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 . . . 4 · 5 · 6 · 7 · 8 · 9 · 10 . . . 20 · Next

AuthorMessage
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1956923 - Posted: 23 Sep 2018, 14:16:15 UTC

Can someone point me to more information that is reliable about the newest AMD cards? I think I am talking about the Vega series.

I am trying to wrap my head around the differences in terminology and mapping the NVidia "CUDA" numbers onto the equivalent under the AMD architecture.

Are there any other resources besides Shaggies analysis that would allow me to guestimate the kind of Seti production something like a Vega 8 gpu which is also part of an AMD CPU has?

Thank you,
Tom
A proud member of the OFA (Old Farts Association).
ID: 1956923 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956946 - Posted: 23 Sep 2018, 16:10:03 UTC - in response to Message 1956923.  

I usually find the best in depth analysis of cpu and gpu architecture done by the writers at AnandTech. You will find the cpu and gpu roadmaps of the various vendors there.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956946 · 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 1956961 - Posted: 23 Sep 2018, 17:01:20 UTC

Tom, also i believe Shaggie's data only considered the standard SETI apps, and not special stuff like petri's app.

anything AMD wont work on petri' s app, so you'd have to run SoG. also amd cards in general, and especially the Vega series are very power hungry.

if you're ok running linux and want the best performance, i'd just stick with nvidia cards.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1956961 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1957011 - Posted: 23 Sep 2018, 22:42:05 UTC - in response to Message 1956961.  

Tom, also i believe Shaggie's data only considered the standard SETI apps, and not special stuff like petri's app.

anything AMD won't work on petri' s app, so you'd have to run SoG. also, AMD cards in general, and especially the Vega series are very power hungry.

if you're ok running Linux and want the best performance, I'd just stick with Nvidia cards.


The reason I am even looking at it is there is another thread where someone has an "AMD Ryzen 2200G CPU" that started out crashing his system. That problem "cured" itself about a week later with no changes (I think) on the operators part. We think that either Win10 updated or the OS got another device driver.

So I have been wondering what kind of performance we could expect from a new version of the AIO (All in One) from AMD.

So far, it is looking like the CPU/gpu throttling I previously experienced on an A-10 5700 is still happening but much less severely. GPU Z's database says a Gtx 750Ti runs at 179% of a Vega 8. If I could get an idea of how much less speed a AMD task processes at vs a SOG task I could support or refute my suspicion that it "oughta" be running at maybe a half per task on the gpu.

While I have the noisy itch to see what a Vega 64 would do (apparently its a power hungry Gtx 1070) I am trying to make a policy of not spending any more hardware money that doesn't go towards the Threadripper2 project. And the hard part is to "stick to it".

I just saw a used high gpu Miner MB (19 gpus?) for $79 on eBay and almost, almost bought it :(

Tom
A proud member of the OFA (Old Farts Association).
ID: 1957011 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1957021 - Posted: 23 Sep 2018, 23:46:58 UTC - in response to Message 1956946.  

I usually find the best in depth analysis of cpu and gpu architecture done by the writers at AnandTech. You will find the cpu and gpu roadmaps of the various vendors there.


Thank you.
A proud member of the OFA (Old Farts Association).
ID: 1957021 · Report as offensive
Profile Shaggie76
Avatar

Send message
Joined: 9 Oct 09
Posts: 282
Credit: 271,858,118
RAC: 196
Canada
Message 1958702 - Posted: 6 Oct 2018, 0:15:55 UTC

I did a scan over night and there are about a dozen RTX cards in the results but few have enough OpenCL results yet to get anything but a preliminary picture.

There is host 8587521 which may have been running another card previously -- I don't check the details of each task to avoid overloading the server -- so taken with the largest grain of salt possible here's some preliminary results for a GeForce RTX 2080 (OpenCL).

     955 Credit / Hour
     95% Core / Task
     122 Tasks

I'll have the rest of my usual scan shortly.
ID: 1958702 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1958705 - Posted: 6 Oct 2018, 0:33:31 UTC - in response to Message 1958702.  

There is host 8587521 which may have been running another card previously 8587521

Nah, that's a new system, created 30/9/2018.
It hasn't been running for even a week, so another 6 or more to go before it's RAC will stabilise (as much as it ever does).
Grant
Darwin NT
ID: 1958705 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958707 - Posted: 6 Oct 2018, 0:39:54 UTC - in response to Message 1958705.  

I just looked for a task of standard AR(0.40-44) and found one from today's 27Sept Arecibo tape and then looked for a 1080Ti running the SoG app and also had done a standard AR task. The 2080 is 30 seconds faster with default SoG tunings while the 1080Ti ran SoG with maxed out SoG tunings.
Host 8587521

Host 8535487
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958707 · Report as offensive
Profile Shaggie76
Avatar

Send message
Joined: 9 Oct 09
Posts: 282
Credit: 271,858,118
RAC: 196
Canada
Message 1958708 - Posted: 6 Oct 2018, 0:40:09 UTC

New this week: there seems to be enough Radeon Pro Vega 64 in circulation now to show up on the charts but sadly it doesn't seem to be any faster than the Vega 56s. The Radeon RX550 edges out the efficiency bracket: maybe the mix of work units or maybe their drivers have been optimized -- who knows?

ID: 1958708 · Report as offensive
Profile Shaggie76
Avatar

Send message
Joined: 9 Oct 09
Posts: 282
Credit: 271,858,118
RAC: 196
Canada
Message 1958711 - Posted: 6 Oct 2018, 0:52:41 UTC - in response to Message 1958705.  

There is host 8587521 which may have been running another card previously 8587521

Nah, that's a new system, created 30/9/2018.
It hasn't been running for even a week, so another 6 or more to go before it's RAC will stabilise (as much as it ever does).

To be clear: my script doesn't use RAC -- it digests all completed tasks. A larger sample size will smooth out averages but after 122 tasks I would expect it to be pretty close.
ID: 1958711 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1958722 - Posted: 6 Oct 2018, 1:14:36 UTC - in response to Message 1958711.  
Last modified: 6 Oct 2018, 1:22:40 UTC

To be clear: my script doesn't use RAC -- it digests all completed tasks. A larger sample size will smooth out averages but after 122 tasks I would expect it to be pretty close.

Interesting.
So, that being the case the RTX 2080 is outperforming the GTX 1080Tis, by quite a bit.
995 Credit/hr v 800 C/hr or so.
Very impressive.


It's also interesting to compare the very first chart with the current one for the amount of Credits/hour.
Pretty sure the first one was when there was plenty of Arecibo work about, the current one mostly GBT (and longer running tasks at that).
Grant
Darwin NT
ID: 1958722 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958725 - Posted: 6 Oct 2018, 1:41:05 UTC - in response to Message 1958722.  

Yes, that is impressive. That goes along with the graphics test reviews that the 2080 is the equivalent/very slightly faster than the 1080Ti. And matches up with those SoG Arecibo standard AR tasks I linked. Will be interesting to see where the RTX 2080Ti falls out in the chart once it has appeared for a few weeks. I expect those to show up in the projects in another week or so.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958725 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958747 - Posted: 6 Oct 2018, 3:46:42 UTC

Looks like his host has gone offline or moved on to other projects since no more tasks have reported since this afternoon. Was looking again through the validated tasks and noticed this one with a BLC22 VLAR with the wingman on a 1080Ti with default SoG tunings.

Workunit 3164456249

The 2080 was a minute~ faster than the 1080Ti.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958747 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1958759 - Posted: 6 Oct 2018, 5:16:45 UTC - in response to Message 1958747.  

The 2080 was a minute~ faster than the 1080Ti.

Very impressive.
And given how much faster the GTX 1080Ti is compared to the GTX 1080, the RTX 2080Ti will most likely qualify as OMG!
Will be very interested to the results of the Special Application on one of those cards.
Grant
Darwin NT
ID: 1958759 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1958814 - Posted: 6 Oct 2018, 12:48:47 UTC

https://www.techpowerup.com/gpu-specs/radeon-rx-vega-64.c2871

If in "theory" (see above) the Vega 64 is the equivalent of a gtx 1070 then a question is why in practice it isn't performing that well.

I am wondering if it might not be an artifac of the performance of the gpu task software?

I think I understand that Shaggies report is basically SOG (Nvidia) vs. the three odd "opencl_ati5_nocal" (AMD) type apps.

I know there has been publically visable development efforts on the the Nvidia side. I wonder, do we have anyone working on the AMD side to enhance AMD gpu software?

Thank you for the report(s) Shaggie!

Tom
A proud member of the OFA (Old Farts Association).
ID: 1958814 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1958849 - Posted: 6 Oct 2018, 15:56:55 UTC - in response to Message 1958814.  

I'm sure that Vega cards are NOT running "opencl_ati5_nocal" applications. They would be running the same exact SoG r3584 code that the Nvidia cards run. Same source code. Just compiled for ATI hardware versus Nvidia hardware. OpenCL is platform neutral.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1958849 · 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 1958869 - Posted: 6 Oct 2018, 18:05:53 UTC

definitely interested to see the the results once someone can get an RTX card on the special app.

but you have to consider cost.

you can get a used 1080ti for under $500 including shipping. (i've personally purchased 5x 1080ti in the last 2 months, 2 for $490 shipped and 3 for $475 shipped, EVGA SC2 models).

the RTX 2080 is over $800. a black friday deal might make things easier to swallow.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1958869 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1958914 - Posted: 6 Oct 2018, 21:44:37 UTC - in response to Message 1958869.  

the RTX 2080 is over $800. a black friday deal might make things easier to swallow.

If i could get a GTX 1080Ti for $800 i'd buy one, they're still around $1,200 here.
RTX 2080Tis start at $2,200 and go up from there.
Grant
Darwin NT
ID: 1958914 · 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 1958946 - Posted: 6 Oct 2018, 23:58:36 UTC
Last modified: 7 Oct 2018, 0:03:00 UTC

$1200 in Aussie bucks is about $850 Freedom bucks. :P
$2200 is about $1550 USD. Definitely higher than the most expensive models here.

But I hope that’s new prices. The new 1080tis are about $700 new here, but it can vary by model and retailer like anything else.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1958946 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1958956 - Posted: 7 Oct 2018, 0:24:07 UTC - in response to Message 1958946.  
Last modified: 7 Oct 2018, 0:26:16 UTC

$1200 in Aussie bucks is about $850 Freedom bucks. :P
$2200 is about $1550 USD. Definitely higher than the most expensive models here.

Yep.
We've always been screwed when it comes to pricing of overseas goods, particularly electronics.


But I hope that’s new prices. The new 1080tis are about $700 new here, but it can vary by model and retailer like anything else.

Yep, new models. There are some a bit over $1,000, others still over $1,600.
$1,200 is the middle of the range going rate.
Grant
Darwin NT
ID: 1958956 · Report as offensive
Previous · 1 . . . 4 · 5 · 6 · 7 · 8 · 9 · 10 . . . 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.