Announcement

Collapse
No announcement yet.

Intel Updates Alder Lake Tuning For GCC, Reaffirms No Official AVX-512

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • coder
    replied
    Originally posted by avem View Post
    We had a separate topic about ADL, I wonder why you weren't as vocal over there instead of here while arguing with seemingly "an invested person".
    The only posts you ultimately control are your own. You can try to flag others' posts, but most attempts by ordinary users to police the forums only make the situation worse.

    Originally posted by avem View Post
    Sorry, I have zero shares of any companies in the world, I'm broke and broken. I've never dealt with major tech companies either aside from using their products.
    Sorry to hear that. I genuinely don't harbor any ill will towards you, but perhaps you can understand why it arouses suspicion when someone seems to advocate/defend a product a bit too vigorously.

    I know a few ex-Intel employees, but that has far less influence on how I view the company than my personal experience with their products. And what I can say about that is that they're reliable and generally do what they claim. When I use Intel products, I also experience very few issues (i.e. stuff tends to work right out of the box).

    How's this for not being an Intel-hater: I even bought an Intel CPU in the "bad old days" of the Pentium 4. Specifically, it was a Prescott.

    Originally posted by avem View Post
    Thirdly, I wonder why you criticise Intel for not including AVX-512 so much.
    It's a point of discussion. There are various aspects of this worth talking about, and some important issues people should be aware of (e.g. the lack of validation).

    Originally posted by avem View Post
    You could make it more personal, instead of making it look like it's some sort of a horrible oversight, a major shortcoming/problem, e.g. "At my work we have uses for AVX-512 and it's sad Intel hasn't made it available despite the previous rumors".
    Again, you're coming across as too defensive. If what I say is true or a legitimate opinion, then it shouldn't matter whether it's directly tied to personal experience.

    Leave a comment:


  • coder
    replied
    Originally posted by avem View Post
    Rocket Lake should have never been released. Of course, it's a major feat retrofitting a 7/10nm uArch to the 14nm node but other than that it was a failure which I guess everyone will quickly forget about. So, again, Intel has yet to make AVX-512 a first class citizen in their desktop parts.
    Be that as it may, it did happen and it does exist. Other than on the efficiency front, it really wasn't bad. Intel did not hold back on selling it, either. I really don't see how you can continue to claim that AVX-512 on the desktop didn't happen.

    Originally posted by avem View Post
    As for ICL/TGL AVX-512 - I guess no one uses laptops for their AVX-512 prowess. Anyways, the conversation about AVX-512 almost looks stupid since we don't even have benchmarks for it and the number of common tasks benefitting from this instruction set seems to be zero despite it having been available for more than three years.
    Why do you say we don't have benchmarks? Michael already posted one benchmark article on it, and the Anandtech article you quoted from includes others. PTS actually has quite a few benchmarks that use it. There are even some deep learning benchmarks in there, such as OpenVINO.

    We can agree that AVX-512 (or the lack thereof) is not terribly consequential for the typical consumer. However, Phoronix users are not typical consumers and this thread is a legitimate place to discuss the pros/cons of the decision.

    Leave a comment:


  • coder
    replied
    Originally posted by avem View Post
    I'll gladly discuss the pros and cons of ADL, I just don't understand why AMD has a cult following seemingly everywhere and that makes arguing quite difficult.
    I think you're seeing the cult of AMD even in places where it doesn't exist. If you view every time someone raises concerns, misgivings, or otherwise points out a negative about some Intel product as coming from a rabid AMD fan, then we have a problem.

    Originally posted by avem View Post
    Pros:
    • A massive IPC increase
    • Very competitive performance despite having fewer threads than comparable AMD parts
    • An option of using AVX-512 at the expense of E-cores
    • You can use DDR4 with it without losing performance in most applications
    • PCI-E 5.0
    • Agree on IPC increase & overall performance.
    • Disagree on AVX-512. If it's not validated and not guaranteed to be available, then people should not count on it. I regard it almost as though it's not even there.
    • Agree on DDR4 option. That seems to have been a wise choice. Too bad motherboard selection forces you to go with one or the other, and that motherboards for it are very expensive (see PCIe 5). Actually, expensive motherboards should be another Con.
    • PCIe 5 is very much an open question, in my book. There's nothing which can take advantage of it, and we might find a raft of issues with various motherboards, once the feature finally does start getting used in anger (which possibly won't happen until after the launch of Raptor Lake). Recall how AMD had to walk back their promise of PCIe 4.0 support on legacy boards, when Ryzen 3000 launched, and how Intel had to reverse course on including PCIe 4.0 in Comet Lake. I'm expecting a similar debacle with PCIe 5.0 in Alder Lake. And even if it works, PCIe 4.0 is already more than enough for today's fastest graphics cards. Therefore, I don't count it as adding much/any value. And to the degree it's adding cost to motherboards, it can really be seen as a Con.
    Originally posted by avem View Post
    Cons:
    • Overclocked out of the box and an insane power consumption at full MT load (can be partially mitigated by setting PL1/PL2) limits
    • Actual TDP is much higher than advertised (almost twice as high for 12900K, 125W vs 241W)
    • AVX-512 is again delayed by a generation or two because E-cores don't support it
    • High-speed DDR5 costs an arm and leg
    • Requires OS support for proper functioning, currently works terribly under Linux
    • Agreed on most points, except...
    • I see you're still in denial that AVX-512 is a retraction, not a delay.
    • Some people clearly see Big.Little as a mistake. In spite of some short-term pain, I don't (provided the OS is scheduling it reasonably). We'll see how things shape up, in the coming year.
    • Expensive motherboards. Not sure how much of this is due to PCIe 5, but the electronics industry press is full of articles about the added costs & challenges it poses.

    Originally posted by avem View Post
    Other thoughts:
    • An anemic iGPU and most likely Intel will not release desktop parts with more EUs. Would be great if Intel designed and released desktop parts with more EUs at the expense of P-cores
    • The maximum power efficiency for 12900K is between 160-180W which probably Intel should have targeted instead of 241W
    This looks much better than saying "ADL is bad", "Intel sucks" or "Insane power consumption!!"
    I don't especially mind the small iGPU. It's still bigger than what was included through the Gen 10 CPUs, and Michael posted benchmarks showing it out-performs Rocket Lake's GPU of the same size. The fact that they put a much bigger iGPU in the laptop parts doesn't oblige them to do the same for desktops. Then again, I should note that one of my desktops is still using the processor graphics of Sandybridge. So, I'm clearly not a very demanding GPU user.

    Leave a comment:


  • Rabiator
    replied
    Originally posted by avem View Post
    As for ICL/TGL AVX-512 - I guess no one uses laptops for their AVX-512 prowess. Anyways, the conversation about AVX-512 almost looks stupid since we don't even have benchmarks for it and the number of common tasks benefitting from this instruction set seems to be zero despite it having been available for more than three years.
    AVX-512 support is awfully fragmented in the sense of "some Intel CPUs support it, others don't".
    I guess that makes it less attractive for the mass market, as you would always have to include and test a fallback code path, perhaps for AVX-256. Also, marketing would have to be restrained to praising AVX-256 performance, otherwise complaints by people with CPUs missing AVX-512 would be guaranteed.
    Alder Lake pushes things into the direction of "no support" again. Because there is a workaround now (switching off the efficiency cores), but you cannot be sure it will be working in future revisions.

    Some applications that profit from really wide SIMD instructions use AVX-512 to good success. There are actually benchmarks, for instance at Phoronix.
    About Windows, I have read some CPU reviews where Intel won the benchmarks in number-crunching application that use AVX-512. (In other applications, Ryzen was leading). So it might still be a good choice for professional use where people (hopefully) know what they want to run and buy an appropriate CPU.
    Edit:
    I think you cannot fully discount laptop use. Laptops are increasingly common among software developers too, and I think those who are working with AVX-512 would appreciate if they can compile and test on their laptop too. Sometimes you might be at a customer site and have no remote access to the office.
    Last edited by Rabiator; 12 November 2021, 11:04 AM.

    Leave a comment:


  • avem
    replied
    Originally posted by coder View Post
    Their Big.Little and AVX-512 are controversial decisions. Some debate of these decisions is entirely natural and should be expected.

    Beyond that, there are legitimate concerns about its efficiency.

    What I don't get is why you seem so threatened by these discussions? Why are you trying to shut them down? Do you have some vested interest that you've not disclosed? Otherwise, you are as entitled to your opinions (and as capable of backing them up with data) as anyone else.


    Right in the article's title: "Reaffirms No Official AVX-512"

    AVX-512 is On. Fucking. Topic.

    Deal with it.
    We had a separate topic about ADL, I wonder why you weren't as vocal over there instead of here while arguing with seemingly "an invested person". Sorry, I have zero shares of any companies in the world, I'm broke and broken. I've never dealt with major tech companies either aside from using their products.

    Secondly, if AVX-512 is not ever there officially, let's just move on, OK? 99.999% of people out there don't know what it is and have no tasks to which could benefit from it.

    Thirdly, I wonder why you criticise Intel for not including AVX-512 so much. You could make it more personal, instead of making it look like it's some sort of a horrible oversight, a major shortcoming/problem, e.g. "At my work we have uses for AVX-512 and it's sad Intel hasn't made it available despite the previous rumors".

    You see, it sounds and looks a ton better and it doesn't make Intel look like a crap company which is there to fuck everyone over.

    I'm threatened by the amount of outright hostility, negativity and fanboyism in all threads related to Intel AMD or NVIDIA. It sucks to see it, it suck to participate in it. Ciao.

    Leave a comment:


  • avem
    replied
    Originally posted by coder View Post
    Rocket Lake is a mainstream consumer desktop CPU, launched in March of this year. Tiger Lake is a mainstream consumer mobile part (launched > 1 year ago), with no contemporaneous alternative. Their entire 11th gen series had AVX-512, both mobile and desktop.
    Rocket Lake should have never been released. Of course, it's a major feat retrofitting a 7/10nm uArch to the 14nm node but other than that it was a failure which I guess everyone will quickly forget about. So, again, Intel has yet to make AVX-512 a first class citizen in their desktop parts.

    As for ICL/TGL AVX-512 - I guess no one uses laptops for their AVX-512 prowess. Anyways, the conversation about AVX-512 almost looks stupid since we don't even have benchmarks for it and the number of common tasks benefitting from this instruction set seems to be zero despite it having been available for more than three years.

    Leave a comment:


  • avem
    replied
    Originally posted by smitty3268 View Post
    This is frankly hilarious, coming from you.


    Not sure if this was directed at me or just generally, but I'm not an AMD fanboy.

    I'll flat out say I think Alder Lake is better than Zen 3 overall. Does that satisfy you?

    However, I think many of the things you are arguing in this thread are flat out crazy, and make it difficult to take anything you say seriously. Why not focus on what Alder Lake is actually good at, rather than trying to argue things like that it's more efficient? What's the need to constantly prove it's 100% better in every possible way?

    It's better at some things, worse at others. That's good for competition and for consumer choice.
    I'll gladly discuss the pros and cons of ADL, I just don't understand why AMD has a cult following seemingly everywhere and that makes arguing quite difficult.

    Pros:
    • A massive IPC increase
    • Very competitive performance despite having fewer threads than comparable AMD parts
    • An option of using AVX-512 at the expense of E-cores
    • You can use DDR4 with it without losing performance in most applications
    • PCI-E 5.0
    Cons:
    • Overclocked out of the box and an insane power consumption at full MT load (can be partially mitigated by setting PL1/PL2) limits
    • Actual TDP is much higher than advertised (almost twice as high for 12900K, 125W vs 241W)
    • AVX-512 is again delayed by a generation or two because E-cores don't support it
    • High-speed DDR5 costs an arm and leg
    • Requires OS support for proper functioning, currently works terribly under Linux
    Other thoughts:
    • An anemic iGPU and most likely Intel will not release desktop parts with more EUs. Would be great if Intel designed and released desktop parts with more EUs at the expense of P-cores
    • The maximum power efficiency for 12900K is between 160-180W which probably Intel should have targeted instead of 241W
    This looks much better than saying "ADL is bad", "Intel sucks" or "Insane power consumption!!"
    Last edited by avem; 12 November 2021, 07:45 AM.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by coder View Post
    What I don't get is why you seem so threatened by these discussions? Why are you trying to shut them down? Do you have some vested interest that you've not disclosed? Otherwise, you are as entitled to your opinions (and as capable of backing them up with data) as anyone else.
    Yeah, exactly. I don't understand why birdie is so emotionally invested in the topic. He does the same thing in every AMD thread too.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by avem View Post

    They have not formally introduced AVX-512 in their consumer lineup yet aside from a few mobile parts.
    I realize most people want to forget Rocket Lake ever existed.... but it did.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by avem View Post
    Not sure what you're talking about. I'd be glad if we discussed products but instead half of the comments about ADL are "But but but Intel is bad!" WTF is wrong with people?
    This is frankly hilarious, coming from you.

    Can you stick to it instead of discussing a person who you personally dislike because he doesn't share your infinite AMD love and devotion?
    Not sure if this was directed at me or just generally, but I'm not an AMD fanboy.

    I'll flat out say I think Alder Lake is better than Zen 3 overall. Does that satisfy you?

    However, I think many of the things you are arguing in this thread are flat out crazy, and make it difficult to take anything you say seriously. Why not focus on what Alder Lake is actually good at, rather than trying to argue things like that it's more efficient? What's the need to constantly prove it's 100% better in every possible way?

    It's better at some things, worse at others. That's good for competition and for consumer choice.
    Last edited by smitty3268; 11 November 2021, 11:45 PM.

    Leave a comment:

Working...
X