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

Best use of 4P machine

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

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Ever since PG shut down the bigadv program my 4P Magny-Corus Opteron 6174s haven't seen much action :(... but I want to bring them back to FAH. Strange that I see some GRO_A5 work in the summary... thought it was dead.

System is running Windows Server 2012 R2 and will probably move to 2016 when it RTMs. I need Windows as I'm doing other Windows specific things with this machine. However, I'm totally open to running Linux VM(s) in Hyper-V.

Curious to know what others are doing with their 4P builds these days.
 

donuts

Member
Joined
Dec 5, 2006
Location
SC (now)
Curious also. I built mine for Rosetta crunching, and earlier this year purchased a Folder's 4p for the same project.
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Ubuntu 16.04 in Hyper-V VM - 16 cores @2.2GHz - Client v7.4.4

Name: Ubuntu VM - Quadteron Slot 00
Number of Frames Observed: 29

Min. Time / Frame : 00:00:51 - 36,056.0 PPD
Avg. Time / Frame : 00:00:53 - 34,034.5 PPD
Cur. Time / Frame : 00:00:52 - 34,791.4 PPD
R3F. Time / Frame : 00:00:54 - 33,060.3 PPD
All Time / Frame : 00:00:53 - 33,907.3 PPD
Eff. Time / Frame : 00:00:54 - 33,060.3 PPD


This outperforms my 4.4GHz 2600k... barely.

Name: Windows - Main Workstation Slot 00
Number of Frames Observed: 300

Min. Time / Frame : 00:00:57 - 30,515.5 PPD
Avg. Time / Frame : 00:01:00 - 28,255.7 PPD


The Ubuntu VM is currently configured with 32 physical cores. I was unable to obtain ANY assignment using client v6.34. I was also unable to obtain an assignment using client v7.4.4 with the CPU slot configured to run all 32 cores. Once I backed it down to 16 the client was assigned to a work server. Maybe there's something in the AS logic that won't assign high core counts to a work server.
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
That's beginning to sound like the reasonable thing to do. However, the PPW (points-per-watt) is probably not cost effective. I guess you're running 6200 series (16-core) Opteron Bulldozers?

Now, this is my thread so I'll side track it a little bit... What the heck kind of GPU is that pushing 574k PPD? I've been out of the loop for far too long.
 

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
Now, this is my thread so I'll side track it a little bit... What the heck kind of GPU is that pushing 574k PPD? I've been out of the loop for far too long.

:rofl::rofl::rofl:
That's a GTX 980 Ti, but that's nothing compared to what the the GTX 1080's get. And yes, 6276's.
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Too expensive for this 'married with children' guy... maybe if I could find a used 970 or 980. :D
 

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
You should see HayesK's GPU farm :) all his points are from GPU's :rofl::comp:
Check out the Meep Meep thread :)
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Yeah, I perused that one earlier today and now just went back and looked at the production of his various GPUs. Guess it's not worth trying to run FAH GPU on my GTX 470 huh? :D
 

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
Ok, here is what 1 of my 4P's is getting (I'll turn both on when it gets cold :)) this is just the benchmarks but it gives you an idea of what a 4P can still do with a4 core WU's. Now when I get my GTX 1080 SC, I'll stop folding on the 4P's and use them as a server :)

9034time.jpg

9034ppp.jpg

9752time.jpg

9752ppd.jpg
 

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
Yeah, I perused that one earlier today and now just went back and looked at the production of his various GPUs. Guess it's not worth trying to run FAH GPU on my GTX 470 huh? :D

Hehehe yep, the days of folding on a GT(x/s) 4xx is over, I have a Win 7 with a GTS 450, and if I get a core 21 on it, it'll take a little over 3 days to do 1 WU, not a good return at all. The GTX 980 Ti I have is on a Win 10 PC.
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Heya Harlam, the GTX 1070 is a bit cheaper and still gets 500k, maybe more.

Yeah, looks like the mid $400s range. Be nice if I could find a used 970 for about $200. :)

@WhiteHawk - have two slots running now, each with 16 cores, frame times on the first slot are holding steady... second slot is a little slower, could be the WU. I'm wondering if 12 cores per slot would be better on my 6174s.

HFM1.png
 
Last edited:

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
@Whitehawk - have two slots running now, each with 16 cores, frame times on the first slot are holding steady... second slot is a little slower, could be the WU. I'm wondering if 12 cores per slot would be better on my 6174s.

I would go for it, better points over all than 2 16 core slots :thup:
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Yeah, I'm just experimenting... want to find the right combination when it comes to production, power consumption, heat output, etc. Running 100% is probably not an option for me right now simply due to heat output. It's already 95 degrees outside with the index into the 100s.

Really wish I could run a single slot and throw 30+ cores at it... I'm puzzled why such a configuration won't receive a valid assignment. I'll probably go complain over at FF.org and see what I can stir up. :D
 

Nathan_P

Member
Joined
Apr 28, 2012
Location
Jersey, Channel islands
24 threads is the most you will currently get work with, all the A5 bigadv projects and core A3 projects running more than 24 threads have finished, or at the very least will not assign to any machine. The project owner, Kasson, has disappeared and no o0ne has been able to contact him.

Some of the old tips and tricks for Linux will still work, the Kraken is still known to work for a start
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Thanks Nathan! Good to know. I'd prefer to run v6 in Linux actually. However, before I cranked up this v7 client I was unable to get ANY assignment with v6 in the same VM (32-cores). Perhaps the AS is seeing the logical core count of the machine as opposed to the number of threads specified in the v6 client arguments, i.e. -smp 16, for example. I'll finish up these units and then adjust the VM core count to 24 and see if I can get an assignment.

Ah, the Kraken... been a while... I have v0.6 here on my server. Looks like v0.7-pre15 is the latest. Says it only works for A3 and A5 cores... I cannot recall the last time I saw an A3 assignment, A5 being dead of course; A4 is all I get. Do you happen to know the flag combination that may run A3 work on v6? Looks like there are plenty of A3 projects still in production.
 

ihrsetrdr

Señor Senior Member
Joined
May 17, 2005
Thanks Nathan! Good to know. I'd prefer to run v6 in Linux actually.

Yea, me too, but those days are gone, but at least v7.4xx runs well in Linux.

Q: how about HFM.NET in Linux, does it run in Linux with Mono installed?
 

WhitehawkEQ

Premium Member
Joined
Dec 6, 2010
Perhaps the AS is seeing the logical core count of the machine as opposed to the number of threads specified in the v6 client arguments, i.e. -smp 16, for example. I'll finish up these units and then adjust the VM core count to 24 and see if I can get an assignment.

I don't think the AS cares how many cores a PC has, I think it has to do with the WU it's self, something I read on [email protected] forums. 1 thing you can do is set 1 slot for 8 or 16 cores, get a a4 WU the pause it and change the core to 48 or 64, but then you would be a slave to it changing it back and forth :)
 
OP
harlam357

harlam357

Senior Fold-a-holic
Joined
Sep 22, 2004
Yea, me too, but those days are gone, but at least v7.4xx runs well in Linux.

Q: how about HFM.NET in Linux, does it run in Linux with Mono installed?

HFM is part of the reason I want to keep v6 running on one of my machines; simply as a live test case. Been working on HFM code quite a bit lately and considered removing the code that is tailored to v6. Doing so would allow me to simplify some less than great abstractions I made when integrating v7 support. I've moved beyond that choice to remove v6 support at the moment, but this experience may prompt be back down that path sooner rather than later.

As for Linux support... I have my latest build of HFM running well on Ubuntu 16.04. The current v0.9.6 will likely fall on its face. Seems there's some regression in Mono with the baked in user settings system, which is what backs HFM's preferences (Edit > Preferences). It worked in Mono 2.8+ through 3.x. The Mono packages for Ubuntu 16.04 are v4.2. This user settings system has also been a long standing source of frustration as even the Microsoft .NET Framework implementation of the system would sometimes fail gloriously and produce unreadable xml. I've found numerous accounts of the same issues with other .NET applications using this feature. So, I finally ditched it and rolled my own preferences storage system. My motivation was to get things running on Linux again and that has been accomplished.

I know I'll sound like a broken record here... but I should have v0.9.7 out as soon as I patch a few issues I've found with my latest changes. If there is enough interest from the community I may look into integrating "fast track" release mechanism into HFM; where you could choose to be notified of the latest "unstable" bits and try them out if you wish. However, I'll have to invest some time in either my local build server infrastructure or migrate to a cloud based build infrastructure. Either way there is a time investment I'd need to make to get that rolling.

I don't think the AS cares how many cores a PC has, I think it has to do with the WU it's self, something I read on [email protected] forums. 1 thing you can do is set 1 slot for 8 or 16 cores, get a a4 WU the pause it and change the core to 48 or 64, but then you would be a slave to it changing it back and forth :)

It doesn't... tried configuring the VM for 24 cores. Still no work on v6. :(