|
|
|
|
|
|
by 0xDEADFED5
544 days ago
|
|
|
Big agree on Intel working on SYCL. I've millions of tasks thru SYCL llama.cpp at this point, and though SYCL reliably does 5-6x the prompt processing speed of the Vulkan builds, current Vulkan builds are now up to 50% faster at token generation than SYCL on my Intel GPU |
|
|