mersenneforum.org  

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

Reply
 
Thread Tools
Old 2013-02-08, 18:12   #12
Redarm
 
Redarm's Avatar
 
Apr 2012
Berlin Germany

3·17 Posts
Default

Tried to find the max FFT length with my GTX 680 4GB
the cufftbench crashes near 30M
is this just normal?
Attached Thumbnails
Click image for larger version

Name:	Neue Bitmap (2).jpg
Views:	161
Size:	233.3 KB
ID:	9305  
Redarm is offline   Reply With Quote
Old 2013-02-08, 18:37   #13
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3×29×83 Posts
Default

Quote:
Originally Posted by Redarm View Post
Tried to find the max FFT length with my GTX 680 4GB
the cufftbench crashes near 30M
is this just normal?
I've seen it crash before, but I don't know enough about how it works to try and fix it. There are no known reports of actual LL tests crashing, so I haven't really put any thought into it.
Dubslow is offline   Reply With Quote
Old 2013-02-08, 20:03   #14
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2×3×1,693 Posts
Default

CUDALucas has shown itself to be sensitive to memory errors if the memory clock is too high. This is known to happen on some GTX 570's even when they are running at nVidia's specified memclock. Get MSI Afterburner and try cutting the memory speed by 100 or 200 MHz and run the test again.

I could not complete the CuLu self-test until I slowed my 570's memory from 1900 to 1800 MHz.
kladner is offline   Reply With Quote
Old 2013-02-08, 20:13   #15
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3·29·83 Posts
Default

Quote:
Originally Posted by kladner View Post
CUDALucas has shown itself to be sensitive to memory errors if the memory clock is too high. This is known to happen on some GTX 570's even when they are running at nVidia's specified memclock. Get MSI Afterburner and try cutting the memory speed by 100 or 200 MHz and run the test again.

I could not complete the CuLu self-test until I slowed my 570's memory from 1900 to 1800 MHz.
This isn't the self test, it's the -cufftbench option. It's crashed before on my 460, and that was a completely stable card that consistently turned in matching double checks and never failed the selftest. This crash is a CUDA thing, with a CUDA API error, not a roundoff thing.
Dubslow is offline   Reply With Quote
Old 2013-02-08, 22:09   #16
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23×53 Posts
Default

Quote:
Originally Posted by Aramis Wyler View Post
I'll post the first search raw, and post the second one after I filter it down to remove the lower increments w/higher runtimes.
First run, from 11M to 33M at 1M increments.
Quote:
CUDALucas-2.03-cuda4.1-sm_21-x86-64.exe -cufftbench 11534336 34603008 1048576
CUFFT bench start = 11534336 end = 34603008 distance = 1048576
CUFFT_Z2Z size= 11534336 time= 10.476141 msec
CUFFT_Z2Z size= 12582912 time= 8.522680 msec
CUFFT_Z2Z size= 13631488 time= 12.516187 msec
CUFFT_Z2Z size= 14680064 time= 9.308846 msec
CUFFT_Z2Z size= 15728640 time= 11.437217 msec
CUFFT_Z2Z size= 16777216 time= 11.563828 msec
CUFFT_Z2Z size= 17825792 time= 17.191065 msec
CUFFT_Z2Z size= 18874368 time= 11.988928 msec
CUFFT_Z2Z size= 19922944 time= 19.665787 msec
CUFFT_Z2Z size= 20971520 time= 14.167406 msec
CUFFT_Z2Z size= 22020096 time= 16.041197 msec
CUFFT_Z2Z size= 23068672 time= 20.987940 msec
CUFFT_Z2Z size= 24117248 time= 26.096746 msec
CUFFT_Z2Z size= 25165824 time= 18.335030 msec
CUFFT_Z2Z size= 26214400 time= 19.484661 msec
CUFFT_Z2Z size= 27262976 time= 25.086594 msec
CUFFT_Z2Z size= 28311552 time= 20.297859 msec
CUFFT_Z2Z size= 29360128 time= 19.867901 msec
CUFFT_Z2Z size= 30408704 time= 32.914940 msec
CUFFT_Z2Z size= 31457280 time= 23.047129 msec
CUFFT_Z2Z size= 32505856 time= 35.032364 msec
CUFFT_Z2Z size= 33554432 time= 28.495123 msec
Aramis Wyler is offline   Reply With Quote
Old 2013-02-08, 22:38   #17
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23×53 Posts
Default

In this run I did 32k increments from 17M to 22M+1. No real surprises here: the M markes are all the lowest numbers. I deleted numbers from the list below that had better numbers further on, but after 18M (11 seconds) I just put down all times < 16 seconds for reference. The full log is attached. After that I ran it till it crashed at 1M increments.

Quote:
CUDALucas-2.03-cuda4.1-sm_21-x86-64.exe -cufftbench 17825792 22020097 32768
CUFFT bench start = 17825792 end = 22020096 distance = 32768
CUFFT_Z2Z size= 17825792 time= 17.196913 msec
CUFFT_Z2Z size= 18022400 time= 17.184216 msec
CUFFT_Z2Z size= 18350080 time= 12.858215 msec
CUFFT_Z2Z size= 18579456 time= 12.395390 msec
CUFFT_Z2Z size= 18874368 time= 11.992032 msec *
CUFFT_Z2Z size= 19267584 time= 14.244596 msec
CUFFT_Z2Z size= 19660800 time= 15.384396 msec
CUFFT_Z2Z size= 20480000 time= 15.386993 msec
CUFFT_Z2Z size= 20643840 time= 14.298823 msec
CUFFT_Z2Z size= 20971520 time= 14.186642 msec
CUFFT_Z2Z size= 21233664 time= 15.297678 msec
CUFFT_Z2Z size= 22020096 time= 16.054712 msec
After that I ran it till it crashed.
Quote:
C:\Users\Bill\Program Files\CudaLucas>CUDALucas-2.03-cuda4.1-sm_21-x86-64.exe -cufftbench 11534336 69206016 1048576
CUFFT bench start = 11534336 end = 69206016 distance = 1048576
CUFFT_Z2Z size= 11534336 time= 10.476063 msec
CUFFT_Z2Z size= 12582912 time= 8.523827 msec
CUFFT_Z2Z size= 13631488 time= 12.510979 msec
CUFFT_Z2Z size= 14680064 time= 9.300248 msec
CUFFT_Z2Z size= 15728640 time= 11.438691 msec
CUFFT_Z2Z size= 16777216 time= 11.566601 msec
CUFFT_Z2Z size= 17825792 time= 17.186182 msec
CUFFT_Z2Z size= 18874368 time= 11.986383 msec
CUFFT_Z2Z size= 19922944 time= 19.662849 msec
CUFFT_Z2Z size= 20971520 time= 14.184292 msec
CUFFT_Z2Z size= 22020096 time= 16.061382 msec
CUFFT_Z2Z size= 23068672 time= 20.971977 msec
CUFFT_Z2Z size= 24117248 time= 26.087456 msec
CUFFT_Z2Z size= 25165824 time= 18.322544 msec
CUFFT_Z2Z size= 26214400 time= 19.491758 msec
CUFFT_Z2Z size= 27262976 time= 25.062517 msec
CUFFT_Z2Z size= 28311552 time= 20.305361 msec
CUFFT_Z2Z size= 29360128 time= 19.874012 msec
CUFFT_Z2Z size= 30408704 time= 32.909336 msec
CUFFT_Z2Z size= 31457280 time= 23.071260 msec
CUFFT_Z2Z size= 32505856 time= 35.012535 msec
CUFFT_Z2Z size= 33554432 time= 28.497805 msec
CUFFT_Z2Z size= 34603008 time= 34.820026 msec
CUFFT_Z2Z size= 35651584 time= 34.420364 msec
CUFFT_Z2Z size= 36700160 time= 26.757685 msec
CUFFT_Z2Z size= 37748736 time= 25.630711 msec
CUFFT_Z2Z size= 38797312 time= 47.039764 msec
CUFFT_Z2Z size= 39845888 time= 39.513092 msec
CUFFT_Z2Z size= 40894464 time= 41.624195 msec
CUFFT_Z2Z size= 41943040 time= 30.448586 msec
CUFFT_Z2Z size= 42991616 time= 53.082886 msec
CUFFT_Z2Z size= 44040192 time= 32.342079 msec
CUFFT_Z2Z size= 45088768 time= 52.830536 msec
CUFFT_Z2Z size= 46137344 time= 43.976074 msec
CUFFT_Z2Z size= 47185920 time= 34.433437 msec
CUFFT_Z2Z size= 48234496 time= 52.412827 msec
CUDALucas.cu(1091) : cudaSafeCall() Runtime API error 6: the launch timed out and was terminated.
C:\Users\Bill\Program Files\CudaLucas>
Attached Files
File Type: txt reference_580.txt (6.1 KB, 142 views)
Aramis Wyler is offline   Reply With Quote
Old 2013-02-08, 22:47   #18
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23×53 Posts
Default

I'll edit this again in a bit, I was trying some math, but I got the inverse of what I wanted. Whoops!

EDIT: I give up, I don't know where the tradeoff is between bigger numbers and lower speed.
EDIT2: Ok, one more try at my crazy efficiency crud.

Quote:
11534336 | 10.476063 | 1.434
12582912 | 08.523827 | 1.069
13631488 | 12.510979 | 1.449
14680064 | 09.300248 | 1.000
15728640 | 11.438691 | 1.148
16777216 | 11.566601 | 1.088
17825792 | 17.186182 | 1.522
18874368 | 11.986383 | 1.002
19922944 | 19.662849 | 1.558
20971520 | 14.184292 | 1.068
22020096 | 16.061382 | 1.151
23068672 | 20.971977 | 1.435
24117248 | 26.087456 | 1.707
25165824 | 18.322544 | 1.149
26214400 | 19.491758 | 1.174
27262976 | 25.062517 | 1.451
28311552 | 20.305361 | 1.132
29360128 | 19.874012 | 1.068
30408704 | 32.909336 | 1.708
31457280 | 23.071260 | 1.158
32505856 | 35.012535 | 1.700
33554432 | 28.497805 | 1.341
34603008 | 34.820026 | 1.588
35651584 | 34.420364 | 1.524
36700160 | 26.757685 | 1.151
37748736 | 25.630711 | 1.072
38797312 | 47.039764 | 1.914
39845888 | 39.513092 | 1.565
40894464 | 41.624195 | 1.607
41943040 | 30.448586 | 1.146
42991616 | 53.082886 | 1.949
44040192 | 32.342079 | 1.159
45088768 | 52.830536 | 1.849
46137344 | 43.976074 | 1.505
47185920 | 34.433437 | 1.152
48234496 | 52.412827 | 1.715

Last fiddled with by Aramis Wyler on 2013-02-08 at 23:47 Reason: bad math!
Aramis Wyler is offline   Reply With Quote
Reply



All times are UTC. The time now is 14:52.


Fri Jul 7 14:52:20 UTC 2023 up 323 days, 12:20, 0 users, load averages: 1.24, 1.15, 1.12

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2023, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔