Really Big GPW Workunits for FAH

tjmagneto

tjmagneto

[H]ard DCOTM x2
Joined
Aug 6, 2008
Messages
3,299
Stanford must be getting ambitious with their workunits as I've got 2 units needing more than 5 hours to fold.

Project 9412 tpf 4:39 with my GTX 780ti @1137 Mhz est ppd: 240966

Project 9430 tpf 7:46 GTX 960 est ppd: 188648
Base speed is reported at 1304 Mhz, but I haven't found the reported boosted speed from within linux yet.

Check out the ppd difference between the 780ti and the 960, the 960 is scoring only 60k ppd less! Not too shabby for a card costing around 1/3rd the price of the 780ti. :eek:

One day later after starting this thread... Oh crap I just noticed my typo in the thread title. :confused:
 
Last edited:
The 9412 and 13 are above avg payers with 12000 atoms and the 9430 is slightly below avg payer. with 21000 atoms Stanford still has the benchmark problem they have always had. It is kind of funny to watch the 4P's with the 970 running on it. At this moment I have 2 - smp wu's running that are totaling 340k and the 970 GPU that is totaling 377k for a total of 717k but just an hour ago the set that was running on it was totaling 540k :rolleyes:

O-WELL that is Stanford benchmarking for you. :D
 
Grandpa_01 said:
At this moment I have 2 - smp wu's running that are totaling 340k
Click to expand...

How do you accomplish this? Remind me please. I'm still running a single 6.34 client with the bigadv flag which I believe means I'll get the biggest work units available even if BA was disbanded. Are you running 6.34 with specific -smp flags? Or are you running two slots on a 7.x client?
 
Grandpa - (thanks for the PM reply BTW) this is pretty much what I was wondering too: how you successfully deal with the FAH client core limit of 32 on a system that has 33+ cores. I tried running a second slot, but all that did was assign 16 cores per slot. Whoops. :p Or maybe I'll stop uni-tasking (which is my usual MO) and let FAH have 32 cores and various BOINC stuff have the other 32+ that are left over. ;)
 
Unfortunately some of these new GPU WUs are impossible to benchmark. That is because they perform differently on different nvidia architectures. On the same project my Kepler might do 20% above its average while at the same time my Maxwell does 20% below. No way to fix that with benchmarking. It all depends on the type of simulation.
 
ChristianVirtual said:
I think the 32 core limit is Windows only. Linux can deal with moaah cores.
Click to expand...

Ahhh. Yeah, that's possible. I might have a Linux image for these systems somewhere; but messing around with images is a bit of a PITFA. :p
 
brilong said:
How do you accomplish this? Remind me please. I'm still running a single 6.34 client with the bigadv flag which I believe means I'll get the biggest work units available even if BA was disbanded. Are you running 6.34 with specific -smp flags? Or are you running two slots on a 7.x client?
Click to expand...

Brilong I am running 7.4.4 and running 2 - smp slots and 1 GPU slot with the 4650's clocked @ 3175Mhz I think I had 2 - 8106 smp WU's running at the time I posted that 0ne at 168k and one at 171k and a GPU 9704 @ 429k for a total of 768k. Most of the time I am between 230k and 290k on the smp's but occasionally the folding gods all line up just right and I get a whopper..

Any way I found that on v7 that I get more PPD running 2 smp slots vs 1 bigadv :eek:

In this screen shot all of the 4P's are running at stock clocks

PPD_zpssc9fz16t.png
[/URL][/IMG]
 
Last edited:
There are some SMP WU that will go upto 48 cores, I run them on my 2p E5 boxes, but over that and you either have BA replacement WU or 2 SMP clients
 
Nathan_P said:
There are some SMP WU that will go upto 48 cores, I run them on my 2p E5 boxes, but over that and you either have BA replacement WU or 2 SMP clients
Click to expand...

So what does the config.xml look like? I don't install fahcontrol on my machines so I've been hacking config.xml manually. :cool:
 
brilong said:
So what does the config.xml look like? I don't install fahcontrol on my machines so I've been hacking config.xml manually. :cool:
Click to expand...

Code: 
<config>
  <!-- Client Control -->
  <fold-anon v='true'/>

  <!-- Folding Slot Configuration -->
  <gpu v='false'/>

  <!-- HTTP Server -->
  <allow v='10.0.0.1-10.0.0.255'/>

  <!-- Network -->
  <proxy v=':8080'/>

  <!-- Remote Command Server -->
  <command-allow-no-pass v='10.0.0.1-10.0.0.255'/>
  <password v='rick'/>

  <!-- Slot Control -->
  <power v='full'/>

  <!-- User Information -->
  <passkey v='fdaa'/>
  <team v='33'/>
  <user v='Grandpa'/>

  <!-- Folding Slots -->
  <slot id='0' type='CPU'>
    <max-packet-size v='big'/>
  </slot>
  <slot id='2' type='CPU'/>
  <slot id='1' type='GPU'/>
</config>

Code: 
<config><!-- Client Control -->
<fold-anon v="true"/><!-- Folding Slot Configuration -->
<gpu v="false"/><!-- HTTP Server -->
<allow v="10.0.0.1-10.0.0.255"/><!-- Network -->
<proxy v=":8080"/><!-- Remote Command Server -->
<command-allow-no-pass v="10.0.0.1-10.0.0.255"/><password v="rick"/><!-- Slot Control -->
<power v="full"/><!-- User Information -->
<passkey v="fdaa
"/><team v="33"/><user v="Grandpa"/><!-- Folding Slots -->
<slot id="0" type="CPU"><max-packet-size v="big"/></slot><slot id="2" type="CPU"/><slot id="1" type="GPU"/></config>
 
Last edited:
Before I saw your post, I went with this instead:
Code: 
  <!-- Folding Slots -->
  <slot id='0' type='CPU'>
    <client-type v='smp'/>
    <cpus v='32'/>
  </slot>
  <slot id='1' type='CPU'>
    <client-type v='smp'/>
    <cpus v='30'/>
  </slot>
  <slot id='2' type='GPU'>
    <max-packet-size v='big'/>
    <client-type v='beta'/>
    <gpu-index v='0'/>
  </slot>
  <slot id='3' type='GPU'>
    <max-packet-size v='big'/>
    <client-type v='beta'/>
    <gpu-index v='1'/>
  </slot>
 
The only difference is v7 chooses to run cpu v=31 which seems odd but it works
 
brilong said:
So what does the config.xml look like? I don't install fahcontrol on my machines so I've been hacking config.xml manually. :cool:
Click to expand...

I'll post one once I'm back from holiday, I'm away from home and the rigs are off
 
You must log in or register to reply here.
Back
Top