search

cbuchner1 / CudaMiner

a CUDA accelerated litecoin mining application based on pooler's CPU miner
Other
692 stars 304 forks source link

CPU cannot validate when using kepler kernel #49

Closed WhiskeyBravo closed 10 years ago

WhiskeyBravo commented 10 years ago

I recently setup a system with Ubuntu 12.04, a 560ti and a 650ti. Using CUDA 5.0 and 319.17. Download from github, follow compile procedures. If I launch using an F kernel, I get about 80kH/s on the 650 and 130 on the 650Ti, everything works.

If I launch using an F kernel on the 560 and a K on the 650, the cpu cannot validate the result for the 650. The 560 continues to work.

If I launch using only the 650 (filtering via the -d flag) with no specified kernel parameter, it crashes.

Thoughts?

Edit: Just tried doing this with CUDA 5.5 and i'm seeing the same behavior.

WhiskeyBravo commented 10 years ago

For some extra data, here is what smi see's (newer driver now that I'm trying 5.5) Mon Dec 16 15:10:01 2013 +------------------------------------------------------+ | NVIDIA-SMI 5.319.37 Driver Version: 319.37 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 560 Ti Off | 0000:01:00.0 N/A | N/A | | 59% 81C N/A N/A / N/A | 1586MB / 2047MB | N/A Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX 650 Ti Off | 0000:02:00.0 N/A | N/A | | 43% 70C N/A N/A / N/A | 1579MB / 2047MB | N/A Default | +-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+ | Compute processes: GPU Memory | | GPU PID Process name Usage | |=============================================================================| | 0 Not Supported | | 1 Not Supported | +-----------------------------------------------------------------------------+

First few lines of cudaminer in the working configuration REMOTE:~/Downloads$ ./launch_cudaMiner.sh * CudaMiner for nVidia GPUs by Christian Buchner * This is version 2013-12-01 (beta) based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler Cuda additions Copyright 2013 Christian Buchner My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-12-16 15:11:03] Starting Stratum on stratum+tcp://ltc-eu.give-me-coins.com:3333 [2013-12-16 15:11:03] Binding thread 0 to cpu 0 [2013-12-16 15:11:03] 2 miner threads started, using 'scrypt' algorithm. [2013-12-16 15:11:03] Binding thread 1 to cpu 1 [2013-12-16 15:11:27] GPU #1: GeForce GTX 650 Ti with compute capability 3.0 [2013-12-16 15:11:27] GPU #1: interactive: 0, tex-cache: 2D, single-alloc: 1 [2013-12-16 15:11:27] GPU #1: using launch configuration F24x16 [2013-12-16 15:11:27] GPU #1: GeForce GTX 650 Ti, 12288 hashes, 0.54 khash/s [2013-12-16 15:11:28] GPU #1: GeForce GTX 650 Ti, 36864 hashes, 57.06 khash/s [2013-12-16 15:11:49] GPU #0: GeForce GTX 560 Ti with compute capability 2.1 [2013-12-16 15:11:49] GPU #0: interactive: 0, tex-cache: 2D, single-alloc: 1 [2013-12-16 15:11:49] GPU #0: using launch configuration F24x16 [2013-12-16 15:11:49] GPU #0: GeForce GTX 560 Ti, 12288 hashes, 0.28 khash/s [2013-12-16 15:11:49] GPU #0: GeForce GTX 560 Ti, 24576 hashes, 91.84 khash/s [2013-12-16 15:11:54] GPU #0: GeForce GTX 560 Ti, 602112 hashes, 135.56 khash/s [2013-12-16 15:11:54] accepted: 1/1 (100.00%), 192.61 khash/s (yay!!!) [2013-12-16 15:11:54] GPU #0: GeForce GTX 560 Ti, 110592 hashes, 121.78 khash/s [2013-12-16 15:11:55] accepted: 2/2 (100.00%), 178.83 khash/s (yay!!!)

First few lines of cudaminer in the broken config... * CudaMiner for nVidia GPUs by Christian Buchner * This is version 2013-12-01 (beta) based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler Cuda additions Copyright 2013 Christian Buchner My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-12-16 15:13:21] Starting Stratum on stratum+tcp://ltc-eu.give-me-coins.com:3333 [2013-12-16 15:13:21] 2 miner threads started, using 'scrypt' algorithm. [2013-12-16 15:13:21] Binding thread 0 to cpu 0 [2013-12-16 15:13:21] Binding thread 1 to cpu 1 [2013-12-16 15:13:42] Stratum detected new block [2013-12-16 15:14:06] GPU #0: GeForce GTX 560 Ti with compute capability 2.1 [2013-12-16 15:14:06] GPU #0: interactive: 0, tex-cache: 2D, single-alloc: 1 [2013-12-16 15:14:06] GPU #0: using launch configuration F24x16 [2013-12-16 15:14:06] GPU #1: GeForce GTX 650 Ti with compute capability 3.0 [2013-12-16 15:14:06] GPU #1: interactive: 0, tex-cache: 2D, single-alloc: 1 [2013-12-16 15:14:06] GPU #1: using launch configuration K16x14 [2013-12-16 15:14:06] GPU #1: GeForce GTX 650 Ti, 7168 hashes, 0.16 khash/s [2013-12-16 15:14:06] GPU #1: GeForce GTX 650 Ti, 14336 hashes, 160.95 khash/s [2013-12-16 15:14:06] GPU #0: GeForce GTX 560 Ti, 12288 hashes, 0.28 khash/s [2013-12-16 15:14:07] GPU #0: GeForce GTX 560 Ti, 24576 hashes, 93.15 khash/s [2013-12-16 15:14:07] Stratum detected new block [2013-12-16 15:14:07] GPU #1: GeForce GTX 650 Ti result does not validate on CPU! [2013-12-16 15:14:07] GPU #1: GeForce GTX 650 Ti, 193536 hashes, 209.13 khash/s [2013-12-16 15:14:07] GPU #0: GeForce GTX 560 Ti, 86016 hashes, 118.62 khash/s [2013-12-16 15:14:07] GPU #1: GeForce GTX 650 Ti result does not validate on CPU! [2013-12-16 15:14:08] GPU #1: GeForce GTX 650 Ti result does not validate on CPU! [2013-12-16 15:14:09] GPU #0: GeForce GTX 560 Ti, 184320 hashes, 124.98 khash/s [2013-12-16 15:14:09] GPU #0: GeForce GTX 560 Ti, 36864 hashes, 100.19 khash/s [2013-12-16 15:14:09] GPU #1: GeForce GTX 650 Ti result does not validate on CPU!

cbuchner1 commented 10 years ago

the 12.01 version still had a bug that would affect reliability. It would not correctly make sure that kernels weren't executed with overlap, causing validation errors.

have you tried the 12.10 version yet?

if you can't compile 12.10, you can fix that particular bug by forcing the "concurrent" boolean flag to true always (salsa_kernel.cu)

Christian

WhiskeyBravo commented 10 years ago

Just tried with the latest from github -- won't compile with 5.5, but it worked with 5.0 and my numbers are much better across the board as well as Kepler kernels working. Thanks!