Quote from: djm34 on January 26, 2015, 12:48:39 PM

Quote from: tsiv on January 26, 2015, 12:40:51 PM

Quote from: Dotcommie on January 26, 2015, 12:38:28 PM

Quote from: tsiv on January 26, 2015, 12:30:14 PM

On a related note, I am I bit concerned about the miner on non-750Ti cards. I'm getting pretty much zero rejects and exactly as many blocks as you could expect against the current network hash rate on my 750 rig, but a fair amount of rejects (but not 100%) on the 970. Can't see any particular reason why the same code would work on 750 and not on 970 though. It is a mystery. Just seems bit odd that near all the rejects I get are on the 970. Could be coincidence but I can't help but wonder...

Code:

[2015-01-23 01:11:53] accepted: 1/1 (100.00%), 10038 khash/s (yay!!!)
[2015-01-23 04:56:03] accepted: 2/2 (100.00%), 10036 khash/s (yay!!!)
[2015-01-23 16:48:20] accepted: 3/3 (100.00%), 10040 khash/s (yay!!!)
[2015-01-24 07:28:52] accepted: 4/4 (100.00%), 10039 khash/s (yay!!!)
[2015-01-24 07:53:30] accepted: 5/5 (100.00%), 10034 khash/s (yay!!!)
[2015-01-24 09:22:16] accepted: 6/6 (100.00%), 10034 khash/s (yay!!!)
[2015-01-25 05:06:02] accepted: 7/7 (100.00%), 10035 khash/s (yay!!!)
[2015-01-25 16:22:12] accepted: 8/8 (100.00%), 10047 khash/s (yay!!!)
[2015-01-25 21:06:44] accepted: 9/9 (100.00%), 10027 khash/s (yay!!!)
[2015-01-26 04:59:57] accepted: 10/10 (100.00%), 10034 khash/s (yay!!!)
[2015-01-26 13:35:12] accepted: 11/11 (100.00%), 10038 khash/s (yay!!!)

I don't believe I had any rejects on your original spreadminer on my 970s. Seemed to work with the same consistency as 750s for me, unless this is newer code you're talking about.

Talking about the original, yep. Oh well, maybe my 970 is just jinxed Cheesy

register pressure can do that at some extend.
I had a similar problem with a previous version of keccak and compute 5.2 where none of the hash were getting validated due to some bad allocation (The content of the register wasn't what it was supposed to be... and the problem wasn't showing at compile time...)

Well, I went ahead and added CPU verification into the original spreadminer. Running in benchmark mode quickly shows if things are working or not. And the verdict for my 970 is that it doesn't work. Windows binary is still in build and will be added later, fuck compiling shavite for four different targets.

The weird part is that people, including me, have been successfully running it on 970s and now suddenly it's all "no, fuck you dad." Maybe some driver upgrade messed something up..? Hnnngh, do I really need to register as a super official extremely important CUDA developer to see if compiling with CUDA 7.0 would make any difference...

Edit: Win32 binary up at https://github.com/tsiv/spreadminer/releases/tag/v0.1r3 and source at https://github.com/tsiv/spreadminer as always