 14%
[23:15:31] Completed 15%
[23:16:05] Completed 16%
[23:16:39] Completed 17%
[23:17:13] Completed 18%
[23:17:48] Completed 19%
[23:18:22] Completed 20%
[23:18:56] Completed 21%
[23:19:31] Completed 22%
[23:20:05] Completed 23%
[23:20:39] Completed 24%
[23:21:13] Completed 25%
[23:21:47] Completed 26%
[23:22:21] Completed 27%
[23:22:56] Completed 28%
[23:23:30] Completed 29%
[23:24:04] Completed 30%
[23:24:38] Completed 31%
[23:25:12] Completed 32%
[23:25:47] Completed 33%
[23:26:21] Completed 34%
[23:26:55] Completed 35%
[23:27:29] Completed 36%
[23:28:03] Completed 37%
[23:28:38] Completed 38%
[23:29:12] Completed 39%
[23:29:46] Completed 40%
[23:30:20] Completed 41%
[23:30:54] Completed 42%
[23:31:29] Completed 43%
[23:32:03] Completed 44%
[23:32:37] Completed 45%
[23:33:11] Completed 46%
[23:33:45] Completed 47%
[23:34:20] Completed 48%
[23:34:54] Completed 49%
[23:35:28] Completed 50%
[23:36:02] Completed 51%
[23:36:36] Completed 52%
[23:37:11] Completed 53%
[23:37:45] Completed 54%
[23:38:19] Completed 55%
[23:38:53] Completed 56%
[23:39:27] Completed 57%
[23:40:01] Completed 58%
[23:40:36] Completed 59%
[23:41:10] Completed 60%
[23:41:44] Completed 61%
[23:42:18] Completed 62%
[23:42:52] Completed 63%
[23:43:27] Completed 64%
[23:44:01] Completed 65%
[23:44:35] Completed 66%
[23:45:09] Completed 67%
[23:45:43] Completed 68%
[23:46:18] Completed 69%
[23:46:52] Completed 70%
[23:47:26] Completed 71%
[23:48:00] Completed 72%
[23:48:35] Completed 73%
[23:49:09] Completed 74%
[23:49:43] Completed 75%
[23:50:17] Completed 76%
[23:50:52] Completed 77%
[23:51:26] Completed 78%
[23:52:00] Completed 79%
[23:52:34] Completed 80%
[23:53:08] Completed 81%
[23:53:43] Completed 82%
[23:54:19] Completed 83%
[23:54:54] Completed 84%
[23:55:31] Completed 85%
[23:56:09] Completed 86%
[23:56:44] Completed 87%
[23:57:20] Completed 88%
[23:57:55] Completed 89%
[23:58:30] Completed 90%
[23:59:05] Completed 91%
[23:59:42] Completed 92%
[00:00:18] Completed 93%
[00:00:54] Completed 94%

Code:
[21:48:36] Project: 3065 (Run 2, Clone 235, Gen 23)
[21:48:36] 
[21:48:37] Assembly optimizations on if available.
[21:48:37] Entering M.D.
[21:48:43] Rejecting checkpoint
[21:48:44] lambda5_99sb_big
[21:48:44] Writing local files
[21:48:44] 
[21:48:44] Writing local files
[21:48:44] Extra SSE boost OK.
[21:48:44] Writing local files
[21:48:44] Completed 0 out of 2500000 steps  (0 percent)
[22:03:45] Timered checkpoint triggered.
[22:13:35] Writing local files
[22:13:36] Completed 25000 out of 2500000 steps  (1 percent)
[22:28:35] Timered checkpoint triggered.
[22:30:56] Writing local files
[22:30:56] Completed 50000 out of 2500000 steps  (2 percent)
[22:41:53] Writing local files
[22:41:54] Completed 75000 out of 2500000 steps  (3 percent)
[22:52:50] Writing local files
[22:52:50] Completed 100000 out of 2500000 steps  (4 percent)
[23:03:47] Writing local files
[23:03:47] Completed 125000 out of 2500000 steps  (5 percent)
[23:14:46] Writing local files
[23:14:46] Completed 150000 out of 2500000 steps  (6 percent)
[23:25:43] Writing local files
[23:25:43] Completed 175000 out of 2500000 steps  (7 percent)
[23:36:46] Writing local files
[23:36:46] Completed 200000 out of 2500000 steps  (8 percent)
[23:47:49] Writing local files
[23:47:49] Completed 225000 out of 2500000 steps  (9 percent)
[23:59:05] Writing local files
[23:59:05] Completed 250000 out of 2500000 steps  (10 percent)

Code:
[01:58:36] Project: 3065 (Run 5, Clone 334, Gen 18)
[01:58:36] 
[01:58:37] Assembly optimizations on if available.
[01:58:37] Entering M.D.
[01:58:55]  on if available.
[01:58:55] Entering M.D.
[01:59:01] Rejecting checkpoint
[01:59:01] lambdExtra SSE boost OK.
[01:59:01] local files
[01:59:01] Extra SSE boost OK.
...............

[16:07:26] Completed 1275000 out of 2500000 steps  (51 percent)
[16:22:25] Timered checkpoint triggered.
[16:28:09] Writing local files
[16:28:09] Completed 1300000 out of 2500000 steps  (52 percent)
[16:43:09] Timered checkpoint triggered.
[16:48:53] Writing local files
[16:48:53] Completed 1325000 out of 2500000 steps  (53 percent)
[17:03:54] Timered checkpoint triggered.
[17:17:12] Writing local files
[17:17:12] Completed 1350000 out of 2500000 steps  (54 percent)
[17:32:12] Timered checkpoint triggered.
[17:47:13] Timered checkpoint triggered.
[17:51:14] Writing local files
[17:51:14] Completed 1375000 out of 2500000 steps  (55 percent)
[18:06:14] Timered checkpoint triggered.
[18:09:12] Writing local files
[18:09:12] Completed 1400000 out of 2500000 steps  (56 percent)
[18:24:12] Timered checkpoint triggered.
[18:25:56] Writing local files
[18:25:56] Completed 1425000 out of 2500000 steps  (57 percent)
[18:40:57] Timered checkpoint triggered.
[18:42:43] Writing local files
[18:42:44] Completed 1450000 out of 2500000 steps  (58 percent)
[18:57:43] Timered checkpoint triggered.
[18:59:28] Writing local files
[18:59:28] Completed 1475000 out of 2500000 steps  (59 percent)
[19:14:29] Timered checkpoint triggered.
[19:16:15] Writing local files
[19:16:15] Completed 1500000 out of 2500000 steps  (60 percent)
[19:31:15] Timered checkpoint triggered.
[19:33:04] Writing local files
[19:33:04] Completed 1525000 out of 2500000 steps  (61 percent)
[19:48:04] Timered checkpoint triggered.

HFM is saying ~3800PPD for the CPU WUs and unknown for the GPU but its still in Beta so meh, but the frame times are listed above.

*edit* I tihnk I figured out the issue with the 285GTX, I recently changed user accounts on this box and reset up the clients, looks like I missed the machine ID and priority option this time around. Going to check over my other 8800 rig to see what it's numbers look like.

*edit* ok that fixed the 285. According to HFM this machine is now doing ~13k PPD I started one of the 8800 machines from scratch, removed the vid card and drivers and the GPU client and started again and its now doing ~3800PPD with jsut the card. I will have to mess with the other machine tomorrow since it is at work but I suspect its a driver issue at this point. Thanks for the tips ChasR, helped point me in the right direction.

*GTX 285 is averaging a nice ~35s per frame now. ~8500ppd according to HFM
 
Last edited:
Woot! production is up where it should be :)

Code:
Team Rank          1,539
[SIZE="3"][B]Last 24hr Points   28,492[/B][/SIZE]
24hr Avg Points    19,416
Today Points       18,289

Thanks again ChasR

oh and the issue with the other 8800 was a very old driver. Updated it to the 185.5 and its rocking at ~1:50 per frame. have a :beer: on me!
 
Back