Anyone who has ever done programming can tell you that one mistake can have very serious consequences. So it is with one of the first drivers for the long-awaited Intel Arc Desktop GPUs. According to the official pull request for the Linux version of one open source driver, one line of code caused a bug that resulted in a 100x performance degradation in ray tracing. The bug was found and merged in the next release.
This is where things get a little technical. The bug was discovered by Intel engineer Lionel Landverlin and posted to the public Mesa gitlab repository. According to his notes, the previous version of Vulkan’s ray tracing implementation used system memory (as in main RAM attached to the motherboard) instead of local memory (GDDR6 RAM soldered directly to the graphics card). Of course, speeding up graphics processing is primarily the point of having memory on a graphics card, so it’s kind of a big miss from the point of view of any graphics driver.
Landverlin reported that with a single line of code change, there is “almost a 100x (no joke) improvement” in ray tracing performance using Vulkan on Linux. Which is not surprising, since the driver now assigns tasks to the memory that is actually dedicated to those tasks, rather than the much more common system RAM. The change has been approved and included in the next release of the Mesa driver, according to Phoronix.
This little episode shows just how far Intel is behind its new competitors in the discrete graphics market. Nvidia and AMD have years of experience writing and tweaking discrete graphics drivers, not only for PCs in general, but also to tweak and improve performance for specific graphics APIs (and sometimes even individual games). It is impossible to argue that the established giants of the industry would not have made the same mistake, especially since we are talking about Linux here. But it’s easy to point to this one-liner as an example of Intel’s immaturity with graphics drivers.
Some of Intel’s latest press releases reflect this. The company hopes that an aggressive, competitively priced, three-tier strategy will help mitigate its poor optimization for some games when it launches globally, with a US release scheduled for later this summer. In particular, Arc GPUs will be priced to compete with similar graphics cards based on the lowest level of performance they can achieve in popular games, not the highest.
That said, the fact that Intel was able to fix this problem long before the international release of its drivers is reassuring. While there is room for improvement in this area, it is, as you know, Intel. A multi-trillion-dollar international mega-corporation can more or less buy its way into a new market if it wants to, although that doesn’t mean it will succeed. If Intel can keep up with improvements in its software and development teams, we can expect a much more even race in the desktop GPU market in a year or two.