Alder Lake and AVX-512

By: anon (anon.delete@this.anontech.anon), July 11, 2020 8:12 pm
Room: Moderated Discussions
Geoff Langdale (geoff.langdale.delete@this.gmail.com) on July 11, 2020 7:49 pm wrote:
> OK, so I'm biased too - I used to work for Intel (not in hardware design, so I certainly don't have a vested
> interest in the particulars of AVX-512). And while I'm not an mouth-frothing Intel loyalist, I did invest a lot
> of time and trouble with SIMD programming and I think I've had a good deal of success making workloads that aren't
> from "Floating Point Benchmark Central Casting" go a lot faster with SIMD - specifically regular expression scanning
> (https://github.com/intel/hyperscan) and JSON parsing (https://github.com/lemire/simdjson).
>
> Granted there's a lot of general bloody-mindedness in both those projects where you wind up going faster because
> you beat on a problem a lot, not necessarily because SIMD Is Magic. But we have some SIMD-based table lookup
> stuff that's way faster than the integer equivalent both because you're doing a lot of stuff at once, but you're
> also doing stuff where there's no integer equivalent (there's no PSHUFB for a GPR register).
>
> Also, AVX-512 has a bunch of weirdnesses that I can't say that I love - the mask registers are a PITA
> and weirdly opinionated - and doing ops on them are so expensive that they displace useful AVX-512
> ops (they use the same units!). So it's weird to save register pressure and replace it with uop pressure;
> I wish they'd stayed with the old-fashioned 0xff or 0x00 compare results at 512-bit.
>
> That being said:
>
> The top level of AVX* has always, IIRC, been a power virus for the first iteration or two. Not sure that's a
> long-term argument against progressing to bigger vector registers. A lot of the area stuff with AVX-512 is due
> to a rather exuberant simultaneous increase in number of vector registers, vector size and FP capacity. It is
> probably a one-time hit - I doubt they'll double the number of registers or their size again any time soon.
>
> AVX-512 done with integers and on smaller registers (128/256) doesn't have the pain of doing
> AVX-512 in the worst case. There's a lot of interesting tricks that can be done with AVX-512
> that allow writing of viable branch-free code for 'irregular' tasks in the integer domain. If
> you have VBMI/VBMI2 you can do all sorts of funky table lookup tricks for things like string
> processing and search that allow you to drag more integer workloads into the SIMD domain.
>
> Scatter/gather may eventually become viable (they have traditionally been disappointing), which
> will also allow some nice tricks. I'm not totally enamored of scatter/gather due to the way
> they tend to slow a vectorized algorithm down to the slowest memory access (unlike GPR stuff,
> where you can slip a lot of useful work past the 'bad iteration'). But if you otherwise had
> a good SIMD algorithm barring that *one* need for gather, it's getting almost useful.
>
> I'm also not clear that the long-term trend in architecture is towards cramming as many "generally useful cores"
> into as small a space as possible. I'm not a chip designer, and don't even play one on TV, but the arguments
> that we're likely to do better with "dark silicon" because of cooling factors doesn't seem ridiculous. So the
> FP weenie gets their big multiply-add unit and we clock down when they use it, but otherwise, we get a bit of
> headroom that they didn't have. IDK that we could otherwise have crammed more cores into that space and flogged
> them at 100%. AMD is on a different process so we're not really seeing a clear counterexample there.

Eh, the actual power density, which is sort of the real issue here, is so much higher in an 512-bit datapath (or even a 256-bit one) compared to everything else in core, including the entire regular integer side of things. I don't think the argument holds that you'll just deal with it by having this giant thing that can suddenly suck down huge amounts of power into tiny areas, but it'll mostly be not on so it's now somehow a win. Get rid of it, cram in some more cores, and your power density across the board still isn't terribly high. Plus, the need to deliver power those huge datapaths will almost certainly have a negative effect on power delivery to the rest of the core. At the least, it'd require upsizing power switches and voltage regulators and such, that wouldn't need to be that large otherwise.

>
> I do agree that the fragmentation that Intel has introduced is absolutely tooth grinding. All those
> clever (?) tricks I've worked on are a really hard sell if you have an alphabet soup of different SIMD
> specifications and a slow uptake of their "latest and greatest". Hard to motivate some weird branch
> of the code that only works on 1% of processors out there (and might well cause clockdowns!).
>
> Some of this fragmentation was clearly accidental - if certain things hadn't happened,
> we could be living in a world with a successful 10nm (no pluses) process and the world
> would be full of CNL/ICL, not endless variants of client Skylake. But the way they keep
> shipping "Pentiums" and so on with no AVX2 (even) or BMI/BMI2 is a bloody nuisance.
>
> But I still think you're wrong here overall. There are workloads outside of "operating systems"
> and lots of gccs. Floating point is not fungible between GPU and CPU and there are plenty of
> FP algorithms that mix branching and computation in a way that doesn't play nice on GPU.
>

Sure, there are domains where it may be more appropriate. But why should the other 99% of the domains have that tax? This ends up being a situation where these things exist because Intel has historically just made a single one-size-fits-all core, not because it made sense for the majority of use cases. We've seen them start to (slightly) diversify with SKL client vs server, but seemingly they've gone back on that a bit by adding the AVX512 unit to ICL anyways.

In a perfect world, the unique capabilities of the AVX512 ISA wouldn't be tied to the vector width.

> If there's a sweet spot, it's almost certainly 256-bit SIMD, not 128-bit. AVX2 is very solid
> and widely available and the "power virus" argument doesn't really hold, particularly if
> you're integer-oriented. I would claim that a 256-bit AVX-512 (not as stupid as it sounds
> - I mean the AVX-512 operations done on 128/256 bit registers, which allow you 3 uops/cycle
> execution) with all the goodies (VBMI, GFNI, etc) might be an even sweeter spot.
>
> Geoff.
>
> Linus Torvalds (torvalds.delete@this.linux-foundation.org) on July 11, 2020 11:41 am wrote:
> > me (me.delete@this.me.com) on July 11, 2020 7:02 am wrote:
> > > https://www.phoronix.com/scan.php?page=news_item&px=GCC11-SapphireRapids-AlderLake
> > >
> > > Hope you didn't get too attached to AVX-512. The GCC 11 compiler target for Alder
> > > Lake doesn't enable it, only AVX2. As Phoronix mentions the target likely is what
> > > the small cores support and not necessarily the big... but it makes me wonder...
> >
> > 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.
> >
> > I hope 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.
> >
> > I've said this before, and I'll say it again: in the heyday of x86, when Intel was laughing all the
> > way to the bank and killing all their competition, absolutely everybody else did better than Intel
> > on FP loads. Intel's FP performance sucked (relatively speaking), and it matter not one iota.
> >
> > Because absolutely nobody cares outside of benchmarks.
> >
> > The same is largely true of AVX512 now - and in the future. Yes, you can find
> > things that care. No, those things don't sell machines in the big picture.
> >
> > And AVX512 has real downsides. 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 destest FP benchmarks, and I realize other people care deeply.
> > I just think AVX512 is exactly the wrong thing to do. It's a pet peeve of mine. It's a prime example
> > of something Intel has done wrong, partly by just increasing the fragmentation of the market.
> >
> > Stop with the special-case garbage, and make all the core common stuff that everybody
> > cares about run as well as you humanly can. Then do a FPU that is barely good enough
> > on the side, and people will be happy. AVX2 is much more than enough.
> >
> > Yeah, I'm grumpy.
> >
> > Linus
>
>

< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Alder Lake and AVX-512me2020/07/11 07:02 AM
  Alder Lake and AVX-512Linus Torvalds2020/07/11 11:41 AM
    informative (NT)blue2020/07/11 12:40 PM
    grumpyMichael S2020/07/11 12:51 PM
      grumpyme2020/07/11 01:27 PM
    area and power cost of AVX-512Michael S2020/07/11 12:58 PM
      area and power cost of AVX-512Anon2020/07/11 04:35 PM
        area and power cost of AVX-512Michael S2020/07/12 04:16 AM
          area and power cost of AVX-512Travis Downs2020/07/12 09:13 AM
      area and power cost of AVX-512Travis Downs2020/07/11 07:19 PM
    Alder Lake and AVX-512Maynard Handley2020/07/11 02:02 PM
      Alder Lake and AVX-512Ungo2020/07/11 05:28 PM
        Alder Lake and AVX-512Maynard Handley2020/07/11 10:16 PM
      Alder Lake and AVX-512Linus Torvalds2020/07/11 06:51 PM
        Alder Lake and AVX-5122020/07/12 01:48 PM
          Alder Lake and AVX-512Michael S2020/07/12 03:07 PM
          HDRAnon32020/07/12 03:42 PM
            HDR10 in Kaby Lake?David Kanter2020/07/12 05:09 PM
              HDR10 in Kaby Lake?Maynard Handley2020/07/12 06:13 PM
                Thanks for the link (NT)David Kanter2020/07/12 06:43 PM
              HDR10 in Kaby Lake?Anon32020/07/13 01:36 AM
        Alder Lake and AVX-512Dummond D. Slow2020/07/12 03:00 PM
        AVX-512 with narrow ex units?m2020/07/23 12:10 PM
          AVX-512 with narrow ex units?Anon2020/07/23 12:53 PM
            AVX-512 with narrow ex units?Paul A. Clayton2020/07/23 06:32 PM
              AVX-512 with narrow ex units?Anon2020/07/23 06:50 PM
                AVX-512 with narrow ex units?Paul A. Clayton2020/07/23 07:45 PM
                  AVX-512 with narrow ex units?Anon2020/07/23 08:15 PM
                    AVX-512 with narrow ex units?Jukka Larja2020/07/24 04:44 AM
                      AVX-512 with narrow ex units?Gabriele Svelto2020/07/24 02:56 PM
                        AVX-512 with narrow ex units?Jouni Osmala2020/07/24 09:22 PM
                          AVX-512 with narrow ex units?Jukka Larja2020/07/25 01:32 AM
                      AVX-512 with narrow ex units?Eugene Nalimov2020/07/25 05:56 PM
                        AVX-512 with narrow ex units?Jukka Larja2020/07/26 01:28 AM
                        AVX-512 with narrow ex units?Gabriele Svelto2020/07/26 02:22 PM
                          AVX-512 with narrow ex units?Jukka Larja2020/07/27 07:00 AM
          AVX-512 with narrow ex units?-.-2020/07/23 06:32 PM
            AVX-512 with narrow ex units?Travis Downs2020/07/24 05:01 PM
    Alder Lake and AVX-512Jörn Engel2020/07/11 04:45 PM
      Alder Lake and AVX-512Chester2020/07/11 05:26 PM
        Alder Lake and AVX-512Jörn Engel2020/07/11 06:22 PM
        Alder Lake and AVX-512Michael S2020/07/12 02:02 AM
        Alder Lake and AVX-512Travis Downs2020/07/13 09:01 PM
      Alder Lake and AVX-512Linus Torvalds2020/07/11 06:54 PM
        Alder Lake and AVX-512Jörn Engel2020/07/11 08:01 PM
          Alder Lake and AVX-512N Owen2020/07/12 12:37 AM
            Alder Lake and AVX-512Michael S2020/07/12 01:48 AM
            Alder Lake and AVX-512anon22020/07/12 07:13 PM
          Alder Lake and AVX-512Travis Downs2020/07/13 09:09 PM
            Alder Lake and AVX-512Jörn Engel2020/07/13 11:42 PM
      Alder Lake and AVX-512Doug S2020/07/11 11:49 PM
        Alder Lake and AVX-512Michael S2020/07/12 01:53 AM
    Alder Lake and AVX-512Travis Downs2020/07/11 07:03 PM
      Alder Lake and AVX-512Veedrac2020/07/11 07:43 PM
        Alder Lake and AVX-512anon22020/07/12 01:31 AM
          Alder Lake and AVX-512Veedrac2020/07/12 04:01 AM
            Alder Lake and AVX-512anon22020/07/12 03:26 PM
              Alder Lake and AVX-512Anon32020/07/12 04:07 PM
                Alder Lake and AVX-512anon22020/07/12 05:39 PM
              Alder Lake and AVX-512Veedrac2020/07/12 04:21 PM
                Alder Lake and AVX-512anon22020/07/12 05:33 PM
                  Alder Lake and AVX-512Veedrac2020/07/12 05:54 PM
                    Alder Lake and AVX-512anon22020/07/12 06:20 PM
                  Alder Lake and AVX-512David Hess2020/07/12 07:32 PM
                    Alder Lake and AVX-512anon22020/07/12 08:41 PM
                Alder Lake and AVX-5122020/07/13 04:02 AM
                  Alder Lake and AVX-512anon22020/07/13 07:25 PM
                    PentiumMMX vs Transmeta's VLIW in hindsight2020/07/19 06:16 AM
                      PentiumMMX vs Transmeta's VLIW in hindsightMaynard Handley2020/07/19 10:47 AM
                      PentiumMMX vs Transmeta's VLIW in hindsightanon22020/07/19 03:24 PM
                      VLIW, OOO, Pairing, and FusionChester2020/07/19 10:16 PM
                        Poulson was in-order (NT)anon22020/07/20 12:20 AM
                        VLIW, OOO, Pairing, and FusionMichael S2020/07/20 12:48 AM
                        Itanium is NOT VLIWHeikki Kultala2020/07/20 02:27 PM
                          Itanium is NOT VLIWAdrian2020/07/20 11:03 PM
                            Itanium crappiness and EPIC - and could EPIC still have something good in it?Heikki Kultala2020/07/21 03:38 AM
                              Itanium crappiness and EPIC - and could EPIC still have something good in it?anon22020/07/21 05:03 AM
                                Itanium crappiness and EPIC - and could EPIC still have something good in it?dmcq2020/07/21 03:27 PM
                              Itanium crappiness and EPIC - and could EPIC still have something good in it?j2020/07/21 08:54 AM
                                Itanium crappiness and EPIC - and could EPIC still have something good in it?Tim McCaffrey2020/07/21 10:30 AM
                              Itanium crappiness and EPIC - and could EPIC still have something good in it?Linus Torvalds2020/07/21 09:13 AM
                                Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureHeikki Kultala2020/07/22 12:31 PM
                                  Turn that on its head?Ray2020/07/22 12:49 PM
                                    Turn that on its head?Anon2020/07/22 01:53 PM
                                    Turn that on its head?Maynard Handley2020/07/22 02:37 PM
                                    Turn that on its head?anon22020/07/22 03:32 PM
                                    Turn that on its head?anon32020/07/22 04:45 PM
                                    Turn that on its head?Heikki Kultala2020/07/23 02:53 AM
                                      Turn that on its head?Anon2020/07/23 10:20 AM
                                        Turn that on its head?Heikki Kultala2020/07/23 11:21 AM
                                          Turn that on its head?Brett2020/07/23 03:26 PM
                                            Turn that on its head?Brett2020/07/24 04:22 AM
                                      Bundling OOO entries does this implicitlyDavid Kanter2020/07/23 10:56 AM
                                      Turn that on its head?anon2020/07/23 11:49 AM
                                  Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureMaynard Handley2020/07/22 02:29 PM
                                  Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architecturewumpus2020/07/22 03:16 PM
                                    Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureDoug S2020/07/22 10:37 PM
                                      what Intel would have doneMichael S2020/07/23 12:46 AM
                                        what Intel would have doneDoug S2020/07/23 09:52 AM
                                        what Intel would have doneAnon2020/07/23 10:25 AM
                                          what Intel would have doneMichael S2020/07/23 11:23 AM
                                            what Intel would have doneMontaray Jack2020/07/23 06:08 PM
                                    Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureHeikki Kultala2020/07/22 11:47 PM
                                      Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architecturewumpus2020/07/23 01:46 PM
                                  Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureMichael S2020/07/23 12:56 AM
                                    Itanium is not synomym of EPIC. Itanium is just the most common EPIC-style architectureHeikki Kultala2020/07/23 02:44 AM
                          thanksChester2020/07/24 03:50 PM
      Alder Lake and AVX-512Linus Torvalds2020/07/11 07:46 PM
        Alder Lake and AVX-512never_released2020/07/11 08:54 PM
          Alder Lake and AVX-512Michael S2020/07/12 02:25 AM
        Alder Lake and AVX-512anon22020/07/12 01:36 AM
      Alder Lake and AVX-512Doug S2020/07/12 12:01 AM
      Alder Lake and AVX-512Michael S2020/07/12 02:41 AM
        Alder Lake and AVX-512rwessel2020/07/12 10:17 AM
      Alder Lake and AVX-512-.-2020/08/18 03:24 AM
        Alder Lake and AVX-512Travis Downs2020/08/18 11:04 PM
    Alder Lake and AVX-512Geoff Langdale2020/07/11 07:49 PM
      Alder Lake and AVX-512anon2020/07/11 08:12 PM
      Alder Lake and AVX-512Jörn Engel2020/07/11 08:33 PM
        Alder Lake and AVX-512Michael S2020/07/12 03:00 AM
        Alder Lake and AVX-512Jukka Larja2020/07/12 08:51 AM
          Alder Lake and AVX-512Maynard Handley2020/07/12 10:30 AM
            Alder Lake and AVX-512Jukka Larja2020/07/13 07:43 AM
              Alder Lake and AVX-512Montaray Jack2020/07/23 07:20 PM
                Alder Lake and AVX-512Jukka Larja2020/07/24 04:57 AM
          Alder Lake and AVX-512Jörn Engel2020/07/12 11:35 AM
            Alder Lake and AVX-512Linus Torvalds2020/07/12 12:01 PM
              Alder Lake and AVX-512Linus Torvalds2020/07/12 12:15 PM
                Alder Lake and AVX-512anonymou52020/07/12 01:50 PM
                  Alder Lake and AVX-512Linus Torvalds2020/07/12 02:31 PM
                    Alder Lake and AVX-512anonymou52020/07/12 03:09 PM
                      Alder Lake and AVX-512Linus Torvalds2020/07/12 04:25 PM
                        Alder Lake and AVX-512anonymou52020/07/12 08:34 PM
                          Alder Lake and AVX-512Jose2020/07/13 01:35 AM
                  Alder Lake and AVX-512gallier22020/07/13 02:11 AM
                Alder Lake and AVX-512gallier22020/07/13 02:01 AM
                  Alder Lake and AVX-512Linus Torvalds2020/07/13 11:06 AM
                    Alder Lake and AVX-512Doug S2020/07/13 12:11 PM
                      Alder Lake and AVX-512Brett2020/07/14 02:34 AM
                        Alder Lake and AVX-512Linus Torvalds2020/07/14 09:02 AM
                          Alder Lake and AVX-512Maynard Handley2020/07/14 12:40 PM
                            Alder Lake and AVX-512Michael S2020/07/14 12:48 PM
                            Alder Lake and AVX-512Linus Torvalds2020/07/15 01:37 AM
                              OS X file names normalizationMichael S2020/07/15 02:26 AM
                                OS X file names normalizationSimon Farnsworth2020/07/15 04:16 AM
                                  OS X file names normalizationMichael S2020/07/15 10:51 AM
                                    OS X file names normalizationSimon Farnsworth2020/07/15 12:27 PM
                                OS X file names normalizationDoug S2020/07/15 10:46 AM
                                  OS X file names normalizationMichael S2020/07/15 11:05 AM
                                    OS X file names normalizationLinus Torvalds2020/07/15 12:58 PM
                                      OS X file names normalizationLinus Torvalds2020/07/15 02:21 PM
                                      OS X file names normalizationgallier22020/07/15 11:57 PM
                                    OS X file names normalizationgallier22020/07/15 11:44 PM
                                  OS X file names normalizationRob Thorpe2020/07/15 11:23 AM
                                    OS X file names normalizationDoug S2020/07/15 01:32 PM
                                      OS X file names normalizationMaynard Handley2020/07/15 05:20 PM
                                        OS X file names normalizationLinus Torvalds2020/07/15 08:37 PM
                                          OS X file names normalizationAnon32020/07/16 01:43 PM
                                            OS X file names normalizationDoug S2020/07/16 03:38 PM
                                              OS X file names normalizationLinus Torvalds2020/07/17 12:21 AM
                                                OS X file names normalizationAnon32020/07/17 02:15 AM
                                                  OS X file names normalizationJukka Larja2020/07/17 06:40 AM
                                                OS X file names normalizationgallier22020/07/17 03:19 AM
                                                  OS X file names normalizationLinus Torvalds2020/07/17 09:41 AM
                                                    OS X file names normalizationDummond D. Slow2020/07/17 09:54 AM
                                                      OS X file names normalizationLinus Torvalds2020/07/17 10:16 AM
                                                      OS X file names normalizationSimon Farnsworth2020/07/18 06:12 AM
                                              OS X file names normalizationAnon32020/07/17 02:04 AM
                                                OS X file names normalizationDoug S2020/07/17 10:15 AM
                              Alder Lake and AVX-512Maynard Handley2020/07/15 10:32 AM
                            File Systems and VC ProblemsRob Thorpe2020/07/15 07:24 AM
                    vectorization of utf8Robert David Graham2020/07/13 02:36 PM
                      vectorization of utf8anon22020/07/13 05:07 PM
                        vectorization of utf8Robert David Graham2020/07/13 08:36 PM
                          vectorization of utf8anon22020/07/13 11:23 PM
                        vectorization of utf8Maynard Handley2020/07/13 10:46 PM
                      vectorization of utf8Gabriele Svelto2020/07/15 03:27 AM
                    Alder Lake and AVX-512gallier22020/07/14 01:13 AM
              Alder Lake and AVX-512Jörn Engel2020/07/12 01:29 PM
                Alder Lake and AVX-512Linus Torvalds2020/07/12 02:08 PM
                  Alder Lake and AVX-512Jörn Engel2020/07/12 06:26 PM
                    Alder Lake and AVX-512-.-2020/07/12 07:11 PM
                      Alder Lake and AVX-512Jörn Engel2020/07/12 07:43 PM
            Alder Lake and AVX-512Jukka Larja2020/07/13 08:38 AM
              Alder Lake and AVX-512Jörn Engel2020/07/13 10:10 AM
                Alder Lake and AVX-512Michael S2020/07/13 11:02 AM
                  Alder Lake and AVX-512Jörn Engel2020/07/13 11:22 AM
                    Alder Lake and AVX-512Michael S2020/07/13 12:10 PM
                      Alder Lake and AVX-512Jörn Engel2020/07/13 04:03 PM
                Alder Lake and AVX-512Jukka Larja2020/07/14 06:53 AM
      Alder Lake and AVX-512Linus Torvalds2020/07/11 08:34 PM
        Alder Lake and AVX-512Brett2020/07/11 09:02 PM
          Alder Lake and AVX-512David Hess2020/07/13 12:36 PM
            Alder Lake and AVX-512anonymou52020/07/13 01:01 PM
              Alder Lake and AVX-512Brett2020/07/13 04:19 PM
        Alder Lake and AVX-512Geert2020/07/11 09:36 PM
          AMD's FPUChester2020/07/12 02:28 AM
            Is 3|5 lower than 4?Michael S2020/07/12 03:59 AM
              Is 3|5 lower than 4?Chester2020/07/12 05:54 AM
        Alder Lake and AVX-512Geoff Langdale2020/07/11 11:45 PM
          Alder Lake and AVX-512me2020/07/12 03:44 AM
          Alder Lake and AVX-512Michael S2020/07/12 04:09 AM
          Alder Lake and AVX-512Linus Torvalds2020/07/12 11:35 AM
            ~80% of details are wrong. So what one can expect from conclusions? :( (NT)Michael S2020/07/12 11:57 AM
              ~80% of details are wrong. So what one can expect from conclusions? :(anonymous22020/07/12 12:50 PM
            Alder Lake and AVX-512nobody in particular2020/07/12 12:25 PM
              Alder Lake and AVX-512Linus Torvalds2020/07/12 12:37 PM
                Alder Lake and AVX-512nobody in particular2020/07/12 12:43 PM
                  Alder Lake and AVX-512me2020/07/12 01:32 PM
                    Alder Lake and AVX-512Maynard Handley2020/07/12 08:51 PM
            Alder Lake and AVX-512UnmaskedUnderflow2020/07/12 12:33 PM
            AVX-512 vs SVE2-.-2020/07/12 06:22 PM
              AVX-512 vs SVE2noko2020/07/13 12:12 AM
                AVX-512 vs SVE2-.-2020/07/13 04:00 AM
            Alder Lake and AVX-512Geoff Langdale2020/07/12 08:18 PM
              Could you please stop top-posting (NT)Jukka Larja2020/07/13 08:45 AM
              Alder Lake and AVX-512Romain Dolbeau2020/07/15 01:00 AM
            Alder Lake and AVX-512Spiteful Sprites2020/07/13 04:59 AM
              Alder Lake and AVX-512nobody in particular2020/07/13 09:12 AM
                Alder Lake and AVX-512Spiteful Sprites2020/07/13 04:21 PM
                  Alder Lake and AVX-512Jouni Osmala2020/07/14 02:55 AM
                  RISC-V & commercial support (was: Alder Lake and AVX-512)Romain Dolbeau2020/07/15 01:11 AM
                    RISC-V & commercial support (was: Alder Lake and AVX-512)Romain Dolbeau2020/07/15 01:13 AM
              Alder Lake and AVX-512Linus Torvalds2020/07/13 11:10 AM
            AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Romain Dolbeau2020/07/14 10:09 AM
              AVX-512/SVE & HPC (was: Alder Lake and AVX-512)anon2020/07/14 10:53 AM
                AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Romain Dolbeau2020/07/14 11:27 AM
                  AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Maynard Handley2020/07/14 12:52 PM
                  AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Doug S2020/07/14 01:43 PM
                  AVX-512/SVE & HPC (was: Alder Lake and AVX-512)anon2020/07/14 03:01 PM
              AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Linus Torvalds2020/07/14 12:00 PM
                AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Romain Dolbeau2020/07/14 11:42 PM
                  Configurable cache line size?Doug S2020/07/15 10:56 AM
                    Configurable cache line size?dmcq2020/07/15 03:43 PM
                    Configurable cache line size?Romain Dolbeau2020/07/15 11:37 PM
                    Configurable cache line size?NoSpammer2020/07/16 01:27 AM
                    Configurable cache line size?Pixie2020/07/16 10:55 AM
                      Configurable cache line size?Etienne2020/07/17 01:03 AM
                        Configurable cache line size?Hugo Décharnes2020/07/18 02:11 AM
                  Cache line sizeMark Roulo2020/07/15 06:10 PM
                    Cache line sizeanon2020/07/15 06:46 PM
                  AVX-512/SVE & HPC (was: Alder Lake and AVX-512)Gabriele Svelto2020/07/17 02:30 AM
                    AVX-512/SVE & HPC (was: Alder Lake and AVX-512)dmcq2020/07/17 03:34 AM
                      AVX-512/SVE & HPC (was: Alder Lake and AVX-512)zArchJon2020/07/17 01:16 PM
            Macro-instructions to the rescue2020/07/24 12:56 PM
              Some fundamentals haven't changedChester2020/07/24 03:59 PM
                Some fundamentals haven't changed2020/07/24 04:24 PM
                  Some fundamentals haven't changeddmcq2020/07/25 07:58 AM
                    Some fundamentals haven't changed2020/07/25 11:05 AM
                    Some fundamentals haven't changedBrett2020/07/25 02:16 PM
                      Some fundamentals haven't changedBrett2020/07/25 02:27 PM
                      What belt is.Heikki Kultala2020/07/26 07:49 AM
                        What belt is.Michael S2020/07/26 10:00 AM
                          What belt is.Brett2020/07/26 11:46 PM
                            What belt is.Michael S2020/07/27 12:52 AM
                              What belt is.Brett2020/07/27 07:25 AM
                                What belt is.Doug S2020/07/27 01:31 PM
                                  What belt is.Andrew Clough2020/07/28 06:11 AM
                                    What belt is.dmcq2020/07/28 08:17 AM
                                      Mill Compiler still MIA?Geoff Langdale2020/07/28 05:04 PM
                                        If they release the compiler, how they will blame the still-in-development compiler for the lacklust (NT)Anon2020/07/28 05:20 PM
                                          If they release the compiler, how they will blame the still-in-development compiler for the lacklustAnon2020/07/28 05:20 PM
                                        Apparently they're busy writing a kernel...Anon2020/07/29 03:03 AM
                                          Apparently they're busy writing a kernel...dmcq2020/07/29 03:39 AM
                        What belt is.2020/07/26 11:44 AM
                          What belt is.anonymous22020/07/26 12:02 PM
                            What belt is.Doug S2020/07/26 03:26 PM
                              What belt is.2020/07/26 04:02 PM
        gooduseruser2020/07/12 10:06 AM
      Alder Lake and AVX-512-.-2020/07/11 09:03 PM
        Alder Lake and AVX-512-.-2020/07/11 09:07 PM
      Alder Lake and AVX-512j2020/07/13 12:29 AM
        Alder Lake and AVX-512Michael S2020/07/13 01:12 AM
          Alder Lake and AVX-512j2020/07/13 02:58 AM
            Alder Lake and AVX-512dmcq2020/07/13 04:53 PM
              Alder Lake and AVX-512Michael S2020/07/14 12:57 AM
                Alder Lake and AVX-512Maynard Handley2020/07/14 10:26 AM
                Alder Lake and AVX-512dmcq2020/07/14 12:33 PM
                  Alder Lake and AVX-512dmcq2020/07/14 03:43 PM
                    Alder Lake and AVX-512Michael S2020/07/15 12:55 AM
                      Alder Lake and AVX-512dmcq2020/07/15 02:19 AM
                        Alder Lake and AVX-512Michael S2020/07/15 02:34 AM
                          Alder Lake and AVX-512dmcq2020/07/15 03:03 AM
                            Alder Lake and AVX-512Michael S2020/07/15 09:43 AM
                              Alder Lake and AVX-512dmcq2020/07/15 09:54 AM
                                Alder Lake and AVX-512Michael S2020/07/15 11:35 AM
                                  Alder Lake and AVX-512dmcq2020/07/15 03:18 PM
                                    GV100 + POWER9Michael S2020/07/16 01:17 AM
                                      GV100 + POWER9dmcq2020/07/16 08:58 AM
                                        GV100 + POWER9dmcq2020/07/16 09:10 AM
                        Alder Lake and AVX-512dmcq2020/07/15 02:48 AM
    Alder Lake and AVX-512o2020/07/12 03:08 AM
    Alder Lake and AVX-5122020/07/12 11:07 AM
      Alder Lake and AVX-5122020/07/12 11:32 AM
      Alder Lake and AVX-512Linus Torvalds2020/07/12 11:39 AM
        Alder Lake and AVX-5122020/07/12 12:47 PM
        Alder Lake and AVX-512Michael S2020/07/12 01:18 PM
          x87 crapHeikki Kultala2020/07/12 01:30 PM
            x87 crapMichael S2020/07/12 01:37 PM
              x87 crapHeikki kultala2020/07/12 02:11 PM
                x87 crapMichael S2020/07/12 02:50 PM
                  Sparc and PA-RISC vs pentium FP performanceHeikki Kultala2020/07/13 01:14 AM
                    Sparc and PA-RISC vs pentium FP performanceanonymous22020/07/13 10:48 AM
          Alder Lake and AVX-512Doug S2020/07/12 03:33 PM
            Alder Lake and AVX-512Michael S2020/07/12 04:10 PM
    Alder Lake and AVX-512David Kanter2020/07/12 05:01 PM
      Alder Lake and AVX-512anon2020/07/12 05:40 PM
      ~0% of users do much FP outside of GPUs for games (NT)anonymous22020/07/12 05:47 PM
        ~0% of users do much FP outside of GPUs for gamesMaynard Handley2020/07/13 12:26 AM
        not trueChester2020/07/13 12:37 AM
          not trueMichael S2020/07/13 01:29 AM
            not trueChester2020/07/13 01:59 AM
              not trueanonymous22020/07/13 10:32 AM
                not trueMaynard Handley2020/07/13 02:30 PM
                  not trueChester2020/07/14 05:47 AM
            not trueDoug S2020/07/13 12:30 PM
              not trueAnon2020/07/13 01:16 PM
                not trueMaynard Handley2020/07/13 02:39 PM
              not trueMaynard Handley2020/07/13 02:38 PM
          not trueLinus Torvalds2020/07/13 11:27 AM
            not trueDummond D. Slow2020/07/13 02:10 PM
              not trueMaynard Handley2020/07/13 02:49 PM
                not trueDummond D. Slow2020/07/13 03:38 PM
            not true (about FP, not avx-512)Chester2020/07/17 10:37 AM
  Alder Lake and AVX-512Travis Downs2020/07/11 06:45 PM
    Alder Lake and AVX-512-.-2020/07/11 06:57 PM
      Alder Lake and AVX-512-.-2020/07/12 04:26 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?