Iris Xe Graphics: Number of Work-Items

[Juan_G_1]

Hello  my name is Juan

Recently I started to code with Data Parallel C++ but in the past I used to code with OpenCL. At this moment I have a computer with an Intel(R) UHD Graphics 630 and only has 256 work-items in each dimension. So, I was thinking of buying a new laptop with an Intel Iris Xe Graphics and my question is:

Does anyone know how many work-items has the Intel Iris Xe Graphics in each dimension? Would it give me better performance if I do scientific computing?

If there's another Intel Graphics that you recommend please let me know

-Juan

[ArpitaP_Intel]

Hi Juan,

Thanks for reaching out to us.

The work item size for each dimension is 512 for Intel Iris Xe Graphics. Please find the screenshot for the same

Regards,

Arpita

[Ben_A_Intel]

Hello,

Please note that the maximum work-group size isn't necessarily correlated with performance.  Some algorithms will benefit from a larger work-group size (say by having more work-items cooperate to perform some work), but all else being equal many algorithms will run just fine with smaller work-groups.

For Intel GPUs specifically a good number to pay attention to is the number of execution units (or "EUs"), which are the the hardware units that run kernels and perform computation.  Your Intel(R) UHD Graphics 630 GPU has 24 execution units.  Newer Xe GPUs can have 96 or more execution units, which can improve the performance of compute-bound applications.

For scientific computing specifically you may also want to choose a GPU with double-precision float support - there are other forum threads on this topic.

Hope this helps!

[Juan_G_1]

Hello Ben,

I really appreciate your help. I can see that the Xe GPU has a good amount of compute units. Do you know if the Xe GPUs has double-precision float support?

[ArpitaP_Intel]

Hi Juan,

Intel Iris Xe GPU supports double-precision floating point emulation only for Linux. For more details you can refer the below link

https://github.com/intel/compute-runtime/blob/master/opencl/doc/FAQ.md#feature-double-precision-emul...

Regards,

Arpita

[ArpitaP_Intel]

Hi Juan,

Please, let us know, if your issue is resolve.

Regards,

Arpita

[ArpitaP_Intel]

Hi Juan,

We assume that your issue is resolved. If you need any additional information, please submit a new question as this thread will no longer be monitored.

Thanks!