It seems 13.1 is now the oldest available ati-drivers, which is known to be broken for OpenCL. Can the actually-working versions get re-added please?
(In reply to comment #0) > which is known to be broken for OpenCL Care to elaborate?
I'm not familiar with the problems enough to be too specific, but the broken drivers all return invalid results with net-misc/bfgminer at least.
This might be a bfgminer-only issue. What version of it are you using? I've noticed that from some time it no longer pulls amd-adl-sdk which I don't know if it's a good or a bad thing. I've done some testings with bfgminer and ati-drivers-13.2_beta7 and: * bfgminer-2.6.5 works without issues. * bfgminer-2.10.5 does not work, shows WAIT all the time and seems like it's frozen. And back to the topic, what ati-drivers version specifically are you asking for?
WAIT suggests some unrelated issue - if you could post a debug log somewhere of that, I'd appreciate it. I'm using ati-drivers 11.3 (w/ stream SDK 2.1) myself, but opened this on behalf of user reports (who I believe are CC'd and may be able to elaborate on their setup) - it is my understanding that only the latest version has this problem, so probably anything before that should work. The classic 11.3 would be a nice bonus since it performs far better for 5xxx cards. When you tested, did you confirm it was using AMD OpenCL device(s)? The easiest way to do this is to run with the -d? option to get a list by name. Since BFGMiner only depends on *some* OpenCL, it could also be satisifed by and use nvidia or Intel...
As this seems to be a bfgminer-only related issue I'm changing the summary for a more accurate one. If bfgminer needs an older ati-drivers version or not, that will be decided by the maintainer(s) of ati-drivers. (In reply to comment #4) > I'm using ati-drivers 11.3 (w/ stream SDK 2.1) myself I'm not sure if the maintainer will agree with adding such old version to the portage tree. Anyways, the sabayon-distro layman repo still keeps some >=11.6 ebuilds so as a workaround people could use that. > it is my understanding that only the latest version has this problem, so probably anything before that should work. ati-drivers-13.1_pre897 is in the portage tree, which technically is an older version of the 13.1 - have these users tested it? > When you tested, did you confirm it was using AMD OpenCL device(s)? I think this confirms it is using an OpenCL device: $ ▶ bfgminer -n [2013-03-16 18:44:14] CL Platform 0 vendor: Advanced Micro Devices, Inc. [2013-03-16 18:44:14] CL Platform 0 name: AMD Accelerated Parallel Processing [2013-03-16 18:44:14] CL Platform 0 version: OpenCL 1.2 AMD-APP (1124.2) [2013-03-16 18:44:14] Platform 0 devices: 1 [2013-03-16 18:44:14] 0 Pitcairn [2013-03-16 18:44:14] GPU 0 AMD Radeon HD 7800 Series hardware monitoring enabled [2013-03-16 18:44:14] 1 GPU devices max detected About a debug log, if I run $ ▶ bfgminer --benchmark I only get this: [2013-03-16 18:52:49] Disabling extra threads due to dynamic mode. [2013-03-16 18:52:49] Tune dynamic intensity with --gpu-dyninterval [2013-03-16 18:52:49] Pool 0 not providing work fast enough If I run: $ ▶ bfgminer --benchmark --debug This repeats infinitely: [2013-03-16 18:55:13] Generated benchmark work [2013-03-16 18:55:13] Work stale due to expiry (1363365512 >= 119) [2013-03-16 18:55:13] Pushing work from pool 0 to hash queue [2013-03-16 18:55:13] Discarded cloned or rolled work [2013-03-16 18:55:13] Generated benchmark work [2013-03-16 18:55:13] Work stale due to expiry (1363365512 >= 119) [2013-03-16 18:55:13] Pushing work from pool 0 to hash queue [2013-03-16 18:55:13] Discarded cloned or rolled work [2013-03-16 18:55:13] Generated benchmark work [2013-03-16 18:55:13] Work stale due to expiry (1363365512 >= 119) [2013-03-16 18:55:13] Discarded cloned or rolled work [2013-03-16 18:55:13] Pushing work from pool 0 to hash queue [2013-03-16 18:55:13] Generated benchmark work [2013-03-16 18:55:13] Work stale due to expiry (1363365512 >= 119) [2013-03-16 18:55:13] Pushing work from pool 0 to hash queue [2013-03-16 18:55:13] Discarded cloned or rolled work [2013-03-16 18:55:13] Generated benchmark work I'm not used to use bfgminer so if you need something more specific and you use IRC (freenode), contact me there.
I have been mining for a long time (near 2 months) on x11-drivers/ati-drivers-13.1 and bfgminer-2.10.6 and have no problems with it.
bfgminer-3.1.* and ati-drivers-13.6* are currently in the tree, please test and see if it works for you. Closing this bug for now, feel free to reopen if you have any pertinent info on this bug.
Still doesn't work here (ati-drivers-13.6_beta, bfgminer-3.1.1) [2013-07-03 02:54:14] OCL 0: invalid nonce - HW error
The problems with newer SDKs are related to the BFI_INT patching (hacking the binary kernel the SDK outputs). I've modified BFGMiner 3.1.2 (soon to be released) to not do this with 13.2+, but the performance takes a hit so I'd still recommend putting some older versions in the tree. 13.1 seems to work fine on 7xxx cards, but 6xxx and 5xxx perform poorly even with its SDK. I'm told AMD-APP 2.5 (not sure which Catalyst this is) works good for both 5xxx and 7xxx, but I haven't tested it myself - going this far back might make it necessary to split the SDK out of ati-drivers (they're independent components, really, but AMD bundles a SDK with newer drivers).
I have radeon hd6850 and hd6870. I use bfgminer 2.10.10. It works fine with ati-drivers 13.1 and give me h\w errors with 13.4.
looks like it works again with 13.9, can anyone else confirm this?
It should work, but doesn't perform as well.
Newer versions seem to work, closing.
