Message Boards Message Boards

2 Replies
3 Total Likes
View groups...
Share this post:

What does the Core Count returned by OpenCLInformation refer to?

Posted 11 years ago
What does the "Core Count" returned by OpenCLInformation[] refer to?
For a particular device, it returns 16. The manufacturer specifications say 384 cores. Note that "Core Count" is not the same as "Compute Units" (which is 2 for this device). Most of the information returned directly corresponds to something the clGetDeviceInfo OpenCL API function can tell me, but I'm not sure what "Core Count" could correspond to.
So what is the precise meaning for "Core Count" here?

For comparison, CUDAInformation[] returns "Core Count: 64", "Multiprocessor Count: 2" for the same device.
I think this information might not have much practical relevance, but I'm not sure as it's the most prominently presented value in the Links/OpenCL section of SystemInformation[].

Here's the full information returned for this device:

 Name->GeForce GT 650M
 Version->OpenCL 1.2
 Driver Version->8.18.22 310.40.05f01
 Vendor ID->16918272
 Compute Units->2
Core Count->16
Maximum Work Item Dimensions->3
Maximum Work Item Sizes->{1024,1024,64}
Maximum Work Group Size->1024
Preferred Vector Width Character->1
Preferred Vector Width Short->1
Preferred Vector Width Integer->1
Preferred Vector Width Long->1
Preferred Vector Width Float->1
Preferred Vector Width Double->1
Maximum Clock Frequency->900
Address Bits->32
Maximum Memory Allocation Size->268435456
Image Support->True
Maximum Read Image Arguments->256
Maximum Write Image Arguments->16
Maximum Image2D Width->16384
Maximum Image2D Height->16384
Maximum Image3D Width->2048
Maximum Image3D Height->2048
Maximum Image3D Depth->2048
Maximum Samplers->32
Maximum Parameter Size->4352
Memory Base Address Align->1024
Memory Data Type Align Size->128
Floating Point Precision Configuration->{Denorms,Infinity,NaNs,Round to Nearest,Round to Infinity,Round to Zero}
Global Memory Cache Type->None
Global Memory Cache Line Size->0
Global Memory Cache Size->0
Global Memory Size->1073741824
Maximum Constant Buffer Size->65536
Maximum Constant Arguments->9
Local Memory Type->Local
Local Memory Size->49152
Error Correction Support->False
Profiling Timer Resolution->1000
Endian Little->True
Compiler Available->True
Execution Capabilities->{Kernel Execution}
Command Queue Properties->{Profiling Enabled}
POSTED BY: Szabolcs Horvát
2 Replies
I will pass some info from my colleagues. It corresponds to the "marketing" value for the GPU (a multiple of CL_DEVICE_MAX_COMPUTE_UNITS ). It should not really be used. It is explained with a little more detail here:

Number of Compute Units corresponding to the number of work groups
POSTED BY: Vitaliy Kaurov
Some additional information I've been able to obtain is that the OpenCL API doesn't provide a way to query the number of hardware processing elements.  This makes me even more curious to find out what this "core count" refers to.
POSTED BY: Szabolcs Horvát
Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
or Discard

Group Abstract Group Abstract