SMT flock of chickens stupidity.

By: wumpus (lost.delete@this.in.a.cave), August 6, 2018 9:54 am
Room: Moderated Discussions
Brett (ggtgp.delete@this.yahoo.com) on August 5, 2018 10:27 pm wrote:
> Brett (ggtgp.delete@this.yahoo.com) on July 28, 2018 12:21 am wrote:
> > Brett (ggtgp.delete@this.yahoo.com) on July 26, 2018 1:03 pm wrote:
> > > Maynard Handley (name99.delete@this.name99.org) on July 26, 2018 12:22 pm wrote:
> > > > Passing Through (ireland.delete@this.web.ie) on July 26, 2018 10:11 am wrote:
> > > > > Per Hesselgren (perhesselgren.delete@this.yahoo.se) on July 26, 2018 8:50 am wrote:
> > > > > > Emil Briggs (me.delete@this.nowherespam.com) on July 26, 2018 4:50 am wrote:
> > > > > > >
> > > > > > > Performance boost from SMT depends on the workload. In many cases it doesn't help and can even hurt.
> > > > > > > This is particularly noticeable for some HPC workloads. That's not really relevant for Apple of course
> > > > > > > and while SMT is probably quite helpful for low core count parts on the desktop it's probably a lot
> > > > > > > less useful for the high core count parts which is typically where Intel has made it available.
> > > > > >
> > > > > > 7-zip benchmark is one example where decompress is more depending on threads than compress.
> > > > >
> > > > > Again,
> > > > >
> > > > > What I'd do is reiterate my original hypothesis to Maynard. Things change, in the software environment,
> > > > > that have knock-on impacts in hardware. I'll provide one example, to illustrate it below.
> > > > >
> > > > > I've been tracking the trajectory of where multi-core computing requirement is going in my own
> > > > > work, for several years now. For my own part, I didn't get anything extra for the kind of computing
> > > > > tasks I do, above and beyond a decent dual core system. Yeah, 7-zip or with anything like that,
> > > > > it is possible to almost 'break' a dual core system. The tasks is very processor bound.
> > > > >
> > > > > My explanation of this, would be based upon an analogy. The way that in phones now, they need to have
> > > > > different sets of processors, for while the device is in 'high activity' and the user interacts a lot
> > > > > with app's on the phone. Versus times, when the phone is on sleep almost, and just receiving new messages
> > > > > and stuff, where you don't want the same levels of power draw down. As I've looked at my work in the
> > > > > area of computer aided design in the last several years, the need for real multi-core processors with
> > > > > ability to power down on certain cores, has increased. Why? Because the workloads that they have to
> > > > > endure now, are not a million miles away from the one described on the smart phone chips.
> > > > >
> > > > > What happened about ten years ago, was that our computer aided design files increasingly started
> > > > > to become 'database' files, of 10x the size, instead of being file based, just collections of binary
> > > > > points of geometry. I.e. They had meta data, and a lot more meta data in them. Instead of files
> > > > > being 3mb in size, they suddenly exploded up to 300mb very easily. However, that was fine for the
> > > > > most part, because one still worked on that 300mb database locally. Yes, when one 'opened' the file
> > > > > at first, it could take up to five or ten minutes to assemble all of the components and relationships
> > > > > between them, together in the computer's memory, from the file on the hard drive.
> > > > >
> > > > > And yes, additional cores were a huge time saving expense for that initial opening of one's project files
> > > > > each morning, or after lunch, or whatever. In between though, the tasks required did not scale much beyond
> > > > > 'single' or 'dual' cores at best. Hence, why one had to be careful, to try and shop around for smaller
> > > > > numbers of cores, and have 'more of' local fast cache or whatever extra goodies in that department that
> > > > > one could get. Weirdly, some processors in the past, that were seen as being pretty bad - such as dual
> > > > > core Phenom2 chips for example - were favored a lot for the tasks and working that I described.
> > > > >
> > > > > However, the software environment even in the area of industry I describe has changed a lot
> > > > > now. What we're seeing more and more each day, with greater network bandwidth in fixed lines
> > > > > especially, and use of 'cloud' for file storage as opposed to local physical servers, and
> > > > > with a greater amount of sophisticated 'work sharing' happening,... the databases are getting
> > > > > larger, more sophisticated and more people are working on them 'at the same time'.
> > > > >
> > > > > This changes things from a point of view of what the processor
> > > > > has to do. Because, nowadays the user is opening
> > > > > and working on a portion of an overall database that is stored
> > > > > somewhere and is shared across a team of people.
> > > > > And those people, are constantly sending checks for updates back and forth to each other. In the past, that
> > > > > used to only happen once a day. It still doesn't happen
> > > > > very regularly (it's not like 'high frequency trading'
> > > > > for example). But still, there are enough times of the day,
> > > > > during which those additional cores in the CPU package,
> > > > > actually get to do something useful with themselves. As
> > > > > opposed to sitting there, taking up space on the die,
> > > > > that could be better used for larger cache, more power individual cores or whatever.
> > > > >
> > > > > It's funny how that balance has altered, only recently in this area of desktop computing. Like, there
> > > > > are enough times in the day, where you want to 'chop' that 10 minutes database synchronization task,
> > > > > down to five minutes or less,... so as not to constantly interrupt one's workflow. In the past, I
> > > > > wouldn't have bothered going from dual core up to quad core even say. Because the quadcore was only
> > > > > valuable to me, maybe two times a day. But I'm finding, that that is more like ten, or twenty times
> > > > > a day now. And that radically alters one's requirements from one's hardware. Don't get me wrong either,
> > > > > there are many times in a day, when faster single processor performance is 'all' that I require. But
> > > > > it's more a hybrid now, than it used to be a choosing between single or multi-core.
> > > >
> > > > You presumably don't know enough history to understand the context of my remarks.
> > > >
> > > > There has been something of a tradition on RWT of justifying SMT as a great technology,
> > > > in fact so great that Apple and ARM should copy it ASAP (with the corollary that,
> > > > obvious Apple's and ARM's CPUs must suck if they don't implement it).
> > >
> > > I don’t recall anyone here actually claiming that lack of SMT hurts Apple.
> > > If Apple had thought SMT was useful to them then SMT would be in Apples CPU’s by now.
> > > The Apple team is in no way lacking talent to do whatever they think is best.
> >
> > SMT takes a high performance single thread core and turns it into a flock of chickens.
> > Apple already competes with flocks of chickens, and crushes them.
> >
> > A real flock of chickens takes 1/4 the die and power of a real CPU, it’s an
> > insane thing to cripple a real CPU by turning it into a flock of chickens.
> > This only makes sence for Intel, as Intel does not have good chickens to bundle as little’s with
> > it’s big cores. Doing a good x86 little would cost money, Intel is cheap. Intel would rather force
> > you to buy more big power hungry cores, perf/watt be damned, the peasants can eat cake.
> >
> > For low priority background tasks you can have little low power companion cores, running at the same time
> > as your big cores, not instead of your big cores. This is my guess at what Apple is doing, it gives the big
> > cores an effective speed boost at little additional power
> > and die size, that competitors do not see. Competitors
> > end up using quad cores and losing in power and performance to Apple’s dual core with chickens.
>
> WOW, this whole tree of SMT threads instantly died after my post. ;)
>
> One can use this information to predict what configuration the powerful but
> cheap laptop Linus will be buying in 2 or 8 years or so will look like.
> A PlayStation configuration of 4-8 real X86 cores surrounded by 16-32 helper cores.
> The helper cores will be more than powerful enough to compile Linux.
> When I say PlayStation I mean it, the very same chip as will be being
> used in that generation of consoles, also available in the PC market.
>
> X86 will be outdated as a CPU architecture by then, but Linus will not mind.
>
> Throwing this out as an alternative to a nearly identical ARM or post
> ARM ARM core configuration of 4-8 real and 16-32 helper cores.
>
> > > > I, in contrast, contend that SMT was and is a bad technology choice whose only justification
> > > > is business/legal (eg IBM's bizarre contortions with POWER to create things that look for all
> > > > the world like dual cores, but which can claim to be single cores for licensing purpose).
> > >
> > > For server workloads the mm2 tradeoff of SMT clearly is a win.
> > > Any other change pales in comparison.
> > >
> > > This is why AMD added SMT even though SMT no longer makes sense on the desk/laptop.
> > > At 2 cores and below one can make a weak case for SMT, not today with plentiful cores.
> > >
> > > For phones there is no case for SMT, higher heat offsets the gains.
> > >
> > > > I base this argument on the following claims
> > > > - SMT is a way to get more work out of execution units. But EU's are the most plentiful resource on
> > > > modern cores! The limited resources are memories (eg L1 caches) and SMT overloads those. This is why,
> > > > on any core that doesn't suck, you get such a pitiful win from SMT -- what you gain in POTENTIAL higher
> > > > instruction throughput you lose in more waiting on memory because of smaller effective caches
> > > > - SMT introduces security concerns, and once you start worrying about things
> > > > like fairness, it also starts to impact your single-threaded pipeline
> > > > - if you want the optionality value of higher throughput, a better way to do that
> > > > is through the provision of a sea of small cores alongside your large cores.
> > > > - if you want to make your large core slightly more performant without the negative
> > > > aspects of traditional SMT, the way to do this is to reduce the problem to co-running
> > > > threads from the *same address space* and with no concerns about fairness.
> > > >
> > > > For this dissension from the party line, I have been repeatedly excoriated, being told, for example, with
> > > > the full politeness that only the internet can offer, that I know nothing of modern CPUs or software.
> > > >
> > > > So yes, it is more than a little amusing to see that Intel seems (IMHO) to have, belatedly, agreed
> > > > (more or less) with my analysis. We'll see where they're headed with this long term...
>
[Warning: most of the hard numbers used below involve floorspace data. These were used as they were far easier to find and compute. Considering Intel's margins, the cost to produce a mm**2 is almost irrelevant, but leakage should scale at least somewhat with it. Trying to compare power and performance between big and small x86 is like apples and oranges, so I can't say I tried (I couldn't even find real "cat" performance data, let alone power consumption). I also seem to repeat the "fabric" argument a few times, some of the bits were written out of order. But the fabric problems are sufficiently ignored enough in the original post to warrant more than a few mentions].

I can't tell what Intel is doing with goldmont/goldmont+/tremont. I was pretty sure it was canceled, but googling leaves me unsure. On the AMD side, puma appears to be the end of the line for the "cat" architecture and all AMD cores are going to be zen based for the immediate future. Maybe you can dig up Intel atom power/size/performance characteristics, but it is pretty clear why the cats are canceled.

To be honest, I had some hopes that AMD might try something like this with their earlier line of APUs, although this was mostly based on a few observations:
1. Bulldozer performance was bad enough to give a jaguar core a chance
2. Bulldozer already had an eight core fabric that wasn't being fully used in the ALU
3. Slipping a few cats in the remaining "fabric holes" wouldn't take up much room at all

I suspect that windows wouldn't properly handle such different core powers, so AMD would likely never consider building such (and they really couldn't afford the risk of a failed chip).

With zen, AMD has a completely different set of variables. A Puma core is said to be 3.1mm on 28nm, and we would expect our chickens to be in a flock of 4 cores tied to a 2M L2 (pumas work in groups of 2-4 tied to a common 1-2M L2 cache). Trying to find some values, we have a few numbers here: https://www.anandtech.com/show/11170/the-amd-zen-and-ryzen-7-review-a-deep-dive-on-1800x-1700x-and-1700
Core: 44mm
L2: 1.5mm (per core) for 512k
8M L3: 16mm
gotcha: the whole chip is 192mm with 16M L3 (zen+ appears to have more, or possibly simply harvested more). 44*4+16*2=208mm, while the cores+L3 clearly make up less than half the die. Near as I can tell, each core is 11mm.

Hypothetical flock of 4 feline chickens:
4*(3.1*(scaling factor)+6mm (for 2M l2@14nm)
using a scaling factor of .60 (for a full node) we get 13.44nm. They simply take up more floorspace than a zen.

So our flock of chickens is no smaller than a SMT zen processor. The real catch is that the per/core [single threaded] performance of a puma+ core is likely 1/4 (or less) of a zen core even before scaling factors come into play (IPC easily one half of zen [it was less than bulldozer which is about half of zen] and clockspeeds slightly higher than half of zen). Maybe the thing can justify itself in power/W, but I wouldn't count on it. Even if the core could do it on it's own, the cost of the rest of the chip to support the core won't drop by more than 25%, and I doubt it will drop at all. Don't forget that nearly all benchmarks dealing with high-thread performance are looking at floating point: Intel's AVX512 experience shows that simply spamming floating point units may simply hit power limits that much sooner, never mind if they are issued by cat or zen. Maybe database applications will work with spammed cats, but I suspect that they are already thread limited with epycs (and if the database fits in memory, the AMD fabric kills performance, I'm pretty sure Intel managed to find such a benchmark).

Don't forget that AMD's memory fabric leaves with a dilemma: if the threads are using the same memory footprint, you beat yourself against AMD's memory fabric. If they are independent, you will need at least 4M (8mm) of L3 per flock just to *only* linearly increase the amount of bandwidth consumed by the memory interface: if you spammed more cores (and are hitting memory limits) you will need a lot more L3. Even with a better fabric (like a presumed Intel chip might have) will still need plenty of L3 to avoid memory bottlenecks.

As far as "spamming these flocks of feline chickens" we run into even worse issues: the zepplin die is 23% core, 17% L3, and 60% "everything else". It would be a miracle if "everything else" merely scaled linearly with cores (either floorspace or power consumption). Also AMD really doesn't have a good interconnect fabric: spamming strong zepplin cores (Epyc and Threadripper) seems to work fine on most jobs, but I doubt you want to push it with weak kittens. Intel might not have this problem, but I'm sure that only means more power used in the "everything else/uncore", especially once you start spamming cores.

The only thing that has changed since Seymour Cray uttered that bit about "flock of chickens vs. strong oxen" is that oxen have hit an upper limit in size. If you are going to spam cores (via sockets or on die), spam oxen. The issues on cache coherent networks, the scaling factors, and simple physics all conspire against spamming cores, and they hurt the less powerful ones a lot more. Intel is making plenty of money spamming oxen, and the obituary for their "flock of chickens" was last week (they built it but nobody came).
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Ha ha ha ha haMaynard Handley2018/07/25 04:42 PM
  Ha ha ha ha haDB2018/07/25 04:55 PM
    Ha ha ha ha haMaynard Handley2018/07/25 05:07 PM
      Ha ha ha ha hawumpus2018/07/25 06:22 PM
        Ha ha ha ha hagallier22018/07/26 01:37 AM
          Ha ha ha ha haDavid Hess2018/07/26 03:41 PM
        Ha ha ha ha haDavid Hess2018/07/26 03:29 PM
          Ha ha ha ha hagallier22018/07/27 12:12 AM
            Ha ha ha ha harwessel2018/07/27 08:48 AM
              Ha ha ha ha hawumpus2018/07/27 09:46 AM
                Ha ha ha ha haDavid Hess2018/07/29 06:01 PM
                  Ha ha ha ha haAdrian2018/07/29 09:18 PM
                    Ha ha ha ha harwessel2018/07/30 12:48 AM
                  Ha ha ha ha hagallier22018/07/30 01:13 AM
                    Ha ha ha ha haMichael S2018/07/30 04:02 AM
      8C/8T > 6C/12TAndrew2018/07/25 06:57 PM
        8C/8T > 6C/12TAndrew2018/07/25 06:59 PM
        8C/8T > 6C/12TPassing Through2018/07/26 02:57 AM
        not all latencies are created equalJeff S.2018/07/26 06:26 AM
      Ha ha ha ha haDB2018/07/25 09:43 PM
      Ha ha ha ha haDoug S2018/07/25 11:29 PM
      Ha ha ha ha haFoo_2018/07/26 03:01 AM
      Ha ha ha ha haDavid Hess2018/07/26 03:39 PM
        Ha ha ha ha haMaynard Handley2018/07/26 03:57 PM
          Ha ha ha ha haDavid Hess2018/07/26 04:20 PM
            Ha ha ha ha haAnon2018/07/27 01:33 AM
              Ha ha ha ha haDavid Hess2018/07/29 05:42 PM
          Ha ha ha ha haPassing Through2018/07/26 04:27 PM
            Simple Arithmetic Might Look LikePassing Through2018/07/26 04:47 PM
            Ha ha ha ha haTravis2018/07/26 09:25 PM
              Ha ha ha ha haPassing Through2018/07/27 04:52 AM
          Ha ha ha ha haJukka Larja2018/07/27 10:17 AM
  Ha ha ha ha haDavid Kanter2018/07/25 11:50 PM
    Ha ha ha ha haEmil Briggs2018/07/26 04:50 AM
      Ha ha ha ha haJukka Larja2018/07/26 07:05 AM
        Ha ha ha ha haEmil Briggs2018/07/26 08:06 AM
          Ha ha ha ha hawumpus2018/07/26 11:38 AM
            Intel MarketingAndrew Clough2018/07/26 12:18 PM
            Intel's End User UpgradesDavid Hess2018/07/26 03:49 PM
              consumer backlash (NT)anonymous22018/07/26 06:50 PM
        Some 'Atom' too (eg. Bonnell) (NT)anonymous22018/07/26 10:10 AM
      Ha ha ha ha haPer Hesselgren2018/07/26 08:50 AM
        Multi-core versus singlePassing Through2018/07/26 10:11 AM
          Multi-core versus singleMaynard Handley2018/07/26 12:22 PM
            Multi-core versus singletarlinian2018/07/26 12:36 PM
            again... 8C/8T > 12C/12TAndrew2018/07/26 12:58 PM
              You mean 8C/8T > 6C/12T (NT)juanrga2018/07/26 07:29 PM
                Correct, thank you (NT)Andrew2018/07/27 03:25 PM
            Multi-core versus singleBrett2018/07/26 01:03 PM
              SMT flock of chickens stupidity.Brett2018/07/28 12:21 AM
                SMT flock of chickens stupidity.Brett2018/08/05 10:27 PM
                  SMT flock of chickens stupidity.wumpus2018/08/06 09:54 AM
                    Former chicken farmer sentimentsJeff S.2018/08/06 11:48 AM
                      Former chicken farmer sentimentsSimon Farnsworth2018/08/06 12:49 PM
                        Former chicken farmer sentimentsJeff S.2018/08/06 01:06 PM
                        Former chicken farmer sentimentsMaynard Handley2018/08/06 01:22 PM
                          Former chicken farmer sentimentsSimon Farnsworth2018/08/07 02:05 AM
                          Former chicken farmer sentimentsFoo_2018/08/07 07:35 AM
                            Former chicken farmer sentimentsJukka Larja2018/08/07 09:17 AM
                              Former chicken farmer sentimentsDoug S2018/08/07 10:05 AM
                                Former chicken farmer sentimentsJukka Larja2018/08/08 06:22 AM
                              Former chicken farmer sentimentsFoo_2018/08/07 12:06 PM
                                Former chicken farmer sentimentsJukka Larja2018/08/08 06:08 AM
                                  Former chicken farmer sentimentsTravis2018/08/08 10:25 AM
                                    Former chicken farmer sentimentsFoo_2018/08/08 01:32 PM
                                      Former chicken farmer sentimentsanonymou52018/08/08 02:13 PM
                                        Former chicken farmer sentimentsFoo_2018/08/08 03:44 PM
                                          Former chicken farmer sentimentsTravis2018/08/08 09:33 PM
                                            Former chicken farmer sentimentsMichael S2018/08/09 12:40 AM
                                    Former chicken farmer sentimentsgallier22018/08/09 02:20 AM
                          Former chicken farmer sentimentsDavid Hess2018/08/11 07:38 PM
                        Former chicken farmer sentimentsLinus Torvalds2018/08/06 02:52 PM
                          Former chicken farmer sentimentsTravis2018/08/06 06:21 PM
                            Former chicken farmer sentimentsanon2018/08/07 03:40 AM
                              Former chicken farmer sentimentsTravis2018/08/07 08:04 AM
                                Former chicken farmer sentimentsanon2018/08/08 08:23 PM
                          Former chicken farmer sentimentsJeff S.2018/08/06 09:00 PM
                            Former chicken farmer sentimentsTravis2018/08/08 08:46 PM
                              Former chicken farmer sentimentsLinus Torvalds2018/08/09 01:20 PM
                                Former chicken farmer sentimentsPassing Through2018/08/09 04:42 PM
                                Former chicken farmer sentimentsBlaine2018/08/10 11:43 AM
                          a better nuanced statement on coherencyJeff S.2018/08/07 08:57 AM
                            a better nuanced statement on coherencywumpus2018/08/07 11:30 AM
                              a better nuanced statement on coherencyJeff S.2018/08/07 11:38 AM
                                a better nuanced statement on coherencydmcq2018/08/07 12:12 PM
                                  a better nuanced statement on coherencyMaynard Handley2018/08/07 03:25 PM
                                    a better nuanced statement on coherencyhobel2018/08/08 08:21 AM
                                      a better nuanced statement on coherencyMaynard Handley2018/08/08 09:33 AM
                                    a better nuanced statement on coherencydmcq2018/08/08 10:03 AM
                                    CTOS - RIPTIm McCaffrey2018/08/10 08:35 AM
                                a better nuanced statement on coherencyanon2018/08/07 01:41 PM
                                  a better nuanced statement on coherencyanon.12018/08/08 09:51 AM
                                    a better nuanced statement on coherencyanon2018/08/08 01:46 PM
                                      a better nuanced statement on coherencyRicardo B2018/08/09 09:32 AM
                                        Do you mean SGI Altix UV?Michael S2018/08/10 09:23 AM
                                          Do you mean SGI Altix UV?Ricardo B2018/08/10 10:45 AM
                                            Do you mean SGI Altix UV?Michael S2018/08/11 12:23 PM
                                              Do you mean SGI Altix UV?Anon HPC2018/08/12 07:04 AM
                                                Do you mean SGI Altix UV?Wes Felter2018/08/12 10:50 AM
                                                  Do you mean SGI Altix UV?Anon HPC2018/08/12 11:43 AM
                                              Do you mean SGI Altix UV?Kevin G2018/08/12 12:00 PM
                                        a better nuanced statement on coherencywumpus2018/08/10 09:31 AM
                                          a better nuanced statement on coherencyRicardo B2018/08/10 11:00 AM
                                            a better nuanced statement on coherencyJeff S.2018/08/10 11:53 AM
                                              a better nuanced statement on coherencyanon2018/08/10 11:59 PM
                                                IBM systemP coherencyDavid Kanter2018/08/11 09:20 AM
                                        a better nuanced statement on coherencyanon2018/08/10 02:49 PM
                                          a better nuanced statement on coherencyRicardo B2018/08/11 04:38 PM
                                            a better nuanced statement on coherencyanon2018/08/12 03:11 AM
                                              a better nuanced statement on coherencyRicardo B2018/08/12 11:39 AM
                                                a better nuanced statement on coherencyMichael S2018/08/13 07:06 AM
                                                  a better nuanced statement on coherencyrwessel2018/08/13 11:39 AM
                                                  a better nuanced statement on coherencyLinus Torvalds2018/08/13 12:21 PM
                                                    a better nuanced statement on coherencyMaynard Handley2018/08/13 01:53 PM
                                                      a better nuanced statement on coherencyLinus Torvalds2018/08/13 05:47 PM
                                                        a better nuanced statement on coherencyMaynard Handley2018/08/13 06:33 PM
                                                          a better nuanced statement on coherencyLinus Torvalds2018/08/13 07:52 PM
                                                            a better nuanced statement on coherencyMaynard Handley2018/08/13 10:04 PM
                                                              a better nuanced statement on coherencyLinus Torvalds2018/08/13 10:42 PM
                                                                a better nuanced statement on coherencyMaynard Handley2018/08/13 11:44 PM
                                                                  a better nuanced statement on coherencydmcq2018/08/14 04:46 AM
                                                                a better nuanced statement on coherencyEtienne2018/08/15 03:11 AM
                                                                  a better nuanced statement on coherencyTravis2018/08/15 03:50 PM
                                                                    a better nuanced statement on coherencyEtienne2018/08/16 02:16 AM
                                                                      a better nuanced statement on coherencyMichael S2018/08/16 04:06 AM
                                                                      a better nuanced statement on coherencyRicardo B2018/08/16 07:54 AM
                                                                        a better nuanced statement on coherencyMichael S2018/08/16 08:33 AM
                                                                          a better nuanced statement on coherencyRicardo B2018/08/16 08:59 AM
                                                                            a better nuanced statement on coherencyMichael S2018/08/17 03:08 AM
                                                                              a better nuanced statement on coherencyRicardo B2018/08/17 11:41 AM
                                                                        a better nuanced statement on coherencyEtienne2018/08/17 04:02 AM
                                                                          a better nuanced statement on coherencyMichael S2018/08/17 05:03 AM
                                                                          a better nuanced statement on coherencyRicardo B2018/08/17 12:00 PM
                                                            ignoring message passing over shared memorywumpus2018/08/15 10:28 AM
                                                              ignoring message passing over shared memoryRicardo B2018/08/15 11:24 AM
                                                              ignoring message passing over shared memoryLinus Torvalds2018/08/15 03:28 PM
                                                                ignoring message passing over shared memorydmcq2018/08/16 06:00 AM
                                                                  ignoring message passing over shared memorydmcq2018/08/16 06:16 AM
                                                                    WinFS (NT)Michael S2018/08/16 06:20 AM
                                                                    ignoring message passing over shared memoryPassing Through2018/08/16 07:06 AM
                                                                  ignoring message passing over shared memoryRicardo B2018/08/16 07:19 AM
                                                                    ignoring message passing over shared memoryPassing Through2018/08/16 07:49 AM
                                                                      ignoring message passing over shared memoryPassing Through2018/08/16 07:56 AM
                                                          a better nuanced statement on coherencyanon2018/08/13 08:18 PM
                                                      a better nuanced statement on coherencyJukka Larja2018/08/14 08:02 AM
                                                      a better nuanced statement on coherencyTravis2018/08/15 12:27 PM
                                                        a better nuanced statement on coherencyMaynard Handley2018/08/15 01:19 PM
                                                          a better nuanced statement on coherencyTravis2018/08/15 02:27 PM
                                                      a better nuanced statement on coherencyRicardo B2018/08/15 02:07 PM
                                                    a better nuanced statement on coherencyDoug S2018/08/13 02:21 PM
                                                  a better nuanced statement on coherencyRicardo B2018/08/14 08:20 AM
                                                a better nuanced statement on coherencyanon2018/08/13 03:16 PM
                                                  a better nuanced statement on coherencyanon2018/08/13 08:28 PM
                                                    a better nuanced statement on coherencyanon2018/08/14 06:07 AM
                                                      a better nuanced statement on coherencyanon2018/08/14 08:38 AM
                                                        a better nuanced statement on coherencyanon2018/08/14 09:05 AM
                                                          a better nuanced statement on coherencyanon2018/08/14 07:49 PM
                                                  a better nuanced statement on coherencyRicardo B2018/08/14 08:57 AM
                                                    a better nuanced statement on coherencyanon2018/08/14 09:18 AM
                                                      a better nuanced statement on coherencyRicardo B2018/08/14 10:13 AM
                                                        a better nuanced statement on coherencyanon2018/08/14 11:28 AM
                                                          a better nuanced statement on coherencyRicardo B2018/08/14 04:14 PM
                                                            a better nuanced statement on coherencyLinus Torvalds2018/08/14 04:35 PM
                                                              a better nuanced statement on coherencyRicardo B2018/08/14 05:22 PM
                                                              Supermicro MicroBlade MBI-6218G-T41X"Mark Roulo2018/08/14 09:02 PM
                                                              a better nuanced statement on coherencyPassing Through2018/08/15 02:08 AM
                                                            a better nuanced statement on coherencyanon2018/08/14 04:53 PM
                                                              a better nuanced statement on coherencyRicardo B2018/08/14 05:39 PM
                                                            a better nuanced statement on coherencyAdrian2018/08/14 11:58 PM
                                                              Value of an Option Theory in Finance Passing Through2018/08/15 01:38 AM
                                                                Value of an Option Theory in Finance Passing Through2018/08/15 03:13 AM
                                                              a better nuanced statement on coherencyRicardo B2018/08/15 09:58 AM
                                                                a better nuanced statement on coherencyAdrian2018/08/15 11:31 AM
                                                                  a better nuanced statement on coherencyRicardo B2018/08/15 12:36 PM
                                                                    a better nuanced statement on coherencyAdrian2018/08/16 11:01 PM
                                              a better nuanced statement on coherencyKevin G2018/08/12 01:26 PM
                                                a better nuanced statement on coherencyanon2018/08/13 02:40 PM
                                                  a better nuanced statement on coherencydmcq2018/08/14 05:15 AM
                                                    a better nuanced statement on coherencyanon2018/08/14 05:51 AM
                                                      a better nuanced statement on coherencydmcq2018/08/14 10:26 AM
                                                        a better nuanced statement on coherencyRicardo B2018/08/14 10:57 AM
                                                          a better nuanced statement on coherencydmcq2018/08/15 09:32 AM
                                                            a better nuanced statement on coherencyRicardo B2018/08/15 10:40 AM
                            a better nuanced statement on coherencySimon Farnsworth2018/08/07 01:06 PM
                        Former chicken farmer sentimentsanon2018/08/07 03:23 AM
                      Former chicken farmer sentimentsPassing Through2018/08/06 01:38 PM
                        Former chicken farmer sentimentsanonymou52018/08/06 04:12 PM
                      zen/zeppelin infinity fabric port count?Heikki Kultala2018/08/06 02:49 PM
                        yep, actually 11Jeff S.2018/08/06 02:56 PM
                    SMT flock of chickens stupidity.Maynard Handley2018/08/06 12:27 PM
                    SMT flock of chickens stupidity.Brett2018/08/06 09:07 PM
                      Zen core has (more than) double the flops/cycle of a cat coreHeikki Kultala2018/08/06 10:31 PM
                        Zen core has (more than) double the flops/cycle of a cat coreMichael S2018/08/07 01:17 AM
                          Register pressure and SMTPaul A. Clayton2018/08/07 06:16 AM
                            Register pressure and SMTEric Bron2018/08/09 01:33 AM
                        Zen core has (more than) double the flops/cycle of a cat coredmcq2018/08/07 07:23 AM
                        Zen core has (more than) double the flops/cycle of a cat corewumpus2018/08/07 08:54 AM
                        Zen core has (more than) double the flops/cycle of a cat coreBrett2018/08/07 11:49 PM
                          revelationMichael S2018/08/08 12:46 AM
                            revelationLinus Torvalds2018/08/08 10:04 AM
                              revelationJeff S.2018/08/08 11:32 AM
                                revelationAnon2018/08/08 08:51 PM
                                  revelationdmcq2018/08/09 06:05 AM
                                  revelationwumpus2018/08/09 06:57 AM
                                    revelationMichael S2018/08/09 07:08 AM
                                      rambling comparison of the 320C080 to cell wumpus2018/08/10 07:59 AM
                                        rambling comparison of the 320C080 to cell none2018/08/10 09:20 AM
                                          rambling comparison of the 320C080 to cell wumpus2018/08/10 10:03 AM
                                        rambling comparison of the 320C080 to cell Jukka Larja2018/08/10 11:37 AM
                                          rambling comparison of the 320C080 to cell Linus Torvalds2018/08/10 02:05 PM
                                            rambling comparison of the 320C080 to cell Jukka Larja2018/08/10 10:21 PM
                                              rambling comparison of the 320C080 to cell Doug S2018/08/11 02:34 AM
                                                rambling comparison of the 320C080 to cell Jukka Larja2018/08/11 05:21 AM
                                              rambling comparison of the 320C080 to cell Kevin G2018/08/12 01:43 PM
                                                rambling comparison of the 320C080 to cell Jukka Larja2018/08/13 06:06 AM
                                        rambling comparison of the 320C080 to cell Action_P2018/08/10 06:20 PM
                                          rambling comparison of the 320C080 to cell anon2018/08/11 08:05 AM
                                            rambling comparison of the 320C080 to cell Action_P2018/08/11 10:53 AM
                                              rambling comparison of the 320C080 to cell anon2018/08/11 03:43 PM
                                                rambling comparison of the 320C080 to cell Action_P2018/08/13 04:39 PM
                                                  rambling comparison of the 320C080 to cell Brett2018/08/13 08:09 PM
                                                    Would Cell have been more tolerable with an OoO core?SoftwareEngineer2018/08/14 01:15 AM
                                                      Would Cell have been more tolerable with an OoO core?wumpus2018/08/14 07:32 AM
                                                        Would Cell have been more tolerable with an OoO core?Joe old timer2018/08/14 08:26 PM
                                                  rambling comparison of the 320C080 to cell anon2018/08/14 06:02 AM
                              revelationanon2018/08/08 07:09 PM
                            revelationGabriele Svelto2018/08/08 12:01 PM
                              revelationBrett2018/08/08 09:05 PM
                                revelationGabriele Svelto2018/08/09 04:15 PM
                                  revelationBrett2018/08/09 08:46 PM
                                    revelationGabriele Svelto2018/08/10 02:07 AM
                                      revelationBrett2018/08/10 11:01 AM
                                        revelationGabriele Svelto2018/08/10 02:25 PM
                                          revelationBrett2018/08/11 02:00 AM
                                            On PlayStation GPUsGabriele Svelto2018/08/11 06:56 AM
                                              On PlayStation GPUsBrett2018/08/11 12:46 PM
                                                On PlayStation GPUsJoe old timer2018/08/14 08:42 PM
                                                  On PlayStation GPUsBrett2018/08/15 12:40 AM
                              revelationgallier22018/08/09 04:53 AM
                                revelationdmcq2018/08/09 06:10 AM
                                  revelationgallier22018/08/09 08:05 AM
                                  revelationJukka Larja2018/08/09 08:34 AM
                        Zen core has (more than) double the flops/cycle of a cat corejuanrga2018/08/08 04:18 AM
                          Really weak cakculations , exrapolating after extrapolation Heikki Kultala2018/08/08 10:28 PM
                            Really weak cakculations , exrapolating after extrapolation juanrga2018/08/10 02:47 AM
                              typojuanrga2018/08/10 02:55 AM
                      SMT flock of chickens stupidity.Anon2018/08/07 04:23 AM
                  SMT flock of chickens stupidity.Linus Torvalds2018/08/06 10:32 AM
                    SMT flock of chickens stupidity.mpx2018/08/07 05:51 AM
                      IBM POWER9 core-counting gamesJeff S.2018/08/07 06:41 AM
                      SMT flock of chickens stupidity.Simon Farnsworth2018/08/07 06:53 AM
                        SMT flock of chickens stupidity.anon.12018/08/07 10:30 AM
                    SMT flock of chickens stupidity.Brett2018/08/07 11:33 PM
                      Marketed prize vs manufacturing cost of SMTHeikki Kultala2018/08/08 12:01 AM
                        Marketed prize vs manufacturing cost of SMTBrett2018/08/08 09:55 PM
                          Marketed prize vs manufacturing cost of SMTHeikki Kultala2018/08/08 10:40 PM
                            Marketed prize vs manufacturing cost of SMTBrett2018/08/09 09:25 PM
            Multi-core versus singlePassing Through2018/07/26 01:29 PM
            Multi-core versus singleanon2018/07/26 08:23 PM
            Multi-core versus singleDoug S2018/07/26 11:41 PM
              SMT already present on smartphonesmpx2018/07/27 03:26 AM
                SMT or FgMT? (NT)Heikki Kultala2018/07/27 09:28 PM
            Multi-core versus singlewumpus2018/07/27 07:05 AM
    Ha ha ha ha haanon.12018/07/26 09:56 AM
      Ha ha ha ha haPassing Through2018/07/26 10:24 AM
        Capture and encoding - real world examplePassing Through2018/07/26 10:30 AM
        Ha ha ha ha haAaron Spink2018/07/26 05:48 PM
          Ha ha ha ha haPassing Through2018/07/27 05:11 AM
          A Top Tier Streamer ? Passing Through2018/07/27 05:42 AM
            A Top Tier Streamer ? Passing Through2018/07/27 06:16 AM
              A Top Tier Streamer ? Passing Through2018/07/27 10:59 AM
                A Top Tier Streamer ? Kevin G2018/07/29 02:03 PM
      Ha ha ha ha haMaynard Handley2018/07/26 07:00 PM
        Ha ha ha ha haJukka Larja2018/07/27 09:59 AM
        Ha ha ha ha haanon.12018/07/27 08:24 PM
  i9-9900k, i7-9700k, and i5-9600k specificationsjuanrga2018/07/26 03:06 AM
    9900k/9700k L3 difference too?Jeff S.2018/07/26 06:07 AM
      9900k/9700k L3 difference too?Anon2018/07/26 06:16 AM
      9900k/9700k L3 difference too?Jukka Larja2018/07/26 07:10 AM
        silicon harvesting?hobold2018/07/26 07:40 AM
  Top chip still has hyperthreadingSeriously2018/07/28 08:22 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?