PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

  1. Read our rules and policies.
  2. Download, install and run the BOINC software used by PrimeGrid.
  3. When prompted, enter the URL:
    http://dev.primegrid.com/
  4. Select which subprojects to run via the PrimeGrid Preferences.

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : General discussion : Will CUDA 3.1 appear soon?

Subscribe Sort
Author Message
Profile Nawiedzony
Avatar
Send message
Joined: 8 Aug 10
Posts: 4
ID: 65128
Credit: 2,812,958
RAC: 0
321 LLR Bronze: Earned 10,000 credits (14,629)PPS LLR Silver: Earned 100,000 credits (170,664)SGS LLR Silver: Earned 100,000 credits (133,666)TRP LLR Bronze: Earned 10,000 credits (10,570)Woodall LLR Bronze: Earned 10,000 credits (11,572)Cullen/Woodall Sieve (suspended) Silver: Earned 100,000 credits (249,640)PPS Sieve Ruby: Earned 2,000,000 credits (2,221,920)
Message 25943 - Posted: 27 Aug 2010 | 20:06:54 UTC

Is it planned to carry CUDA 3.1 technology into effect within the project?

ID: 25943 | Report as offensive
Ken_g6Project donor
Volunteer developer
Avatar
Send message
Joined: 4 Jul 06
Posts: 915
ID: 3110
Credit: 183,164,814
RAC: 0
Discovered 1 mega primeFound 2 primes in the 2018 Tour de PrimesFound 1 prime in the 2019 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Ruby: Earned 2,000,000 credits (2,010,094)Cullen LLR Ruby: Earned 2,000,000 credits (2,022,806)ESP LLR Amethyst: Earned 1,000,000 credits (1,193,202)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (767,102)PPS LLR Jade: Earned 10,000,000 credits (17,537,174)PSP LLR Ruby: Earned 2,000,000 credits (4,060,731)SoB LLR Ruby: Earned 2,000,000 credits (3,860,053)SR5 LLR Ruby: Earned 2,000,000 credits (2,061,736)SGS LLR Amethyst: Earned 1,000,000 credits (1,923,102)TPS LLR (retired) Bronze: Earned 10,000 credits (19,376)TRP LLR Ruby: Earned 2,000,000 credits (2,520,745)Woodall LLR Ruby: Earned 2,000,000 credits (2,021,413)321 Sieve Ruby: Earned 2,000,000 credits (2,915,071)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (8,584,236)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (2,461,309)PPS Sieve Emerald: Earned 50,000,000 credits (83,501,701)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (352,129)TRP Sieve (suspended) Gold: Earned 500,000 credits (776,202)AP 26/27 Turquoise: Earned 5,000,000 credits (5,851,049)GFN Sapphire: Earned 20,000,000 credits (34,062,697)PSA Ruby: Earned 2,000,000 credits (4,154,617)
Message 26113 - Posted: 4 Sep 2010 | 19:04:52 UTC - in response to Message 25943.
Last modified: 4 Sep 2010 | 19:44:56 UTC

There isn't anything I see CUDA 3.1 can do that CUDA 2.3 isn't already doing equally well in PPSieve-CUDA. I don't need memory accesses except to load and save registers at the start and end of a kernel. I have no need to run multiple kernels in parallel. And the 64-bit multiplies already get improved on Fermi without changing CUDA versions. If you see another particular innovation not involving the memory systems (Edit: or floating point), you're welcome to point it out.
____________

ID: 26113 | Report as offensive
Profile Nawiedzony
Avatar
Send message
Joined: 8 Aug 10
Posts: 4
ID: 65128
Credit: 2,812,958
RAC: 0
321 LLR Bronze: Earned 10,000 credits (14,629)PPS LLR Silver: Earned 100,000 credits (170,664)SGS LLR Silver: Earned 100,000 credits (133,666)TRP LLR Bronze: Earned 10,000 credits (10,570)Woodall LLR Bronze: Earned 10,000 credits (11,572)Cullen/Woodall Sieve (suspended) Silver: Earned 100,000 credits (249,640)PPS Sieve Ruby: Earned 2,000,000 credits (2,221,920)
Message 26156 - Posted: 6 Sep 2010 | 23:29:00 UTC - in response to Message 26113.

"Math Libraries Performance Improvements, including:
- Improved performance of selected transcendental functions from the log, pow, erf, and gamma families
- Significant improvements in double-precision FFT performance on Fermi-architecture GPUs for 2^n transform sizes
- Streaming API now supported in CUBLAS for overlapping copy and compute operations
- CUFFT Real-to-complex (R2C) and complex-to-real (C2R) optimizations for 2^n data sizes
- Improved performance for GEMV and SYMV subroutines in CUBLAS
- Optimized double-precision implementations of divide and reciprocal routines for the Fermi architecture "

Some people try cuda3.1 vs cuda2.3 in Collatz project, and 3.1 is a bit faster...

ID: 26156 | Report as offensive
jjwhalen
Avatar
Send message
Joined: 28 Dec 09
Posts: 78
ID: 52778
Credit: 476,139,121
RAC: 0
321 LLR Bronze: Earned 10,000 credits (35,831)Cullen LLR Bronze: Earned 10,000 credits (35,150)ESP LLR Bronze: Earned 10,000 credits (35,604)PPS LLR Silver: Earned 100,000 credits (103,518)PSP LLR Silver: Earned 100,000 credits (153,683)SoB LLR Bronze: Earned 10,000 credits (17,742)SGS LLR Bronze: Earned 10,000 credits (54,444)TRP LLR Silver: Earned 100,000 credits (137,791)Woodall LLR Bronze: Earned 10,000 credits (50,603)321 Sieve Silver: Earned 100,000 credits (202,830)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,071,626)PPS Sieve Double Silver: Earned 200,000,000 credits (466,804,040)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Bronze: Earned 10,000 credits (26,614)TRP Sieve (suspended) Silver: Earned 100,000 credits (363,752)AP 26/27 Silver: Earned 100,000 credits (101,981)GFN Ruby: Earned 2,000,000 credits (2,943,912)
Message 26160 - Posted: 7 Sep 2010 | 1:54:59 UTC - in response to Message 26156.
Last modified: 7 Sep 2010 | 2:05:35 UTC


Some people try cuda3.1 vs cuda2.3 in Collatz project, and 3.1 is a bit faster...


I can confirm that empirically on my GTX 465SC, though the CPU load is correspondingly higher (probably not a big shocker). By empirically I mean average elapsed time of Collatz v2.05 (cuda 3.1) vs. 2.03 (cuda 2.3).
____________

ID: 26160 | Report as offensive
Mutiny32*
Send message
Joined: 17 Aug 10
Posts: 2
ID: 65772
Credit: 1,417,690
RAC: 0
PSP LLR Silver: Earned 100,000 credits (357,702)TRP LLR Gold: Earned 500,000 credits (509,796)321 Sieve Bronze: Earned 10,000 credits (25,158)PPS Sieve Gold: Earned 500,000 credits (524,651)
Message 26595 - Posted: 25 Sep 2010 | 18:27:46 UTC

CUDA 3.2 RC is out and it appears to be touting a big jump in performance for most libraries and it looks like it has been highly optimized for Fermi over 3.1. Some notes:


  • CUSPARSE, a new library of GPU-accelerated sparse matrix routines for sparse/sparse and dense/sparse operations
  • CURAND, a new library of GPU-accelerated random number generation (RNG) routines, supporting Sobol quasi-random and XORWOW pseudo-random routines for in both host and device code
  • CUFFT performance tuned for radix-3, -5, and -7 transform sizes on Fermi architecture GPUs
  • CUBLAS performance improved 50% to 300% on Fermi architecture GPUs, for matrix multiplication of all datatypes and transpose variations
  • Support for malloc() and free() in CUDA C compute kernels
  • Added cuda-memcheck support for Fermi architecture GPUs

ID: 26595 | Report as offensive

Message boards : General discussion : Will CUDA 3.1 appear soon?

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.15, 0.04, 0.01
Generated 18 Aug 2022 | 7:03:59 UTC