Browse Source

Extensively document the cause of GPU device issues and the use of --gpu-map.

nfactor-troky
Con Kolivas 13 years ago
parent
commit
6b0d767aca
  1. 78
      README

78
README

@ -184,6 +184,7 @@ GPU only options:
--gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity (default: 7) --gpu-dyninterval <arg> Set the refresh interval in ms for GPUs using dynamic intensity (default: 7)
--gpu-engine <arg> GPU engine (over)clock range in Mhz - one value, range and/or comma separated list (e.g. 850-900,900,750-850) --gpu-engine <arg> GPU engine (over)clock range in Mhz - one value, range and/or comma separated list (e.g. 850-900,900,750-850)
--gpu-fan <arg> GPU fan percentage range - one value, range and/or comma separated list (e.g. 25-85,85,65) --gpu-fan <arg> GPU fan percentage range - one value, range and/or comma separated list (e.g. 25-85,85,65)
--gpu-map <arg> Map OpenCL to ADL device order manually, paired CSV (e.g. 1:0,2:1 maps OpenCL 1 to ADL 0, 2 to 1)
--gpu-memclock <arg> Set the GPU memory (over)clock in Mhz - one value for all or separate by commas for per card. --gpu-memclock <arg> Set the GPU memory (over)clock in Mhz - one value for all or separate by commas for per card.
--gpu-memdiff <arg> Set a fixed difference in clock speed between the GPU and memory in auto-gpu mode --gpu-memdiff <arg> Set a fixed difference in clock speed between the GPU and memory in auto-gpu mode
--gpu-powertune <arg> Set the GPU powertune percentage - one value for all or separate by commas for per card. --gpu-powertune <arg> Set the GPU powertune percentage - one value for all or separate by commas for per card.
@ -217,7 +218,7 @@ FPGA mining boards(BitForce, Icarus) only options:
(where n = the correct device number for the FPGA device) (where n = the correct device number for the FPGA device)
CPU only options: CPU only options (deprecated, not included in binaries!):
--algo|-a <arg> Specify sha256 implementation for CPU mining: --algo|-a <arg> Specify sha256 implementation for CPU mining:
auto Benchmark at startup and pick fastest algorithm auto Benchmark at startup and pick fastest algorithm
@ -836,6 +837,81 @@ miner.php - an example web page to access the API
Read the top of the file (miner.php) for details of how to tune the display Read the top of the file (miner.php) for details of how to tune the display
and also to use the option to display a multi-rig summary and also to use the option to display a multi-rig summary
---
GPU DEVICE ISSUES and use of --gpu-map
GPUs mine with OpenCL software via the GPU device driver. This means you need
to have both an OpenCL SDK installed, and the GPU device driver RUNNING (i.e.
Xorg up and running configured for all devices that will mine on linux etc.)
Meanwhile, the hardware monitoring that cgminer offers for AMD devices relies
on the ATI Display Library (ADL) software to work. OpenCL DOES NOT TALK TO THE
ADL. There is no 100% reliable way to know that OpenCL devices are identical
to the ADL devices, as neither give off the same information. cgminer does its
best to correlate these devices based on the order that OpenCL and ADL numbers
them. It is possible that this will fail for the following reasons:
1. The device order is listed differently by OpenCL and ADL (rare), even if the
number of devices is the same.
2. There are more OpenCL devices than ADL. OpenCL stupidly sees one GPU as two
devices if you have two monitors connected to the one GPU.
3. There are more ADL devices than OpenCL. ADL devices include any ATI GPUs,
including ones that can't mine, like some older R4xxx cards.
To cope with this, the ADVANCED option for --gpu-map is provided with cgminer.
DO NOT USE THIS UNLESS YOU KNOW WHAT YOU ARE DOING. The default will work the
vast majority of the time unless you know you have a problem already.
To get useful information, start cgminer with just the -n option. You will get
output that looks like this:
[2012-04-25 13:17:34] CL Platform 0 vendor: Advanced Micro Devices, Inc.
[2012-04-25 13:17:34] CL Platform 0 name: AMD Accelerated Parallel Processing
[2012-04-25 13:17:34] CL Platform 0 version: OpenCL 1.1 AMD-APP (844.4)
[2012-04-25 13:17:34] Platform 0 devices: 3
[2012-04-25 13:17:34] 0 Tahiti
[2012-04-25 13:17:34] 1 Tahiti
[2012-04-25 13:17:34] 2 Cayman
[2012-04-25 13:17:34] GPU 0 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 1 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 2 AMD Radeon HD 6900 Series hardware monitoring enabled
[2012-04-25 13:17:34] 3 GPU devices max detected
Note the number of devices here match, and the order is the same. If devices 1
and 2 were different between Tahiti and Cayman, you could run cgminer with:
--gpu-map 2:1,1:2
And it would swap the monitoring it received from ADL device 1 and put it to
opencl device 2 and vice versa.
If you have 2 monitors connected to the first device it would look like this:
[2012-04-25 13:17:34] Platform 0 devices: 4
[2012-04-25 13:17:34] 0 Tahiti
[2012-04-25 13:17:34] 1 Tahiti
[2012-04-25 13:17:34] 2 Tahiti
[2012-04-25 13:17:34] 3 Cayman
[2012-04-25 13:17:34] GPU 0 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 1 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 2 AMD Radeon HD 6900 Series hardware monitoring enabled
To work around this, you would use:
-d 0 -d 2 -d 3 --gpu-map 2:1,3:2
If you have an older card as well as the rest it would look like this:
[2012-04-25 13:17:34] Platform 0 devices: 3
[2012-04-25 13:17:34] 0 Tahiti
[2012-04-25 13:17:34] 1 Tahiti
[2012-04-25 13:17:34] 2 Cayman
[2012-04-25 13:17:34] GPU 0 AMD Radeon HD 4500 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 1 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 2 AMD Radeon HD 7900 Series hardware monitoring enabled
[2012-04-25 13:17:34] GPU 3 AMD Radeon HD 6900 Series hardware monitoring enabled
To work around this you would use:
--gpu-map 0:1,1:2,2:3
--- ---
FAQ FAQ

Loading…
Cancel
Save