mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-02-04, 22:04   #111
masser
 
masser's Avatar
 
Jul 2003
Behind BB

176710 Posts
Default

Quote:
Originally Posted by nomead View Post
That is only valid if you're building a system just for that purpose, not upgrading some pre-existing one (like I did - GT430 out, RTX2060 in - on an old 6-core Phenom system from 2011...)
Nice!
masser is offline   Reply With Quote
Old 2019-02-05, 22:27   #112
dragonbud20
 
dragonbud20's Avatar
 
Mar 2014

24×5 Posts
Default

Quote:
Originally Posted by nomead View Post
That is only valid if you're building a system just for that purpose, not upgrading some pre-existing one (like I did - GT430 out, RTX2060 in - on an old 6-core Phenom system from 2011...)
still an opportunity cost that you lose because you could be using the same empty slot for more throughput and to attain the same throughput as a 2080 you'd need to spend more for just the slots.
dragonbud20 is offline   Reply With Quote
Old 2019-02-06, 03:39   #113
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

317 Posts
Default

Quote:
Originally Posted by dragonbud20 View Post
still an opportunity cost that you lose because you could be using the same empty slot for more throughput and to attain the same throughput as a 2080 you'd need to spend more for just the slots.
Of course. But then there are a couple more limiting factors to consider.

At the moment, I can't afford to get a 2080 card for myself, but the 2060 was within the budget. Actually even it was a bit expensive compared to what I would have otherwise bought, but a major factor (pun intended) was the mfaktc performance.

Also there are rumours about the soon-to-be-released GTX 1650, 1660 and 1660 Ti. Even less price, but also less CUDA cores. The 1660 Ti, if it has 1536 cores at $280 as rumoured, is about the same price per core as the 2060. The 1660, with 1280 cores and slower memory at $230 is not such a good deal anymore. Not much details on the 1650 yet.

With a bigger budget, why settle for a 2080, why not get a 2080 Ti instead... and so on.

Another thing was the old case the system was housed in. There are many hard disks inside it, and longer cards wouldn't fit in. Even the shorter 2080 cards seem to be 268 mm. This wouldn't fit in without rearranging the disks or removing a couple, or buying a new case. The Palit 2060 card I got was one of the longer models at 235 mm, and it was already a bit difficult to fit it in, but I didn't need to rearrange anything permanently. And there are some really short models meant for HTPC / Mini-ITX builds that are just 168 mm, but I'd assume the cooling suffers a bit there.
nomead is offline   Reply With Quote
Old 2019-02-16, 01:14   #114
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

317 Posts
Default

Quote:
Originally Posted by nomead View Post
I don't have a Windows build environment yet, but maybe now there's some motivation for me to set one up.
Well... like I posted in the mfaktc thread, I now have a working build environment on Windows as well. Although it somehow feels like a miracle or accident that it works at all. But anyway, now I have Windows 64-bit executables in both more and less classes flavour, where the GPUSieveSize can be set all the way up to 2047. The same as on Linux, 2048 won't work anymore. Still, going from 128 to 2047 seems to give about 5-7% extra performance on an RTX 2060. It is slightly less than on that RTX 2080. I can only guess why. On the 2080, GPU utilization seemed to be about 94% before the change, but on the 2060 it was about 95-96%. So maybe there's less of that unused reserve to draw upon.

The Makefile had code generation enabled for CUDA Compute Capability versions 5.0, 5.2 (Maxwell, GTX9xx - is there much difference between 5.0 and 5.2?), 6.0 (Pascal, GTX10xx), 7.0 (Volta, mostly just Titan V) and 7.5 (Turing, RTX20xx). I'm guessing that the gains on anything before 7.0 / Volta will be relatively small, for architectural reasons.

So, does anyone feel brave (or foolish) enough to try out random exe files off the Internet? Where should I put the zip file, if anyone wants it?
nomead is offline   Reply With Quote
Old 2019-02-16, 01:40   #115
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

10111001001002 Posts
Default

Quote:
Originally Posted by nomead View Post
Where should I put the zip file, if anyone wants it?
Congrats on beating the build gremlins into submission or at least temporary retreat. For gpuowl, I've posted zip files of Windows builds and makefiles as attachments to thread posts (See "Manage attachments", a ways below the composition or edit window for a thread post). Another place to put executables is on James Heinrich's mirror site. Another way to go is to post the complete process of setting up a build environment and the make file.

Last fiddled with by kriesel on 2019-02-16 at 01:44
kriesel is offline   Reply With Quote
Old 2019-02-16, 01:46   #116
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

317 Posts
Default

Ok the path of least effort is probably as an attachment, then. So, here it is.
Attached Files
File Type: zip 2047-bin.zip (815.2 KB, 129 views)
nomead is offline   Reply With Quote
Old 2019-02-16, 18:25   #117
Chuck
 
Chuck's Avatar
 
May 2011
Orange Park, FL

2·449 Posts
Default

Need to put the runtime dlls into the zip file too.
Chuck is offline   Reply With Quote
Old 2019-02-16, 18:34   #118
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

317 Posts
Default

Really? I don't have that cudart64_100.dll in the running directory of mfaktc at all. Anyway, it is included in mfaktc-0.21.win.cuda100.zip (the official binary distribution package) but not in the extra versions package, so I decided to follow that and only include the executable files.

Packages of course available at https://www.mersenneforum.org/mfaktc/mfaktc-0.21/

Or is some other DLL missing?
nomead is offline   Reply With Quote
Old 2019-02-16, 19:03   #119
Chuck
 
Chuck's Avatar
 
May 2011
Orange Park, FL

2×449 Posts
Default

Thanks, I forgot it was located there.
Chuck is offline   Reply With Quote
Old 2019-02-16, 21:39   #120
Chuck
 
Chuck's Avatar
 
May 2011
Orange Park, FL

2×449 Posts
Default

This is giving at least a 5% improvement on a GTX 1080 Ti
Chuck is offline   Reply With Quote
Old 2019-02-17, 08:26   #121
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

317 Posts
Default

Quote:
Originally Posted by Chuck View Post
This is giving at least a 5% improvement on a GTX 1080 Ti
Ok, good to know that 1) it works, and 2) it makes a difference. Thanks for testing!
nomead is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Nvidia GTX 745 4GB ??? petrw1 GPU Computing 3 2016-08-02 15:23
Nvidia Pascal, a third of DP firejuggler GPU Computing 12 2016-02-23 06:55
AMD + Nvidia TheMawn GPU Computing 7 2013-07-01 14:08
Nvidia Kepler Brain GPU Computing 149 2013-02-17 08:05
What can I do with my nvidia GPU? Surge Software 4 2010-09-29 11:36

All times are UTC. The time now is 09:11.


Sun Dec 5 09:11:45 UTC 2021 up 135 days, 3:40, 0 users, load averages: 1.98, 1.99, 1.78

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.