Intel GPUs Got 100 Times Slower Due To Coding Error

Intel announced the Arc GPUs as a revolutionary budget-focused card. We were equally excited that a third company was finally standing against the duopoly of Nvidia and AMD. But the soon-to-launch GPUs from Intel fall short of their promises. An incident happened where a coding error caused the GPUs to become 100 times slower than they actually were.

This happed while the update promised to make the GPUs 100 times faster than they were. A minor coding-related issue caused the GPU to perform poorly in Ray Tracing enabling tasks. This issue was fixed with a slight change to the codebase, after which the Arc GPU got back to working great again.

The code error with Intel GPUs

Unknowingly, the developers added a line of code that would save the instructions in slower memory instead of the GPU memory. As a result, the instructions were stored in system memory, which witnessed a massive delay in response. If they were stored on the GPU memory, the Ray Tracing would have performed much better because it would be more accessible.

Phoronix reported that the change to the open-source Intel Mesa Vulkan driver was made by Lionel Landwerlin. He fixed the issue with a single line of code which removed the sluggishness, and the tasks would now be allocated to the discrete GPU. The update will roll out to end users by the end of August.

Intel even compared the Arc GPU to the modern RTX 3060 and claimed it to be a winner by a considerable margin. But Nvidia has a decade worth of driver development under the belt, and its GPUs are generally compatible with most modern devices. Arc is trying to make a dent in the GPU market, but it won’t be able to usurp Nvidia’s dominance with a mediocre GPU like the latest and greatest Intel Arc.