mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2020-08-08, 08:48   #320
Cheetahgod
 
May 2020

23·3 Posts
Default

Is it possible to run PRP tests on a nvidia Gpu? Radeon VII on ebay are going for ridiculous prices. Hopefully the new Big Navi 2 Amd cards will have support for OpenCl.
Cheetahgod is offline   Reply With Quote
Old 2020-08-08, 10:06   #321
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

3·11·19 Posts
Default

Quote:
Originally Posted by Cheetahgod View Post
Is it possible to run PRP tests on a nvidia Gpu? Radeon VII on ebay are going for ridiculous prices. Hopefully the new Big Navi 2 Amd cards will have support for OpenCl.
Yes, at least some nvidia cards can run PRP tests using gpuowl. There are Radeon VII at ridiculous prices on ebay but I doubt they're selling, the ones I see now are the same unabashed speculation that have been sitting there for months unsold. Big Navi probably won't have timely OpenCL support, small Navi still isn't supported on ROCm and it's still unclear to me if small Navi has OpenCL support on the non-pro cards. A Radeon VII will probably beat Big Navi at PRP, RDNA/2 has gaming characteristics much like the turing nvidia cards which means they're probably better suited for TF. I hope to be proven wrong and as big navi is still poorly defined there's a reasonable chance I am.
M344587487 is offline   Reply With Quote
Old 2020-08-08, 10:07   #322
Viliam Furik
 
Jul 2018
Martin, Slovakia

3×43 Posts
Default

Yes, it is! Most nVidia cards throughout the history support OpenCL. If you have RTX 20XX, you are good to go.

My RTX 2080 Ti has about the same throughput in gpuOwl 90M test as my Ryzen 9 3900X (which is pretty fast on its own because of the huge amount of cache. In numbers, it is about 2.3 ms/iter.
Viliam Furik is online now   Reply With Quote
Old 2020-08-09, 18:29   #323
Cheetahgod
 
May 2020

23·3 Posts
Default

Just saw a used radeon 7 sell on ebay for $860
Cheetahgod is offline   Reply With Quote
Old 2020-08-09, 19:18   #324
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

114516 Posts
Default

Quote:
Originally Posted by Cheetahgod View Post
Is it possible to run PRP tests on an nvidia Gpu?
Definitely possible, on a sufficiently recent model. GTX1050Ti and newer work since gpuowl v6.5, and some older gpus will too. Not ancient Quadro 2000 or 4000 though. Those can run LL DC with CUDALucas, or do TF with mfaktc.

https://www.mersenneforum.org/showth...076#post489076
kriesel is online now   Reply With Quote
Old 2020-08-10, 21:05   #325
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
República de California

2·13·443 Posts
Default

As noted in post #307, on 3 Aug I ordered 2 twin-packs of 2x-SATA-to-6-pin-PCIe adapters for my intended Frankencable scheme for combining power from a pair of the Corsair 850RT's Peripheral&SATA 6-pin power-outs to drive a fourth Radeon VII. At order time seller was alas down to "1 left in stock", so ETA was 2 weeks from then, Aug 17-19. Well, one of the 2-packs - clearly the "in stock" one - arrived yesterday, so while ETA for the 2nd remains a week+ from now, I figured I'd at least try a basic does-this-work-and-is-it-stable test of the scheme. Couple hours ago hooked the pair of 2 adapters to 4 SATA-cable 15-pin plugs (all along a single cable hooked into 1 of the aforementioned power-outs), those combined into a single 6-pin PCIe male-plug at the output ends of the adapters, which I plugged into the pair of 6-pin PCIe female-plug inputs of one of the 6-pin-to-8-pin-PCIe Y-adapters which shipped with my one and only bought-new R7. Plugged the 8-pin PCIe male-plug at the output end of that final power-recombining adapter into one of the two 8-pin PCIe power ports on one of the 3 R7s the system is currently driving, with the other power ports still being fed by one output of a standard 8-pin PCIe 12V power cable hooked into the PSU. I used the R7 farthest from the PSU, which also happens to be one of the 2 plugged into a full-width 16X PCIe plug on the mobo. (Card 3 is hooked into a 1X and gets its 16X PCIe connection via powered-riser card), figuring that the up-to-75W the GPU can draw via the full-width PCIe bus would reduce the power draw on the 2 external plugs, thus minimizing the wattage needing to flow through the Frankencable.

System has been running stably, more or less same total-wattage-at-wall as before, ever since. Gonna let it run 24 hours, if still stable, will try hooking up GPU #4 via standard 8-pin PCIe 12V power cable hooked into the PSU, in hopes that having replaced power to one of the eight total 8-pin PCIe power ports on the 4 cards will be just enough load alleviation on the 12V portion of the PSU power bus to allow it to support a suitably-downclocked set of 4 R7s. If not, I'll have to wait for the 2nd twin-pack of SATA 2x-to-1 adapters to retry.

Last fiddled with by ewmayer on 2020-08-10 at 21:05
ewmayer is offline   Reply With Quote
Old 2020-08-10, 21:12   #326
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

931010 Posts
Default

Quote:
Originally Posted by ewmayer View Post
If not, I'll have to wait for the 2nd twin-pack of SATA 2x-to-1 adapters to retry.
And those run from the PSU directly to the cables to power the GPUs?

Do you have a datastream from the PSU, or only the wall?
chalsall is offline   Reply With Quote
Old 2020-08-10, 21:35   #327
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
República de California

2·13·443 Posts
Default

Quote:
Originally Posted by chalsall View Post
And those run from the PSU directly to the cables to power the GPUs?

Do you have a datastream from the PSU, or only the wall?
See post #309 for the precise setup.

In addition to a wattmeter-at-wall-plug, I have per-GPU wattages as output by the ROCm software, but those appear to understate the actual draw - currently the ROCm data show the 3 cards using ~500W total. I'm seeing 750W-at-wall, and after boot-up but prior to firing up jobs on the GPUs I see ~100W-at-wall. So ~150W of the TDP are missing from the ROCm output. If I manage to run a 4th card via the Frankencable scheme I'll downclock one of the cards further, from the current sclk=3 to 2. That should keep the TDP to ~900W. I've run this PSU for weeks at its rated 850W so I feel OK with just 50W more, but over 900W would really be pushing things.
ewmayer is offline   Reply With Quote
Old 2020-08-10, 21:45   #328
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×5×72×19 Posts
Default

Quote:
Originally Posted by ewmayer View Post
That should keep the TDP to ~900W. I've run this PSU for weeks at its rated 850W so I feel OK with just 50W more, but over 900W would really be pushing things.
Please forgive me if I'm telling you how to chew gum, but...

I've found value in instrumenting (and reading at high temporal resolution) everything possible.

Instruments can sometimes give inaccurate readings. Cross-examination of the data streams can be useful.

I am derived from a family of pilots and engineers. Intense is almost the entire job description.
chalsall is offline   Reply With Quote
Old 2020-08-10, 23:02   #329
PhilF
 
PhilF's Avatar
 
Feb 2005
Colorado

3·132 Posts
Default

Quote:
Originally Posted by ewmayer View Post
Couple hours ago hooked the pair of 2 adapters to 4 SATA-cable 15-pin plugs

Gonna let it run 24 hours
Do me a favor and feel each of those flat 15-pin power connectors before you go to bed. If any are even warm, it means this scheme isn't going to last in the long term anyway. The heat will prematurely oxidize the contacts, increasing their resistance, which adds more heat, etc. So if any are warm you might as well shut it down. We'll both sleep better.
PhilF is online now   Reply With Quote
Old 2020-08-11, 00:01   #330
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
República de California

1151810 Posts
Default

Quote:
Originally Posted by PhilF View Post
Do me a favor and feel each of those flat 15-pin power connectors before you go to bed. If any are even warm, it means this scheme isn't going to last in the long term anyway. The heat will prematurely oxidize the contacts, increasing their resistance, which adds more heat, etc. So if any are warm you might as well shut it down. We'll both sleep better.
All feel cool to the touch - or better, not appreciably warmer than ambient. However, I don't know how much wattage the Frankencable-supplied 8-pin PIe power plog is carrying, compared to its conventionally-cabled neighbor. I figured using just a single Frankencabled plug would be the safest way to start, since if said cable is having any excess-resistance issues, it would [a] be spottable via a warm cable and [b] naturally lead to the other supply cable taking up more of the load. I suspect both-plugs-via-Frankencable will be a stiffer test in a week, though if things get warm in that scenario I have the option of splitting power to both GPUs 3 and 4 the same way as GPU 3 at present, one plug conventional 8-pin PCIe power, the other Frankencable. Are there reasonably cheap mini-watt-meters one can insert in between an 8-pin PCIe power plug and its input receptacle?
ewmayer is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
AMD Radeon Pro WX 3200 ET_ GPU Computing 1 2019-07-04 11:02
Radeon Pro Vega II Duo (look at this monster) M344587487 GPU Computing 10 2019-06-18 14:00
What's the best project to run on a Radeon RX 480? jasong GPU Computing 0 2016-11-09 04:32
Radeon Pro Duo 0PolarBearsHere GPU Computing 0 2016-03-15 01:32
AMD Radeon R9 295X2 firejuggler GPU Computing 33 2014-09-03 21:42

All times are UTC. The time now is 17:34.

Tue Sep 22 17:34:55 UTC 2020 up 12 days, 14:45, 1 user, load averages: 3.13, 2.46, 2.03

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.