-
Notifications
You must be signed in to change notification settings - Fork 825
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error -4: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel) #246
Comments
There is no pool-worksize setting. Can we try a few things:
Also, it seems people had this error before: https://www.weminecryptos.com/forum/topic/2299-getting-error-4-enqueueing-kernel-onto-command-queue/ - it seems to be related with low system RAM. I think you might just be getting this because alexkarnew might need different settings, so you need to try ckolivas first. |
Ok this has been mitigated by replacing alexkarnew with ckolivas as algorithm; however, this now predictably crashes every 60s from this error. [22:28:03] Started sgminer 4.2.1 [22:29:19] Started at [2014-06-08 22:28:04] [22:29:19] Stale submissions discarded due to new blocks: 0 [22:29:19] Summary of per device statistics: [22:29:19] GPU0 | (5s):937.7K (avg):944.7Kh/s | A:1024 R:0 HW:0 WU:1006.016/m Then it crashes out. This is with pool-gpu-threads : 1 - I continue to get -4 Errors if we do not specify pool-gpu-threads to 1 since my main gpu-threads is 2. gpu-threads 2 will always lead to an -4 Error. |
I am trying with "no-restart" : true right now. |
[22:33:41] Started sgminer 4.2.1 |
Just for start, can you try around line 6260 in Edit: Oh right you are on Windows. Are you able to recompile? |
tbh I wait on the Windows compiles provided by Elun on bitcointalk. I have On Sun, Jun 8, 2014 at 10:49 PM, Jan Berdajs notifications@github.com
|
Yes I also had problems with Windows build. Here is some discussion about it, it seems they figured it out: #229 I wouldn't like to add it to the branch, because I don't want to add code that has no effect. |
I think I figured out how to reproduce this. Seems it is indeed a bug. Only happens when |
@mrbrdo Have you managed to build SGminer with the latest commits? I tried last night and I had some major issues. But I don't know if it's my fault or it's something from the source. |
"I think I figured out how to reproduce this. Seems it is indeed a bug. Only happens when "pool-gpu-threads" is configured." Ditto - any thoughts on how to side-step this bug? Edit: hardcoding gpu-threads 1 in the bottom portion of .conf file & removing all instances of pool-gpu-threads appears to have allowed me to build ckolivas24000nf11 and start hashing on scrypt-N for more than a minute; however, if you switch pools down to scrypt with an nf10 it immediately fails. [00:42:46] Started sgminer 4.2.1 |
Here is an attempt to trick the miner into using 2 different kernel.cl files as the algorithm [00:47:33] Started sgminer 4.2.1 Again, trying to from from nf10 -> nf11; or nf11 -> nf10 throws failures. For now; I can add scrypt as a backup without an issue, since ckolivas works with nf10 |
As it is right now, this miner can effectively algo flip between scrypt-N, keccak, x11, x13 [no scrypt nf10] Which, for right now, is all of the feasible mining algorithms for GPUs. It looks like ASICs have increased scrypt difficulty to a phenomenal level already. |
If you don't use "pool-gpu-threads" in your config (at all) then it should not happen. Does it? Also, I experience this problem with Scrypt-N too. |
"If you don't use "pool-gpu-threads" in your config (at all) then it should not happen. Does it?" Confirmed - we are past the 60second shut off. However, I have removed all instances of pool-gpu-threads, and hardcoded gpu-threads to equal 1 in the .conf file. It allows the miner to run, but when you flip from scrypt to scrypt-N it throws -4 Error as provided above. |
It's not the "pool-nfactor" setting, because that one works effectively. However, I have NOT tried to flip back to an X13 after that. Experimenting now. |
Yeah it algo flips amongst nfactors ok. The problem lies in how it flips amongst nfactors within the same algo (ie scrypt). It fails consistently trying to switch only from scrypt10 to scrypt11 and/or either way back. |
@platinum4 How much RAM do you have installed? What is configured pagefile size? 3x R9 290, right? |
Well, it seems somehow it cannot allocate enough memory for kernel/buffers. This seems to work fine when pool-gpu-threads is not set (which means soft restart, mining threads are not completely stopped and restarted). But when it is set (hard restart, mining threads completely stopped and then restarted), it seems like there is some memory left reserved in the devices. But I cannot figure what it could be and why it is not happening with soft restart. It definitely happens because it is out of memory, for example if I set gpu-threads to 1 (which means less VRAM use), then it works fine. (also http://www.popekim.com/2012/07/opencl-getting-outofresource-or.html this is the error we get - -4 CL_MEM_OBJECT_ALLOCATION_FAILURE) |
@troky 8GB RAM, no problem mining any algos with it. Pagefile is set to automatic I am assuming. I'll try 12GB pagefile size and report back. Pagefile shouldn't matter. The same error occurs on these rigs - 2x 290X, 3x 290X, 3x 390X, so pagefiles should ideally be 8gb,12gb,12gb Expanded all page files to 12228MB still -4 Errors when switching amonst scrypt and scrypt-N |
And still getting SICK -> DEAD errors on a few cards (mainly only R9 290X Tri-X OC 1040/1300) |
That's probably unrelated, would probably happen with sph-sgminer-x11mod too. It's probably just misconfiguration. |
Well, not when Elun had extended that SICK timer... ;D But yeah, we'll go with a 'continuous configuration error' after all this time if it's easier to stomach. ;) |
Extending SICK timer is not a solution, it's a workaround... And if we start with those, we will get nowhere again. But I might just throw that restarting SICK GPUs out some day because it doesn't seem to ever help anything. One guy was complaining about SICK/DEAD on X13-mod, but then he found some different settings on some forum and no more SICK, and he even got better hashrate/WU out of it. I'm not saying the X13-mod kernel doesn't have problems, but we use the same that everyone does (from girino), so the problems should be the same no matter which miner you use. Also for example someone told me that he got better hashrate with intensity 18 instead of 20. So it's not necessary to always put everything to absolute max. Similarly keccak seems to work better with gpu-threads 1 instead of 2 on R9 280X (without changing any other settings). But back to the point, I spent some time looking at what could be causing this weird bug with scrypt kernels, but I have no idea at all yet :/ |
This issue is still open until more reports in on a successful flip to nscrypt and back. |
@platinum4 so are you saying you yourself do not experience the problem anymore? |
Still with nf11, and I think others do as well, reports if it from a few members in sgminer-dev IRC and on bitcoin talk still Sent from my iPhone
|
Hm, well the new v5_0 does not do a hard reset unless gpu-threads is really different. Does this still happen for you when switching from scrypt to scrypt-n or vice versa (assuming that you use the same gpu-threads for both)? Or does it only happen when switching from some algo where you use a different gpu-threads (e.g. Keccak)? |
I'll have to test on rigs later tonight I will get back to you ok Sent from my iPhone
|
@mrbrdo I cannot replicate this issue anymore [Windows binaries, possibly fixed by the slew of commits in June 2014]; closing this issue for now unless others can repeat it. |
Re-opened this issue as it is being experienced by others, such as @evolvia31 #308 |
Hi all, i just try this morning the last commit to branch v5_0 and issue continue :( |
@evolvia31 I have time to look into it this week, can you confirm it is still a problem, or has it been fixed? |
@evolvia31 can you paste your full config not just the profiles section please? |
Also your specs might be helpful. What model GPUs are you using? How much system memory? |
Hi, yes I try this morning again and bug is already with crash. My sgminer config file: |
Where are your pools in the config? Do you only have 1 GPU? I could have sworn the enqueue error was with GPU 1 and 2 while 0 was ok. Also using gpu threads 2 across the board seems a bit dangerous. I haven't dealt with non Xn algorithms in a while but I'm pretty sure some of the more intense algorithms needed only 1 thread. I would recommend you test each algorithm individually with only 1 pool and no switching to make sure you have the correct settings before putting them all in 1 file. One last thing... Do you run as root? I notice you have /home/sgminer/ |
@ystarnaud it's an older issue.. I was able to reproduce it too. It seems to happen when switching from Scrypt to Scrypt-N. |
Scrypt-N is working in a very strange way. Scrypt-N or its kernel are very capricious. |
Hi, i have 3 ou 4 GPU card by server, i use more than 20 different pool so i don't publish them but, i have build my config with one pool and each algo config works fine since few months. The problem begin during may but i don't know with which exactly sgminer version. All works fine during long weeks except if i try to switch from algo x11, x13 or x15 to scrypt-N. I have no problem when i switch from Scrypt-N to any other algo. The only problem is from X11, X13 or X15 to Scrypt-N. Sgminer restart don't work to re-enable GPU, i need to quit sgminer stay 30 sec and re-launch sgminer. Yes all instance works as root. |
I've given up on nscrypt until at least the winter time; I can build bins fine now but with 14.7RC drivers the TC was sliced in half, down to a maximum of 8192. Even then, hardware errors were experienced and the hashrate sucked. If I ever go back to nscrypt it will be with the 13.12 drivers on a dedicated rig, unless AMD comes out with better ones in the mean time. The only error I notice now when building nf11 is Error -61 memory size, which is decrease TC or increase lookup gap error; so it's unrelated. |
Also, as an aside note; I've found this particular kernel https://github.com/exeminer/exeminer/blob/master/scrypt140202.cl to not cause any HW errors, and it's different from bufius, ckolivas, alexkarnew & alexkarold |
If you intend to merge this kernel into SGminer, please try to keep some consistency. We have different commits in master, 5.0 and developer. |
This is largely dependent on scrypt n-factor 11, which sucks balls anyway; for the time being, I shall close this. |
When building a bin file for scrypt kernel [zuikkis and/or alexkarnew] on a Hawaii chipset R9 290/X architecture, sgminer5 throws this error.
[00:30:50] Probing for an alive pool
[00:30:51] Switching to NiceHash_Scrypt_backup - first alive pool
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:52] Initialising kernel alexkarnew.cl with bitalign, unpatched BFI, nfactor 10, n 1024
[00:30:54] Error -4: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)
[00:30:54] Error -4: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)
[00:30:54] Error -4: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)
[00:30:54] GPU 0 failure, disabling!
[00:30:54] GPU 2 failure, disabling!
[00:30:54] GPU 1 failure, disabling!
This issue is not found present when running scrypt bins under kalroth's cgminer.
The settings for the scrypt bin are as follows, which are the preferred/max settings for a Hawaii R9 290.
worksize 256, TC 32765 - bin's initialize just fine on cgminer and provide 995Kh/s
under this new sgminer - nothing but Error -4; no change in memory or architecture, as I can close sgminer and go directly over to cgminer
This is the current working alex scrypt bin file that I have scrypt130511_alexeyHawaiiglg2tc32765w256l4
I can get sgminer to build
alexkarnewHawaiiglg2tc32765w128l4, it never respects the pool-worksize setting.
And regardless, if you force it into 256, it still -4 Error.
The text was updated successfully, but these errors were encountered: