GPU FLOPS: Theory vs Reality

Message boards : Number crunching : GPU FLOPS: Theory vs Reality
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 20 · Next

AuthorMessage
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13833
Credit: 208,696,464
RAC: 304
Australia
Message 1937329 - Posted: 26 May 2018, 23:25:48 UTC - in response to Message 1937326.  

I hope you know I was completely kidding :)
But damn, I can't believe I forgot power usually goes up with optimizations :P

But it generally results in improved Credit/W-Hr.
ie more WUs processed for each Watt of power used, ie improved efficiency.
Grant
Darwin NT
ID: 1937329 · Report as offensive
Mark Seeger

Send message
Joined: 16 May 99
Posts: 47
Credit: 16,558,494
RAC: 116
United States
Message 1944758 - Posted: 16 Jul 2018, 16:24:43 UTC

Hi, has anyone looked at other (more bespoke) hardware performance-per-watt? Such as using raspberry pi’s or other scalable hardware?

I’d like to build a hobby SETI rig that optimizes compute performance per watt of power.

Thank you,

Mark
ID: 1944758 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22436
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1944763 - Posted: 16 Jul 2018, 16:45:00 UTC

Back of the proverbial envelope:
A Pi will do a task in something like 20 hours, and drawing about 5 watts from the wall, thus a single task will consume about 100W-hours.
A single GTX1050ti, in a suitable low power computer will do a single task in about 10-15 minutes, drawing about 125 watts from the wall, thus a single task will consume about 32W-hours.

So, if my figures are anywhere near correct the computer plus GTX1050ti wins by a factor of three.

However if one considers the capital cost, we have about $50 for the Pi, and about $300 for the computer plus GPU, and the Pi wins by a factor of six. What the pay-back time of the computer+GPU is compared to the Pi will depend on the local electricity cost.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1944763 · Report as offensive
Mark Seeger

Send message
Joined: 16 May 99
Posts: 47
Credit: 16,558,494
RAC: 116
United States
Message 1944768 - Posted: 16 Jul 2018, 17:03:09 UTC - in response to Message 1944763.  

Bob, super helpful thank you.

So, leaving CAPEX aside, I’d be better off buying a single NVIDIA GP100 (~$7500 GPGPU) for my PC, rather than tons of Pi’s, it would seem. Is the GP100 compatible with
Seti? I would think so...has any ever owned one/run seti on it?

Thank you,

Mark
ID: 1944768 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14672
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1944769 - Posted: 16 Jul 2018, 17:05:19 UTC - in response to Message 1944763.  

So, if my figures are anywhere near correct...
Pretty close. Host 8121358 is such a machine, running SETI as I type. The current tasks run about 10 - 11 minutes, and the watt meter I've just plugged it into is wobbling +/- 100 watts: the highest power draw I've seen is 107 watts during the quick parts near the beginning of a task - it drops as you head towards the sludge at the end.

Three cores of the i5 CPU are also 'crunching for science', though not SETI. Power is for the whole system base, excluding monitor. Machine is a Dell Optiplex 5040, with SSD boot drive supplied by Dell: I added the 1050 Ti.
ID: 1944769 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1944778 - Posted: 16 Jul 2018, 18:52:23 UTC - in response to Message 1944768.  
Last modified: 16 Jul 2018, 18:58:21 UTC

Bob, super helpful thank you.

So, leaving CAPEX aside, I’d be better off buying a single NVIDIA GP100 (~$7500 GPGPU) for my PC, rather than tons of Pi’s, it would seem. Is the GP100 compatible with
Seti? I would think so...has any ever owned one/run seti on it?

Thank you,

Mark

Yes the GP100 would run on Seti. But why you spend the $7500 for it when a GTX 1080Ti or Nvidia Titan V would perform much better at 1/4 or 1/2 the cost respectively. Seti doesn't use the FP64 performance of the Tesla's or Quadros. Other projects do but the GPGPU performance is not needed on Seti. Seti likes lots of shaders and high clocks.

The highest performing card currently being run on Seti as far as production goes is the Titan V at $3000. It runs best on the Petri CUDA 9.3 Linux application.
Petri's Host

[Edit]
Found the power draw for the Titan V from one of Petri's posts'

168/300W
That is the highest I've seen a TITAN V use. Things may change. This is with BLC 16 2bit guppi that run in 34 seconds.
Petri

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1944778 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1945156 - Posted: 18 Jul 2018, 13:32:37 UTC - in response to Message 1944763.  

Back of the proverbial envelope:
A Pi will do a task in something like 20 hours, and drawing about 5 watts from the wall, thus a single task will consume about 100W-hours.
A single GTX1050ti, in a suitable low power computer will do a single task in about 10-15 minutes, drawing about 125 watts from the wall, thus a single task will consume about 32W-hours.

So, if my figures are anywhere near correct the computer plus GTX1050ti wins by a factor of three.

However if one considers the capital cost, we have about $50 for the Pi, and about $300 for the computer plus GPU, and the Pi wins by a factor of six. What the pay-back time of the computer+GPU is compared to the Pi will depend on the local electricity cost.


. . I am running a 1050ti in a Core2Duo under Linux with Cuda 8.0. It draws typically 115W at full crunch and completes Blc01 tasks in 4.8 mins and Blc16 tasks in 5.9 mins. This comes to 5.9 * 115 / 60 or 11.3 w/Hrs per task ...

. . I am not interested in any Pi thank you :)

Stephen

:)
ID: 1945156 · Report as offensive
Profile Shaggie76
Avatar

Send message
Joined: 9 Oct 09
Posts: 282
Credit: 271,858,118
RAC: 196
Canada
Message 1945872 - Posted: 22 Jul 2018, 20:08:10 UTC

I collected data for the last two weeks and generated a new set of graphs. Interestingly the average credit/hour for all cards is lower -- maybe we're getting more VLARs these days? My data has long shown that non-vlar Arecibo work units were generally more profitable but it's curious to see how this doesn't seem to favor any particular card. I also noticed that there's much more variance in the Vega data -- I'm wondering if there is ambiguity in the device name returned and as far as the script can tell there's only one type when in fact there's a family of different variants?

ID: 1945872 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13833
Credit: 208,696,464
RAC: 304
Australia
Message 1945920 - Posted: 23 Jul 2018, 6:07:46 UTC

Thanks for the update. Always makes for interesting viewing.
Grant
Darwin NT
ID: 1945920 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1946514 - Posted: 27 Jul 2018, 15:43:43 UTC
Last modified: 27 Jul 2018, 15:44:07 UTC

i'm considering trading up 2 of my systems from dual 1060's (3gb) to single 1080ti's

looks like a single 1080ti will do about the same or maybe a little more work than 2x 1060's, is that right? I'm using the lunatics optimizations currently with some pretty aggressive command line options, and will probably go more aggressive on the 1080ti's

thoughts?
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1946514 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1946516 - Posted: 27 Jul 2018, 15:54:45 UTC - in response to Message 1946514.  

i'm considering trading up 2 of my systems from dual 1060's (3gb) to single 1080ti's

looks like a single 1080ti will do about the same or maybe a little more work than 2x 1060's, is that right? I'm using the lunatics optimizations currently with some pretty aggressive command line options, and will probably go more aggressive on the 1080ti's

thoughts?


If on Linux, then your 1080Ti will do guppi vlars in 48 seconds with latest software.
http://setiathome.berkeley.edu/workunit.php?wuid=3069924624
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1946516 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1946517 - Posted: 27 Jul 2018, 16:15:57 UTC - in response to Message 1946514.  
Last modified: 27 Jul 2018, 16:16:46 UTC

looks like a single 1080ti will do about the same or maybe a little more work than 2x 1060's


Yes without having to change OS
ID: 1946517 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1946536 - Posted: 27 Jul 2018, 18:08:32 UTC - in response to Message 1946517.  

looks like a single 1080ti will do about the same or maybe a little more work than 2x 1060's


Yes without having to change OS


you mean staying on what they currently run? windows 7 + lunatics.

and even more work done if i switch to linux + special app
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1946536 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1946539 - Posted: 27 Jul 2018, 18:20:01 UTC - in response to Message 1946536.  

Running Windows 7 and 1080Ti you could run 3 at a time and get better output than 2 1060s.

If you want to learn Linux and handhold it everyday then yes, you could install Linux, install the special cuda app and babysit it for faster turn around time. (I'm more for install and forget, let it run on it's own)
ID: 1946539 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1946541 - Posted: 27 Jul 2018, 18:40:04 UTC - in response to Message 1946539.  
Last modified: 27 Jul 2018, 18:42:28 UTC

why do you think it needs babysitting?

I have 2 other systems running linux + special app and they are super stable. i'm already fairly familiar with linux/unix. setting it up is as easy as dropping files in the right directory.

one runs dual 1050ti's
and the other one has 2x 750ti and a single 1060.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1946541 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1946543 - Posted: 27 Jul 2018, 18:57:46 UTC - in response to Message 1946541.  

Oh I don't know....

Maybe the (at least) 5 threads with anywhere from 80-100 pages relating to issues with installing, fixing, reinstalling, finding commands....etc that has been going on for at least 1.5 years here.....

Every time someone updates a driver or kernal, there is the rush to try to find out what was broken and get the machines to work again.

If I had a week off I could learn Linux but I rather spend that week on a beach with frozen drinks watching women in bikinis.

So, until someone comes up with a simple installation program that installs it, installs the cuda and prevents any updates that break it, it's not for a novice cruncher. The dozen or so people running it here have spent long, long hours trying to work out the bugs. And as you point out, you are familiar with it. As are most of those dozen people running.

I approach seti as a person who wants to contribute but doesn't have a lot of familiarity with computing. For them, telling them to install linux isn't fair unless you are playing to go to their houses to install it and return every time it breaks.


Z
ID: 1946543 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1946544 - Posted: 27 Jul 2018, 19:13:59 UTC

Hey, Z-Man. I'm back crunching after two days of no A/C and no crunching. Just before I lost the A/C I had my only Windows computer decide it didn't want to boot Windows anymore. Couldn't recover it after a day of trying. So I said goodbye to Windows once and for all and installed the current Ubuntu LTS. I was up and running in ten minutes for the OS. I was up and running BOINC in 30 minutes with just some app_config and cc_config files to get copied over from the other machines. I've never had to write or edit an app_info file and just run the stock provided one. As Ian stated it is literally drag and drop files into the BOINC folder on the Desktop. I have not had a single one of my Linux machines blow up, ever. I never could get the Windows machines to run longer than a couple of days before a Windows or Defender update trip it up. I have had my Linux machines run for weeks on end without ever looking at them.

So it is not any harder to get Linux/Boinc installed than Windows/Boinc. The added benefit of running the special app is just the cherry on the top. My $0.02.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1946544 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1946545 - Posted: 27 Jul 2018, 19:34:56 UTC - in response to Message 1946543.  

Oh I don't know....

Maybe the (at least) 5 threads with anywhere from 80-100 pages relating to issues with installing, fixing, reinstalling, finding commands....etc that has been going on for at least 1.5 years here.....

Every time someone updates a driver or kernal, there is the rush to try to find out what was broken and get the machines to work again.

If I had a week off I could learn Linux but I rather spend that week on a beach with frozen drinks watching women in bikinis.

So, until someone comes up with a simple installation program that installs it, installs the cuda and prevents any updates that break it, it's not for a novice cruncher. The dozen or so people running it here have spent long, long hours trying to work out the bugs. And as you point out, you are familiar with it. As are most of those dozen people running.

I approach seti as a person who wants to contribute but doesn't have a lot of familiarity with computing. For them, telling them to install linux isn't fair unless you are playing to go to their houses to install it and return every time it breaks.


Z


those all sound like issues of 2 kinds of people.

1. people who dont know how to use linux, or understand how things work differently than windows.
2. people who fix stuff that isn't broken.

like i said, my linux boxes are set it and forget it. once they are up and running, they can stay up without touching it, and that's usually the selling point of linux. windows is the thing that seems to have stability issues. automatic updates breaking things, needing to roll back, etc.

it's ok if you dont have knowledge about linux. i agree that windows is definitely easier to use for the end user. but to say that linux requires more babysitting is incorrect.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1946545 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1946547 - Posted: 27 Jul 2018, 19:42:27 UTC - in response to Message 1946544.  

Not had a problem with windows trying to update anything once I handicapped it by removing all the updates that made it try and update the computer. Usually the first thing I do after clean install. Don't use Defeneder but Norton but even that doesn't bother the machines.
ID: 1946547 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1946550 - Posted: 27 Jul 2018, 20:17:50 UTC - in response to Message 1946547.  

I had Windows set for no updates of drivers. Only thing I allowed was security updates and Defender definition updates. Still seems every week there was a update that left me with a rebooted machine with BOINC not running. The last straw was a reboot that left me with the inability to load Windows. I tried to go back to a restore point. Said it was doing it and then said it was unsuccessful. Tried to use the recovery mechanism. Said it was doing it and then was unsuccessful. Tried recovery again and tried to use the restore Windows while leaving your files and apps alone. That is what I had to do a couple of months ago. Microsoft lied. The recovery deleted all my apps and BOINC. I had to reinstall BOINC and Lunatics last time and all my utilities. Figured I would just do the same again. Windows couldn't recover again after many minutes of disk activity. Took out the original install media. Booted it and tried recovery from the USB stick. Again much disk activity and again failed to recover. So I figured if I had to start from complete scratch and completely wipe the the disk to reinstall a virgin Windows 10 and then suffer through endless hours of 3 major platform updates since release I decided I might as well say I was done with Windows. As I stated, booted my Ubuntu 18.04 USB stick and installed a fully working Linux platform in ten minutes.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1946550 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 20 · Next

Message boards : Number crunching : GPU FLOPS: Theory vs Reality


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.