Why Torvalds Wishes AVX-512 a “Painful Death”

Linus Torvalds, the man behind the development of Linux, criticized “Advanced Vector Extensions 512 (AVX-512)” in an email sent to his mailing list. He appears to deem that the problem lies in the direction toward which Intel is heading and not the technology.

Advanced Vector Extensions 512 (AVX-512), the successor of AVX2, is a set of SIMD extensions. With 512-bit vector processing capabilities, it is available on Intel Xeon Phi processor and co-processor and Intel Xeon Scalable processor. According to Intel website, AVX-512 “can accelerate performance for workloads and usages such as scientific simulations, financial analytics, artificial intelligence (AI)/deep learning, 3D modeling and analysis, image and audio/video processing, cryptography and data compression.”

Torvalds’ harsh comment on AVX512 is based on the article “GCC 11 Compiler Lands Intel Sapphire Rapids + Alder Lake Support” posted on Phoronix, a website that offers news regarding free/open-source software. The article pointed out that while the upcoming “GNU Compiler Collection 11” is going to support Intel Alder Lake and Sapphire Rapids, these don’t support AVX-512 instructions set.

In the email sent to his mailing list, regarding this article, he said, “Hope you didn’t get too attached to AVX-512. The GCC 11 compiler target for Alder Lake doesn’t enable it, only AVX2,” and added that “I hope AVX 512 dies a painful death.”

He revealed his dislike for AVX512 saying that “I hope AVX512 dies a painful death, and that Intel starts fixing real problems instead of trying to create magic instructions to then create benchmarks that they can look good on.” He went on to say that he hopes “Intel gets back to basics: gets their process working again, and concentrate more on regular code that isn’t HPC or some other pointless special case.”

Furthermore, Torvalds pointed out that even in the heyday of x86, when Intel was monopolizing the market, Intel’s competitors always had better FP performance, and wrote that “The same is largely true of AVX512 now – and in the future.” He demanded that “I’d much rather see that transistor budget used on other things that are much more relevant. Even if it’s still FP math (in the GPU, rather than AVX512). Or just give me more cores (with good single-thread performance, but without the garbage like AVX512) like AMD did.” “I want my power limits to be reached with regular integer code, not with some AVX512 power virus that takes away top frequency (because people ended up using it for memcpy!) and takes away cores (because those useless garbage units take up space).”

“Yes, yes, I’m biased. I absolutely detest FP benchmarks, and I realize other people care deeply. I just think AVX512 is exactly the wrong thing to do.” “It’s a prime example of something Intel has done wrong, partly by just increasing the fragmentation of the market”

Torvalds demanded Intel to work on the “core common stuff” instead of other special cases and ended his message with “Yeah, I’m grumpy.”

The email Linus Torvalds sent to his mailing list
https://www.realworldtech.com/forum/?threadid=193189&curpostid=193190
The article on Phoronix
https://www.phoronix.com/scan.php?page=news_item&px=GCC11-SapphireRapids-AlderLake