Old [H] Warrior, new tips needed

S

Strikemaster

[H]ard|Gawd
Joined
Mar 29, 2001
Messages
1,264
As seen here (http://hardforum.com/showpost.php?p=1037597555&postcount=84), I've been nudged to get back into active duty with the [H]orde. Given the mixed bag of machinery, what can be done with each to maximize their potential without making them unreliable for their primary tasks?

Phenom II x6 1090T / ATI 6870 / 8GB RAM / Win7 64-bit (gaming system)
Phenom x4 / NV 8800GT / 4GB RAM / Ubuntu 10.10 64-bit (Cacti)
Opteron dual-core 2.2ghz (AM2 socket) / ATI Rage onboard / 2GB / XP Pro SP3
Core 2 Duo 2.8ghz non HT / NV420 / 4GB / Win7 64-bit (workstation)

Back in the day, it was 5.04, penciled Durons in MP chassis and a whole lot of heat in the basement. I can't do that anymore, but I'll throw what I can at the Beast.


 
Well run SMP on all the rigs, that is your best use for power used and points per day.

If you want to get frisky, you could try and run the GPU client on the 8800gt in whine.

ATI cards burn a lot of power and the ppd is on the meh side.

SMP is your happy place. There are some great guides sticked.
 
I would recommend running SMP on the two Phenom IIs and the C2D. The Opteron isn't worth it from a performance/power consumption perspective. Other than that, try running the GPU client on the 8800GT like Kendrak suggested. If you really want to put the Opteron to use, I would suggest running a pair of uniprocessor clients as well, and as for the 6870, it would be better suited for a BOINC project or for Bitcoin mining if you're into that sort of thing.
 
So, the V7 Beta client for the two Phenoms, skip the 6870 and just go for SMP, while tweaking the idling Opteron with the old-style 6.23 uniproc client.

Sounds like a good jumping-off point. :D

I'd love to get my hands on the Opteron quad-core 2-CPU Socket F rig that hosts our backup email server, but it runs CentOS 5 and the boss might not care for that much heat in the rack. :p
 
The V7 beta isn't necessary. If I were you, I'd just stick with V6.
 
Welcome back Strikemaster... I do remember seeing you vaguely in the past ;)
 
Welcome Back Strikemaster.... It has been a very long while...

If you want to use that Opty for folding and maximize points, you could move the 8800GT into it. That way it will not rob processing power from the SMP unit in the Phenom. As mentioned, run the SMP client on the phenoms and this way the Opty helps out too.

 
jfb9301 said:
Welcome Back Strikemaster.... It has been a very long while...

If you want to use that Opty for folding and maximize points, you could move the 8800GT into it. That way it will not rob processing power from the SMP unit in the Phenom. As mentioned, run the SMP client on the phenoms and this way the Opty helps out too.

Click to expand...

Good idea, add a uniproc client and it will put out a respectable 5-5.5k PPD, A4 WU and the bonus will add to that
 
"Hiya!" back at Xilicon, DwDawg, been waay too long fellows...

The Opty / 8800GT combo can't happen because they are different owners. The Opty station is company property, while the Phenoms are personal gear. The machine is based on an Asus M2N-LR motherboard, and pretty much stands idle for days at a time. I think that needs to change. :p

For the 4-core Phenom that drives the 8800GT, would it work out to tie one Phenom core to a GPU-sited folding slot, leave the other three to be used as necessary? I can always try something like that with the V7 Beta / Ubuntu native client (I think)...

As for the 1090T, looks like the ~1650 PPD someone predicted is coming to pass. :D
 
Strikemaster said:
For the 4-core Phenom that drives the 8800GT, would it work out to tie one Phenom core to a GPU-sited folding slot, leave the other three to be used as necessary? I can always try something like that with the V7 Beta / Ubuntu native client (I think)...
Click to expand...
That would be unnecessary. Let the SMP client run on all four cores and run the GPU client at the same time. The SMP client will take a bit of a hit performance-wise, but not as much as it would if it only had three cores to run on.
Strikemaster said:
As for the 1090T, looks like the ~1650 PPD someone predicted is coming to pass. :D
Click to expand...
You should be getting about 12-14k PPD with that CPU. Use this guide to set up the SMP client: http://hardforum.com/showthread.php?t=1511766
 
Zero82z said:
Yes, I suppose it's possible that he's using an old monitoring program that doesn't calculate bonus points.
Click to expand...

Beta 7, and watching the Extreme Folding website results. Not a lot of movement yet.

From the logfile:

00:51:23:Server responded WORK_ACK (400)
00:51:23:Final credit estimate, 3620.00 points
00:51:23:Cleaning up Unit 00

We shall see once I get a few units through here. For the Ubuntu machine, I'll give the V6 client install a shot as you pointed out.
 
Zero82z said:
Yes, I suppose it's possible that he's using an old monitoring program that doesn't calculate bonus points.
Click to expand...

See below. No monitoring program. Last monitor I ran was Electron Microscope III.
Yeah, I've been a part of the [H]orde *THAT* long. Youngsters. Where's my danged cane? :D

jebo_4jc said:
Welcome back to the ranks, sir.
Click to expand...

Danke!

You know, the more I look around at the current Folding@Home project, the less I understand it. What is this bonus point award Zero82z mentioned? What is Langouste, and why should I care? It looks like a modernized Genome@Home "Dump" competition manager of sorts. ;)

OK, the 1090T Phenom has been running all weekend without errors using the V7 client ( no GPU, 6-thread SMP ) and has a registered Stanford passkey. What should I be looking for in the client, or where on the Web to see these "bonus points"?
 
Strikemaster said:
See below. No monitoring program. Last monitor I ran was Electron Microscope III.
Yeah, I've been a part of the [H]orde *THAT* long. Youngsters. Where's my danged cane? :D
Click to expand...
Some of us were also folding back then ;).
Strikemaster said:
You know, the more I look around at the current Folding@Home project, the less I understand it. What is this bonus point award Zero82z mentioned? What is Langouste, and why should I care? It looks like a modernized Genome@Home "Dump" competition manager of sorts. ;)

OK, the 1090T Phenom has been running all weekend without errors using the V7 client ( no GPU, 6-thread SMP ) and has a registered Stanford passkey. What should I be looking for in the client, or where on the Web to see these "bonus points"?
Click to expand...
Bonus points aren't counted separately. They're added to the credit you get for bonus-qualifying units once you upload them. Once you've submitted 10 SMP work units successfully, you'll start receiving bonus points.
 
Zero82z said:
Some of us were also folding back then ;).

Bonus points aren't counted separately. They're added to the credit you get for bonus-qualifying units once you upload them. Once you've submitted 10 SMP work units successfully, you'll start receiving bonus points.
Click to expand...

more clearly, 10 complete, and 80% of those completed satisfactory within deadlines. Then you just maintain the 80% there after

Additionally you need to complete them with your user name (public) and a 20 or so digit passkey issued to you (private), one of the stickies has the how to get the passkey.

without bonus, my 2600K would get about 4000 PPD ish.... with bonus its around 40000PPD
 
Zero82z said:
Some of us were also folding back then ;).

Bonus points aren't counted separately. They're added to the credit you get for bonus-qualifying units once you upload them. Once you've submitted 10 SMP work units successfully, you'll start receiving bonus points.
Click to expand...

Roger, thanks for clearing that up.

Gods and deities, sooooo tempting to get into the horsepower race again. :)

Edit: Looks like the 10-wu gate has been passed, the 12pm update showed 2 wu credited with 4525 points. The next was 69 pts, apparently from the single-core P4 (no HT) SNMP server that has generated 1384 work units for the [H]orde non-stop since 2006. Tiny gains, but a beach is made from single grains of sand...
 
Last edited:
Been doing a bit more reading in the Ubuntu setup thread, saw the "ext4 no-no" and thought "Aw %^&$, I do NOT want to reload this machine!", since it is running a Cacti status monitor app. Further in, I see what appears to be the reason why; the core "takes 40 minutes to write out to the drive in ext4" vs. moments in ext3.

Uhh, I'm not seeing it. :confused:

This machine is running Ubuntu "Natty Narwhal" Desktop 11.04, 64-bit, and using the 6.34 64-bit client. At the most so far, I've seen a 2 min. 30 second delay from past end to "core shutdown", as seen here in the log file:

[18:10:48] Finished Work Unit:
[18:10:48] - Reading up to 3701664 from "work/wudata_06.trr": Read 3701664
[18:10:48] trr file hash check passed.
[18:10:48] edr file hash check passed.
[18:10:48] logfile size: 59031
[18:10:48] Leaving Run
[18:10:49] - Writing 3795631 bytes of core data to disk...
[18:10:50] Done: 3795119 -> 3522653 (compressed to 92.8 percent)
[18:10:50] ... Done.
[18:13:14] - Shutting down core
[18:13:14]
[18:13:14] Folding@home Core Shutdown: FINISHED_UNIT
[18:13:34] CoreStatus = 64 (100)
[18:13:34] Sending work to server
[18:13:34] Project: 6979 (Run 0, Clone 25, Gen 180)
[18:13:34] + Attempting to send results [August 8 18:13:34 UTC]
[18:13:38] + Results successfully sent

Dumb luck, future screwup, or a non-issue in the 6.34 client?
 
Strikemaster said:
Been doing a bit more reading in the Ubuntu setup thread, saw the "ext4 no-no" and thought "Aw %^&$, I do NOT want to reload this machine!", since it is running a Cacti status monitor app. Further in, I see what appears to be the reason why; the core "takes 40 minutes to write out to the drive in ext4" vs. moments in ext3.

Uhh, I'm not seeing it. :confused:

This machine is running Ubuntu "Natty Narwhal" Desktop 11.04, 64-bit, and using the 6.34 64-bit client. At the most so far, I've seen a 2 min. 30 second delay from past end to "core shutdown", as seen here in the log file:

[18:10:48] Finished Work Unit:
[18:10:48] - Reading up to 3701664 from "work/wudata_06.trr": Read 3701664
[18:10:48] trr file hash check passed.
[18:10:48] edr file hash check passed.
[18:10:48] logfile size: 59031
[18:10:48] Leaving Run
[18:10:49] - Writing 3795631 bytes of core data to disk...
[18:10:50] Done: 3795119 -> 3522653 (compressed to 92.8 percent)
[18:10:50] ... Done.
[18:13:14] - Shutting down core
[18:13:14]
[18:13:14] Folding@home Core Shutdown: FINISHED_UNIT
[18:13:34] CoreStatus = 64 (100)
[18:13:34] Sending work to server
[18:13:34] Project: 6979 (Run 0, Clone 25, Gen 180)
[18:13:34] + Attempting to send results [August 8 18:13:34 UTC]
[18:13:38] + Results successfully sent

Dumb luck, future screwup, or a non-issue in the 6.34 client?
Click to expand...

Possibly due to the size of the file needing to be written, The -bigadv files run to about 100mb where as normal smp are usually 3-5mb, I only run linux on my -bigadv rigs so i haven't experimented with my x6
 
Ugh.

I just killed a completed 6903 that must have been in the middle of preparing for upload. I heard my hard drive absolutely chugging away on something like I had never heard before. I thought something had screwed up, so I rebooted the box. Looking at the logs, it must have been trying to save the WU.

The drive is formatted ext4.
 
jebo_4jc said:
I just killed a completed 6903 that must have been in the middle of preparing for upload.
Click to expand...
It is a good thing Uncle Tobit isn't around anymore... I'm sure he'd have a few choice words for you. :p
 
:)

Uncle Tobit doesn't need to worry. I said a few choice words to myself.
 
You must log in or register to reply here.
Back
Top