mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-12-19, 18:42   #2641
moebius
 
moebius's Avatar
 
Jul 2009
Germany

547 Posts
Default

Quote:
Originally Posted by moebius View Post
Is gpuowl able to calculate together on the two GK210 chips of the K80, or should two instances have roughly the same throughput? Each chip seems to be able to use 12 GB of memory! The above benchmarks are for one instance.
I think gpuowl only uses one of the 2 GPU's, otherwise it should indeed be twice as fast as a K-40.My own mistake in thinking.
moebius is offline   Reply With Quote
Old 2020-12-19, 19:18   #2642
moebius
 
moebius's Avatar
 
Jul 2009
Germany

547 Posts
Default

Unfortunately completely underrated in the list on mersenne.ca

Asus ROG STRIX RX VEGA64 O8G GAMING
Code:
2020-12-19 20:10:53 config: -prp 57885161 
2020-12-19 20:10:53 device 0, unique id ''
2020-12-19 20:10:53 AMD_RXVega64 57885161 FFT: 3M 1K:6:256 (18.40 bpw)
2020-12-19 20:10:53 AMD_RXVega64 Expected maximum carry32: 42500000
2020-12-19 20:10:54 AMD_RXVega64 OpenCL args "-DEXP=57885161u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=6u -DPM1=0 -DAMDGPU=1 -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0x8.3b39c2879b8p-4 -DIWEIGHT_STEP_MINUS_1=-0xa.decf1cf0a51e8p-5  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-12-19 20:10:54 AMD_RXVega64 ASM compilation failed, retrying compilation using NO_ASM
2020-12-19 20:10:58 AMD_RXVega64 OpenCL compilation in 4.29 s
2020-12-19 20:10:59 AMD_RXVega64 57885161 OK        0 loaded: blockSize 400, 0000000000000003
2020-12-19 20:10:59 AMD_RXVega64 validating proof residues for power 8
2020-12-19 20:10:59 AMD_RXVega64 Proof using power 8
2020-12-19 20:11:00 AMD_RXVega64 57885161 OK      800   0.00%;  827 us/it; ETA 0d 13:18; 5727fe6a7225c273 (check 0.39s)
2020-12-19 20:13:47 AMD_RXVega64 57885161 OK   200000   0.35%;  838 us/it; ETA 0d 13:26; de62d6db1ad5092d (check 0.40s)
2020-12-19 20:16:36 AMD_RXVega64 57885161 OK   400000   0.69%;  844 us/it; ETA 0d 13:29; 45e043b36f3556e1 (check 0.40s)
2020-12-19 20:16:39 AMD_RXVega64 Stopping, please wait..
Attached Thumbnails
Click image for larger version

Name:	vega64.gif
Views:	37
Size:	30.9 KB
ID:	23994   Click image for larger version

Name:	vega642.gif
Views:	44
Size:	22.6 KB
ID:	23995  

Last fiddled with by moebius on 2020-12-19 at 19:25
moebius is offline   Reply With Quote
Old 2020-12-20, 16:47   #2643
DrobinsonPE
 
Aug 2020

22×3×7 Posts
Default

Just for testing, I installed Windows on another computer to see if gpuowl and mfakto will run on it. It has been running mprime on Linux for a while now and will probably go back to that when I am finished getting data. Mfakto data will be posted in the other thread.

GB-BRi5H-8250, i508250U, UHD 620, 16GB DDR-4, SSD, Windows 10.

gpuowl V6.11-380

Code:
C:\Users\user\gpuowl\v611380>gpuowl-win -iters 200000 -prp 77936867
2020-12-19 21:13:44 gpuowl v6.11-380-g79ea0cc
2020-12-19 21:13:44 Intel(R) UHD Graphics 620-0 77936867 FFT: 4M 1K:8:256 (18.58 bpw)
2020-12-19 21:13:54 Intel(R) UHD Graphics 620-0 OpenCL compilation in 9.44 s
2020-12-19 21:14:10 Intel(R) UHD Graphics 620-0 77936867 OK        0 loaded: blockSize 400, 0000000000000003
2020-12-19 21:14:10 Intel(R) UHD Graphics 620-0 validating proof residues for power 8
2020-12-19 21:14:10 Intel(R) UHD Graphics 620-0 Proof using power 8
2020-12-19 21:14:54 Intel(R) UHD Graphics 620-0 77936867 OK      800   0.00%; 37240 us/it; ETA 33d 14:12; 1579c241dc63eca6 (check 15.06s)
2020-12-19 23:19:40 Intel(R) UHD Graphics 620-0 Stopping, please wait..
2020-12-19 23:20:10 Intel(R) UHD Graphics 620-0 77936867 OK   200000   0.26%; 37655 us/it; ETA 33d 21:06; f0b04b45b0855bd2 (check 15.12s)
2020-12-19 23:20:10 Intel(R) UHD Graphics 620-0 Exiting because "stop requested"
2020-12-19 23:20:10 Intel(R) UHD Graphics 620-0 Bye
gpuowl V7.2-21

Code:
C:\Users\user\gpuowl\v702021>gpuowl-win -log 10000 -iters 200000 -prp 77936867
2020-12-20 07:49:11 GpuOwl VERSION v7.2-21-g28dbf88
2020-12-20 07:49:11 Intel(R) UHD Graphics 620-0 77936867 FFT: 4M 1K:8:256 (18.58 bpw)
2020-12-20 07:49:20 Intel(R) UHD Graphics 620-0 77936867 OpenCL compilation in 8.79 s
2020-12-20 07:49:20 Intel(R) UHD Graphics 620-0 77936867 maxAlloc: 0.0 GB
2020-12-20 07:49:20 Intel(R) UHD Graphics 620-0 77936867 You should use -maxAlloc if your GPU has more than 4GB memory. See help '-h'
2020-12-20 07:49:20 Intel(R) UHD Graphics 620-0 77936867 P1(0) 0 bits
2020-12-20 07:49:20 Intel(R) UHD Graphics 620-0 77936867 PRP starting from beginning
2020-12-20 07:49:35 Intel(R) UHD Graphics 620-0 77936867 OK         0 on-load: blockSize 400, 0000000000000003
2020-12-20 07:49:35 Intel(R) UHD Graphics 620-0 77936867 validating proof residues for power 8
2020-12-20 07:49:35 Intel(R) UHD Graphics 620-0 77936867 Proof using power 8
2020-12-20 07:50:21 Intel(R) UHD Graphics 620-0 77936867 OK       800   0.00% 1579c241dc63eca6 37369 us/it + check 15.09s + save 0.18s; ETA 33d 17:00
2020-12-20 07:56:22 Intel(R) UHD Graphics 620-0 77936867 OK     10000   0.01% fc4f135f7cf4ad29 37613 us/it + check 15.15s + save 0.18s; ETA 33d 22:11
2020-12-20 08:02:56 Intel(R) UHD Graphics 620-0 77936867 OK     20000   0.03% 3cd1bd9d5e09cbc5 37904 us/it + check 15.16s + save 0.18s; ETA 34d 04:23
2020-12-20 08:09:28 Intel(R) UHD Graphics 620-0 77936867 OK     30000   0.04% c4e0ff35e3290d98 37649 us/it + check 15.16s + save 0.18s; ETA 33d 22:46
2020-12-20 08:09:58 Intel(R) UHD Graphics 620-0 77936867 Stopping, please wait..
2020-12-20 08:10:14 Intel(R) UHD Graphics 620-0 77936867 OK     30800   0.04% eb1faf892fb6b838 37670 us/it + check 15.17s + save 0.17s; ETA 33d 23:12
2020-12-20 08:10:14 Intel(R) UHD Graphics 620-0 Exiting because "stop requested"
2020-12-20 08:10:14 Intel(R) UHD Graphics 620-0 Bye
DrobinsonPE is offline   Reply With Quote
Old 2020-12-21, 14:15   #2644
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

28·3 Posts
Default

This in the readme of ROCm 4.0 about the MI100 caught my eye:
Quote:
Extended matrix core engine with Matrix Fused Multiply-Add (MFMA) for mixed-precision arithmetic and operates on KxN matrices (FP32, FP16, BF16, Int8)
FP32 MFMA sounds like a potentially very useful thing for gpuowl but maybe I'm just naive. Is this AMD playing catch-up with Nvidia's compute offerings or is it some new FP32 extension to ML functionality? Is it even a card feature or is it just a ROCm library that gpuowl has nothing to do with anyway? Even in the best case that it's a new game changing feature, it's still mostly academic being on a card we're unlikely to be able to use and not worth implementing specifically for.
M344587487 is online now   Reply With Quote
Old 2020-12-21, 19:04   #2645
Xyzzy
 
Xyzzy's Avatar
 
"Mike"
Aug 2002

24×499 Posts
Default

They don't list a price for a MI100. Any ideas how expensive they are? $25K?
Xyzzy is offline   Reply With Quote
Old 2020-12-21, 19:18   #2646
moebius
 
moebius's Avatar
 
Jul 2009
Germany

547 Posts
Default

Quote:
Originally Posted by Xyzzy View Post
They don't list a price for a MI100. Any ideas how expensive they are? $25K?
I heard that the price should be around 6.5K, but graphics card prices are currently rising.
moebius is offline   Reply With Quote
Old 2020-12-21, 23:15   #2647
Xyzzy
 
Xyzzy's Avatar
 
"Mike"
Aug 2002

24·499 Posts
Default

How many "R7 units" will an MI100 replace?

Maybe $6.5K isn't so crazy?
Xyzzy is offline   Reply With Quote
Old 2020-12-22, 01:35   #2648
moebius
 
moebius's Avatar
 
Jul 2009
Germany

547 Posts
Default

Quote:
Originally Posted by Xyzzy View Post
Maybe $6.5K isn't so crazy?
The prices at the partners are of course much higher. Dell e.g. wants to have at present $17379.

Last fiddled with by moebius on 2020-12-22 at 02:27
moebius is offline   Reply With Quote
Old 2020-12-22, 05:24   #2649
DrobinsonPE
 
Aug 2020

22·3·7 Posts
Default

GTX 1650 Super

Code:
C:\Users\user\gpuowl\v702021>gpuowl-win -log 10000 -iters 200000 -prp 77936867
2020-12-21 20:48:42 GpuOwl VERSION v7.2-21-g28dbf88
2020-12-21 20:48:42 GeForce GTX 1650 SUPER-0 77936867 FFT: 4M 1K:8:256 (18.58 bpw)
2020-12-21 20:48:45 GeForce GTX 1650 SUPER-0 77936867 OpenCL compilation in 2.51 s
2020-12-21 20:48:45 GeForce GTX 1650 SUPER-0 77936867 P1(0) 0 bits
2020-12-21 20:48:45 GeForce GTX 1650 SUPER-0 77936867 PRP starting from beginning
2020-12-21 20:48:48 GeForce GTX 1650 SUPER-0 77936867 OK         0 on-load: blockSize 400, 0000000000000003
2020-12-21 20:48:48 GeForce GTX 1650 SUPER-0 77936867 validating proof residues for power 8
2020-12-21 20:48:48 GeForce GTX 1650 SUPER-0 77936867 Proof using power 8
2020-12-21 20:48:56 GeForce GTX 1650 SUPER-0 77936867 OK       800   0.00% 1579c241dc63eca6 6531 us/it + check 2.69s + save 0.13s; ETA 5d 21:23
2020-12-21 20:49:59 GeForce GTX 1650 SUPER-0 77936867 OK     10000   0.01% fc4f135f7cf4ad29 6584 us/it + check 2.73s + save 0.14s; ETA 5d 22:31
2020-12-21 20:51:09 GeForce GTX 1650 SUPER-0 77936867 OK     20000   0.03% 3cd1bd9d5e09cbc5 6627 us/it + check 2.73s + save 0.14s; ETA 5d 23:26
2020-12-21 20:52:18 GeForce GTX 1650 SUPER-0 77936867 OK     30000   0.04% c4e0ff35e3290d98 6653 us/it + check 2.74s + save 0.14s; ETA 5d 23:58
2020-12-21 20:53:27 GeForce GTX 1650 SUPER-0 77936867 OK     40000   0.05% dffe1b1b0d748128 6630 us/it + check 2.73s + save 0.13s; ETA 5d 23:28
2020-12-21 20:54:36 GeForce GTX 1650 SUPER-0 77936867 OK     50000   0.06% 52e286945371ed29 6626 us/it + check 2.73s + save 0.14s; ETA 5d 23:21
2020-12-21 20:55:45 GeForce GTX 1650 SUPER-0 77936867 OK     60000   0.08% 0945da4dc08bdd95 6626 us/it + check 2.73s + save 0.14s; ETA 5d 23:20
2020-12-21 20:56:55 GeForce GTX 1650 SUPER-0 77936867 OK     70000   0.09% 7131fa4eb77f4bb2 6645 us/it + check 2.75s + save 0.14s; ETA 5d 23:43
2020-12-21 20:58:04 GeForce GTX 1650 SUPER-0 77936867 OK     80000   0.10% 8d76071d27ee4221 6679 us/it + check 2.75s + save 0.15s; ETA 6d 00:27
2020-12-21 20:59:14 GeForce GTX 1650 SUPER-0 77936867 OK     90000   0.12% 0bacff453b2f470e 6682 us/it + check 2.75s + save 0.20s; ETA 6d 00:30
2020-12-21 21:00:24 GeForce GTX 1650 SUPER-0 77936867 OK    100000   0.13% 6d7296b9e2830f50 6680 us/it + check 2.75s + save 0.14s; ETA 6d 00:25
2020-12-21 21:01:34 GeForce GTX 1650 SUPER-0 77936867 OK    110000   0.14% 8cbfd4435622bda7 6679 us/it + check 2.75s + save 0.23s; ETA 6d 00:24
2020-12-21 21:02:43 GeForce GTX 1650 SUPER-0 77936867 OK    120000   0.15% 79ae5dad855057ad 6681 us/it + check 2.75s + save 0.15s; ETA 6d 00:25
2020-12-21 21:03:53 GeForce GTX 1650 SUPER-0 77936867 OK    130000   0.17% 50c97bcbf876231f 6680 us/it + check 2.75s + save 0.23s; ETA 6d 00:23
2020-12-21 21:05:03 GeForce GTX 1650 SUPER-0 77936867 OK    140000   0.18% e1db15f897271496 6682 us/it + check 2.76s + save 0.16s; ETA 6d 00:24
2020-12-21 21:06:13 GeForce GTX 1650 SUPER-0 77936867 OK    150000   0.19% 127631386c6a9b17 6682 us/it + check 2.75s + save 0.14s; ETA 6d 00:22
2020-12-21 21:07:22 GeForce GTX 1650 SUPER-0 77936867 OK    160000   0.21% 25b7b6206fc6f085 6681 us/it + check 2.75s + save 0.19s; ETA 6d 00:21
2020-12-21 21:08:32 GeForce GTX 1650 SUPER-0 77936867 OK    170000   0.22% 416816b0d9f4bba8 6699 us/it + check 2.76s + save 0.14s; ETA 6d 00:43
2020-12-21 21:09:42 GeForce GTX 1650 SUPER-0 77936867 OK    180000   0.23% 6bee5d054f770861 6681 us/it + check 2.75s + save 0.14s; ETA 6d 00:19
2020-12-21 21:10:52 GeForce GTX 1650 SUPER-0 77936867 OK    190000   0.24% f37f068f014b18a0 6680 us/it + check 2.75s + save 0.14s; ETA 6d 00:16
2020-12-21 21:11:59 GeForce GTX 1650 SUPER-0 77936867 Stopping, please wait..
2020-12-21 21:12:02 GeForce GTX 1650 SUPER-0 77936867 OK    200000   0.26% f0b04b45b0855bd2 6679 us/it + check 2.77s + save 0.23s; ETA 6d 00:14
2020-12-21 21:12:02 GeForce GTX 1650 SUPER-0 Exiting because "stop requested"
2020-12-21 21:12:02 GeForce GTX 1650 SUPER-0 Bye
DrobinsonPE is offline   Reply With Quote
Old 2020-12-22, 15:23   #2650
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

613 Posts
Default https://www.techpowerup.com/gpu-specs/radeon-pro-vii.c3575#:~:text=It%20features%203840%20shading%20

Quote:
Originally Posted by Xyzzy View Post
How many "R7 units" will an MI100 replace?

Maybe $6.5K isn't so crazy?
I would think that a Radeon Pro VII is pretty good AFA "bang for the buck":
It has 6.4 F64 Tflops .
They should be good for 2X performance of the previous Radeon VII and it is possible to actually buy them retail for 1900 US dollars from B&H video.
However, that cooler on it looks pretty whimpy

Last fiddled with by tServo on 2020-12-22 at 15:27
tServo is offline   Reply With Quote
Old 2020-12-22, 19:04   #2651
moebius
 
moebius's Avatar
 
Jul 2009
Germany

547 Posts
Default

Quote:
Originally Posted by tServo View Post
It has 6.4 F64 Tflops .They should be good for 2X performance of the previous Radeon VII and it is possible to actually buy them retail for 1900 US dollars from B&H video.
A user of this forum posted a picture of his new build about 1 year ago which had included a Radeon PRO VII. Maybe he can give us concrete values. Unfortunately I can't find the post anymore.
moebius is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
mfakto: an OpenCL program for Mersenne prefactoring Bdot GPU Computing 1668 2020-12-22 15:38
GPUOWL AMD Windows OpenCL issues xx005fs GpuOwl 0 2019-07-26 21:37
Testing an expression for primality 1260 Software 17 2015-08-28 01:35
Testing Mersenne cofactors for primality? CRGreathouse Computer Science & Computational Number Theory 18 2013-06-08 19:12
Primality-testing program with multiple types of moduli (PFGW-related) Unregistered Information & Answers 4 2006-10-04 22:38

All times are UTC. The time now is 21:27.

Sun Mar 7 21:27:06 UTC 2021 up 94 days, 17:38, 0 users, load averages: 2.96, 2.56, 2.47

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.