4 ms·
Yes, experts are real. Unfortunately, experts are only expert in a few subjects, and experts (well... all humans even) are bad at knowing when they've left the
by dragontamer 6y ago
Yes, experts are real.
Unfortunately, experts are only expert in a few subjects, and experts (well... all humans even) are bad at knowing when they've left their field of expertise.
Linus Torvalds for example is clearly an expert programmer. And then suddenly he talks about AVX512 and I wonder if he knows anything about low-level programming or optimization. (https://www.realworldtech.com/forum/?threadid=193189&curpostid=193190 https://www.realworldtech.com/forum/?threadid=193189&curpost...).
Case in point: with Rocket lake (on 14nm nonetheless), the power-limits associated with AVX512 have already been largely lifted. It didn't even take a few months before Linus's words became obsolete.
https://travisdowns.github.io/blog/2020/08/19/icl-avx512-freq.html https://travisdowns.github.io/blog/2020/08/19/icl-avx512-fre...
100MHz difference single-core, 0MHz difference multicore. Virtually no penalty for using AVX512 on new systems. Linus was ultimately trapped in the Skylake-X way of thinking, and not looking forward to the next chips.
- zug_zug 6y agoDid you read the article? The article is specifically talking about cases where experts are incorrect within their field of expertise and then cited as a source of truth in published papers. The question of the article isn't really a boolean, it clearly shows some experts are real and some aren't and asks the more important question -- how do we discern when the institutionalized "expert" in an area is really an expert?
- dragontamer 6y ago> Did you read the article? The article is specifically talking about cases where experts are incorrect within their field of expertise and then cited as a source of truth in published papers. Yes. But I still think that Linus is a good character to discuss within the scope of the article. Linus is clearly considered an expert, and many people follow his bombastic words (retweeting, or using them to form the basis of their argument around here). "Because Linus said so" is a common way to end various arguments. Linus's opinion is truly "expert", people copy it without further analysis because... well... Linus kinda deserves it. He is an expert after all. It doesn't mean Linus is always right. -------- The "feedback loop" of the open source world is pretty quick. If you're a bad programmer, people figure it out when they start working with your code. But the "feedback loop" of "chip ISA opinions" is slow. There's not really any way to prove, or disprove, whether or not AVX512 is a good or bad idea. Linus's expertise in programming doesn't necessarily translate to chip ISA design. And as such, Linus can get details (expected power-characteristics of future chips) dramatically wrong.
- vlovich123 6y agoI think you’re over-focusing on the AVX512 argument. You even said yourself it was a product release that happened months after he made the comments. Was he supposed to understand what changes Intel made to AVX512 before they announced it? Also, he’s not even talking about clock downscaling. He’s saying that AVX512 is the wrong way to get significantly more performance vs the amount of space and power it consumes (definitely true for most applications and 100% true for kernel work) and that Intel would be better served focused on general perf (which isn’t wrong given how they’ve been trounced by AMD and Apple). Consider also that Linus worked at Transmetta for a relatively long time trying to build a CPU company. I would trust his opinions on chip ISA design more than most software developers. You’ve basically taken his post out of context, misinterpreted it to be saying something he isn’t, and claiming it’s wrong because some details have changed on the ground after he made it. This is the problem with expert opinion. It’s so easy to misinterpret. It’s also possible to get some details wrong without invalidating the larger point. Figuring out when the expert is wrong is hard, but claiming that Linus doesn’t understand chip design just factually doesn’t understand the context of Linus’ perspective and his actual professional experience.
- dragontamer 6y ago> Also, he’s not even talking about clock downscaling Quote the post in question: > not with some AVX512 power virus that takes away top frequency (because people ended up using it for memcpy!) This is pretty obviously talking about the down-clocking issue. > You’ve basically taken his post out of context, misinterpreted it to be saying something he isn’t, and claiming it’s wrong because some details have changed on the ground after he made it. AVX on Sandy Bridge was worthless: it wasn't any faster than SSE because Sandy Bridge was still 128-bit (even on AVX code). Anyone who has followed ISA-development for decades would recognize that the first implementation of an ISA is often sub-par and sub-optimal. It wasn't until Haswell before AVX was faster than SSE in practice. AVX512 was first implemented on Xeon Phi. Then it was ported over to servers in Skylake-X. AVX512 from there on remained a "1st generation" product for a few years (much like AVX on Sandy Bridge, before Haswell). From the history of these ISA-extensions, we generally can expect downclocking issues or inefficiencies in the first few implementations. ----- In fact, some portions of AVX still remain suboptimal. vgather for example, isn't really much faster than standard load/stores. Maybe they'll be suboptimal forever, or maybe some hypothetical future CPU could optimize it. Either way, I know that I shouldn't make an argument based on the "stupidity of vgather" instruction. There's an "obvious" optimization that could happen if that instruction becomes popular enough to deserve the transistors. At best, if I were to argue about vgather instructions, I'd preface it with "current implementations of vgather on Skylake-X", or similar wording. CPUs change over time, much like programming languages or compilers. I expect a degree of forward-looking ability from experts, not for someone to get lost in the details of current implementations.
- exmadscientist 6y agoIf you read his argument as less about the power virus effect and more about the uselessness of AVX512 for most general-purpose integer code, and accordingly about the opportunity cost of developing and fabricating AVX512 for clients who really will not use it, then, unfortunately, he's right. The quote "I'd much rather see that transistor budget used on other things that are much more relevant," in particular, is the key to the argument. Because that seems to be basically exactly what the Apple M1 does, and look what people are saying about it.
- dragontamer 6y agoBut Linus is talking about power-budget. And Ice-lake / Rocket lake are proving that AVX512 is well within the power-budget of these chips. M1 dedicates more chip-area to integer units, and also uses less power (thanks to the 5nm advantage). So what's going on with M1 isn't really a power-issue at all. When Linus says things like: > 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). Its clear that he was unable to be forward thinking. He was unable to see that a future chip, like Icelake or Rocket Lake, could fix the AVX512 power-throttling so dramatically. (without even going to an advanced node !!!).
- topspin 6y ago> But Linus is talking about power-budget. Linus covered a lot of ground in that message. He was correct on his history, whether that is history you'd like to remember or not. His transistor budget argument is very plausible, but without playing out an alternative history we can't know, and by we I mean you. At the time he wrote it he was correct on the power 'virus,' and I can't think of any good reason why he or anyone else owes Intel the benefit of the doubt that some future device would improve anything. The AVX512 fanboi contingent will forever seethe about what Linus wrote. Meanwhile Apple and AMD are eating Intel's lunch with no 512 bit SIMD instruction units at all. Linus had it right and and he is looking more correct every day.
- rhino369 6y agoExpert opinion isn't really worth anything until it's been stress tested by other experts. I worked with experts in the context of patent and trade secret litigation. I pick people at the top of their field from prestigious universities with impeccable credentials. If you ask them off the cuff questions they often are wrong. But if they investigate and think about it, they'll correct themselves. Experts are also just as susceptible to bias, motivated reasoning, groupthink, etc. as anyone else. This is why academic papers are supposed to be a back and forth discussion. "TRUST THE EXPERTS" has a plural for reason, and should be understood to mean trust the experts after they've done a thoughtful analysis with the best available evidence and have considered dissenting expert views.
- hebrox 6y agoI think term "expert opinion" here is key. We want the results of the work of experts, not their opinion.
- rhino369 6y agoOther experts should look at their results and their work. Lay people should be concerned with their opinion, but only after other experts have looked at the results/work.