Forum > GPU crunching

Intel HD GPU crunching

<< < (2/3) > >>

benool:
Speaking of which, anyone has a Pentium G3258 ?

I'm running Intel GPU apps on one and while working fine I get some strange entries in the logs namely regarding frequencies and raw performance.
I should add that the G3258 is overclocked at 4.2GHz.

Example of a recent task: http://setiathome.berkeley.edu/result.php?resultid=3815477962

Already the reported "Device peak FLOPS   16.00 GFLOPS" seems quite low.


--- Code: --- OpenCL Platform Name: Intel(R) OpenCL
Number of devices: 1
  Max compute units: 10
  Max work group size: 256
  Max clock frequency: 200Mhz
  Max memory allocation: 353579827
  Cache type: Read/Write
  Cache line size: 64
  Cache size: 2097152
  Global memory size: 1414319309
  Constant buffer size: 65536
  Max number of constant args: 8
  Local memory type: Scratchpad
  Local memory size: 65536
  Queue properties:
    Out-of-Order: No
  Name: Intel(R) HD Graphics
  Vendor: Intel(R) Corporation
  Driver version: 10.18.10.3960
  Version: OpenCL 1.2
  Extensions: cl_intel_accelerator cl_intel_ctz cl_intel_d3d11_nv12_media_sharing cl_intel_dx9_media_sharing cl_intel_motion_estimation cl_intel_simultaneous_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_khr_depth_images cl_khr_dx9_media_sharing cl_khr_gl_depth_images cl_khr_gl_event cl_khr_gl_msaa_sharing cl_khr_gl_sharing cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_icd cl_khr_image2d_from_buffer cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_spir
--- End code ---

The funny bit is:  Max clock frequency: 200Mhz
while in GPU-z it shows frequency up to 1150MHz.


On another host using an i7 3770, we can compareto say with this task http://setiathome.berkeley.edu/result.php?resultid=3818166251 :
The i7 entry is also showing: Device peak FLOPS   147.20 GFLOPS

G3258 - Run time   1 hours 23 min 2 sec
i7 3770 - Run time   1 hours 4 min 30 sec


Not so bad given a 16 vs 147 GFLOPS performance!

So just wondering what's happening here.

Richard Haselgrove:
You're using the 10.18.10.3960 driver. There have been reports at both Einstein (http://einstein.phys.uwm.edu/forum_thread.php?id=10955) and SETI (http://setiathome.berkeley.edu/forum_thread.php?id=76036) of problems with that driver - it may explain some of the odd readings too. Have your tasks been validating reliably?

benool:

--- Quote from: Richard Haselgrove on 06 Nov 2014, 11:41:45 am ---Have your tasks been validating reliably?
--- End quote ---
Yes, I don't have tasks failing or not validating.

benool:
By the way are there any guidelines regarding the amount of memory to allocate for iGPU?

I'm talking about options in the BIOS letting one choose among values like 128, 224, 256 MB to reserve for integrated GPU.
Is there any actual influence in terms of crunching speed ?

Raistmer:

--- Quote from: benool on 17 Nov 2014, 04:53:46 pm ---By the way are there any guidelines regarding the amount of memory to allocate for iGPU?

I'm talking about options in the BIOS letting one choose among values like 128, 224, 256 MB to reserve for integrated GPU.
Is there any actual influence in terms of crunching speed ?

--- End quote ---

If memory amount not enough modern WDM driver allocates additional memory from system memory. Perhaps that additional memory should somehow be of less performance (possibility of different caching scheme), but AFAIK no-one did robust testing in this area.
Would be interesting to consider 2 extreme cases (64 and 256 M ) with 2 or more tasks in play. with total amout of memory reported by app (-v 8 switch for OpenCL)  exceeded amount of dedicated memory set in BIOS.

Navigation

[0] Message Index

[#] Next page

[*] Previous page