Coding Mistake Made Intel GPUs 100X Slower in Ray Tracing

Intel Linux GPU driver builders have launched an replace that ends in a large 100X increase in ray tracing efficiency. That is one thing to be celebrated, after all. Nevertheless, on the flip aspect, the motive force was 100X slower than it ought to have been due to a reminiscence allocation oversight. The information comes amid reviews that Intel’s transport drivers for its Arc GPUs are fraught with points in Home windows which are akin to “[…]residing in the course of a minefield – thoughts you, whereas taking part in drunk.” The corporate has additionally admitted that Arc efficiency is sub-par with older APIs, like DX11, in Home windows. 

Linux-centric information web site Phoronix reviews {that a} repair merged into the open-source Intel Mesa Vulkan driver was applied by Intel Linux graphics driver engineering stalwart Lionel Landwerlin on Thursday. The developer wryly commented that the merge request, which already landed in Mesa 22.2, would ship “Like a 100x (not joking) enchancment.” Intel has been engaged on Vulkan raytracing help since late 2020, however this repair is healthier late than by no means.

Even readers who aren’t (driver) builders will in all probability have the ability to perceive the difficulty with the earlier code. Often, the Vulkan driver would guarantee non permanent reminiscence used for Vulkan raytracing work could be in native reminiscence, i.e., the very quick graphics reminiscence onboard the discrete GPU.