08-17-2017, 11:01 PM
(This post was last modified: 08-18-2017, 04:25 AM by BobbyMac99.)
(08-13-2017, 04:29 AM)epixoip Wrote: 0. Wait for new cards. We're speculating end of Q3/early Q4, but we don't know for sure because Nvidia is always notoriously tight-lipped about when new products will drop.
1. GTX 1080 Ti gets along just fine with OpenCL, and Nvidia's OpenCL is solid. So I'm not sure what you are referring to.
2. The basis for the objection to OEM cards is that they are garbage. They are designed to provide the bare minimum functionality for gaming workloads ("bursty" utilization for only a few hours per day) and cannot cope with compute workloads (hammering the ALUs with constant sustained load for days/weeks/months/years on end.) They will die a painful, sometimes fiery, death. Keep in mind that I work with GPGPU on a scale that very few people in the world do, and thus I have a lot of datapoints. When I say "only buy reference design", you would be wise indeed to heed my words.
3. GTX 970's memory wasn't botched. The only people that made a big deal out of it are people who don't understand hardware. And the GTX 970 certainly wasn't the first card with this "issue."
4. Of course the GTX 970 is slow compared to the GTX 1080 Ti. You're comparing 1664 Maxwell cores @ 1250 Mhz to 3584 Pascal cores @ 1645 Mhz. 970 doesn't even stand a chance.
==========================================
I was referring to this: https://hashcat.net/forum/thread-6772.html
So either OpenCl has a problem, or Hashcat does - possible the card itself, although it works in video mode just fine, it also works in benchmark with OpenCL as the driver just fine. The 970 I have doesn't suffer from this issue - only the 1080ti. It sometimes works - about 5% of the time, the rest - dies. I can only make references to that hardware which I have actually owned. You however, have the position of being a reference point for a lot of posts. I won't argue the 970 - it does no good and serves no purpose.
So I am at a crossroad, return the 1080ti to NEWEGG (for the 2nd time) that potentially has no hardware issues because it won't work for me - and then wait until maybe the end of the year for a newer card. Or try and find a fix. Nvidia is worthless on support, i suspect no employee there even looks at their forums and have not had a response to support messages to them.
So, being as you are the center of a lot of communications you might know: Has anyone compiled under windows using MinGW or MSYS2 the current Hashcat code, and have those instructions been posted - I can't seem to find a reference. If that can be done - I at least could debug the code and find out what calls are forcing the OpenCL error. Thus ruling out Hashcat, or OpenCl/drivers, or the hardware.