ZECMiner
ZECMiner copied to clipboard
Genoil
Will there be support for non GCN cards?
Hi,
I would like to use my old 6870s to build a small rig but when I try to run this miner I see OC errors about creating the buffer an it just shows 0 in all the stats.
As I saw that the support for GCN 1.0 is being released now Im curious to know if even older GPU's will be supported.
Thanks in advance!
The reason is it tries to allocate around 1664 MB of video memory for its working process.
Just look at the function at 13FB46C90, see a screenshot (in blue rectangles):
Do you see these two calls to clCreateBuffer and then failing (exception raising) in case of error during these calls? The miner calls it twice (at 13FB46EAE and 13FB46F4A) with 0x34000000 as "size" argument (at R8, as a third argument). So, 0x34000000 = 872415232. (872415232 * 2) / 1024 / 1024 = 1664 (as initial argument is in bytes). And we get the "zec-sa#0clCreateBuffer(-61)" (clCreateBuffer error -61), which translates to "GPU can't allocate the DAG in a single chunk".
So we need at least 2GB of video card memory here for it to work. No luck for us here (I have AMD HD8490 with only 1 GB).
Hopefully that is not principal algorithm limitation. As of now i have to use NiceHash miner. But it seems somewhat slow.
With NiceHash I have another problem and is that the GPU is at 100% but no jobs are sent back to the pool thats why I wanted to use this miner instead.
Which miner server u use? NiceHash or flypool or other? I run against NiceHash server and it works. But is uses a non-standard server protocol (extra-nonce) so probably it can only communicate with its own server, not with flypool or anything else.
That said, in Saturday night there were few hours when there was no connection to Nice servers, next day it was fixed though. I'll check when at home if it is still hammering there....
I wonder... All videocards since AGP can address system memory.
So i wonder, even if those large buffers are required, maybe seing lesst han 2GB videoram the miner could allocate 800MB buffer in OpenCL-GPU and another 800MB in OpenCL-CPU and then flip-flop them when needed.....
next release will support lower memory usage, maybe it works then but i doubt
