# Mill *is* a speculation

By: anon (spam.delete.delete.delete@this.this.this.spam.com), January 7, 2019 1:16 pm
juanrga (juju.delete@this.jujuju.com) on January 7, 2019 9:22 am wrote:
> anon (spam.delete.delete.delete@this.this.this.spam.com) on January 7, 2019 5:24 am wrote:
>
> > If they can get their 33 or 37, can't remember which, instructions
> > per cycle all is well. Branch mispredictions
> > won't affect the significantly more or less than anyone
> > else. But they actually need to get that IPC. I can't
> > remember the exact numbers but I think it was something like
> > 8 ALUs, 8 load units, 4 store units on a Mill Gold.
> > So if we give them the benefit of the doubt, that they can
> > move instructions around in a way that is comparable
> > to OoOE, then assume linear scaling, so 8 ALUs instead of
> > 4 doubles the IPC, then add a bit on top of that because
> > they don't share ports we get maybe 3 times the IPC at 1/3
> > the clockrate. So it's not enough to get about the
> > same IPC per ALU that OoOE would get, the other slots that
> > bring up the count to 30+ need to be useful enough
> > to get it way past that or it won't actually get anywhere close to the 2x advantage they need.
> >
>
>
>
> "37-wide cores for get what? Which is the average IPC on general code?"
>
> I am expecting average IPC to be about one order of magnitudes
> smaller than the width. Michael expects IPC of about 2

Would you also expect 4 (or 8 if you count the back-end) OoOE to get 0.4 IPC?

If your argument is "it can't work as well as OoOE because it's not OoOE" then there's no need to bringt width into the argument at all. Your premise is that it can't work, therefore it can't work, the perfect circular argument.

What I'm saying is that if it works as well as OoOE and we give it the benefit of the doubt with linear scaling and ALU ports never being blocked you still only get around 3x. How much that actually is depends on the code, I mean an OoO architecture with 8 wide back-end and 4 wide front-end can't get 8 IPC either, but going from 3.x IPC to 10+ IPC with a backend about 3 times as wide seems on the upper end of realistic expectations. It require everything to work out and near linear scaling, but it's not absurd. But it's the upper end for when things work out. 3x IPC sounds great but if it runs at 1/3 clockrate then we're right back to where we started.

So instead of the circular argument I'm saying that if things work out well then it only ends up being as fast as OoOE, which is nowhere near revolutionary since we already got that.

Or let's look at some simple math. What's the average percentage of ALU operations? 40%? 50%? The usual numbers thrown around are 50/20/10/20 for ALU/load/store/branch. So where do you end up with 8 ALUs? 16-20 IPC, theoretical max I'd say. Similarly Skylake could to 8 with it's 4 ALUs and 8 ports, but the front-end is going to limit it to 6 or less. If there's a 3x difference in clockrate the IPC advantage melts away. So on some well behaved 4+ IPC code running out of the uop I don't see Skylake or Zen being much slower. The ~2 IPC on OoOE cases are where it's at. Can you get 6+ IPC out of those at 1/3 the frequency? Possibly. Can you get 10+ or 12+ IPC? Just working as well as OoOE is not nearly enough for that, it needs to be way better. And that's where the "magic compiler" doubt kicks in.
TopicPosted ByDate
Independent micro threadsTravis Downs2018/12/30 04:18 PM
Independent micro threadsTravis Downs2018/12/30 07:34 PM
Independent micro threadsTravis Downs2018/12/30 09:06 PM
Independent micro threadsTravis Downs2018/12/31 12:42 AM
Independent micro threadsMichael S2019/01/02 02:01 AM
Independent micro threadsMaynard Handley2019/01/02 03:29 PM
Independent micro threadsMontaray Jack2019/01/01 02:12 AM
Independent micro threadsPaul A. Clayton2018/12/30 05:15 PM
Independent micro threadsDavid Hess2018/12/30 06:47 PM
Independent micro threadsDavid Hess2018/12/30 07:26 PM
Independent micro threadsDoug S2018/12/31 10:26 AM
Independent micro threadsDavid Hess2018/12/31 09:32 PM
Independent micro threadsDoug S2019/01/01 12:40 AM
Independent micro threadsDavid Hess2019/01/01 11:41 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Domaldel2019/01/01 12:40 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Domaldel2019/01/01 12:44 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Domaldel2019/01/01 12:49 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Domaldel2019/01/01 12:51 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Simon Farnsworth2019/01/01 06:05 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Domaldel2019/01/01 08:01 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")Maynard Handley2019/01/01 01:17 PM
Taking things to extremes.Domaldel2019/01/01 05:44 PM
Yes, I know, the forum is named *Real* World Tech, but I'm thinking that perhapsDomaldel2019/01/01 05:52 PM
Yes, I know, the forum is named *Real* World Tech, but I'm thinking that perhapsMontaray Jack2019/01/02 07:26 AM
Yes, I know, the forum is named *Real* World Tech, but I'm thinking that perhapsMontaray Jack2019/01/02 08:21 AM
Taking things to extremes.Maynard Handley2019/01/01 06:55 PM
Taking things to extremes.Kevin G2019/01/04 08:57 AM
Overcomming thermal limits of a high dencity 3D arcitecture (Formerly "Independent micro threads")David Hess2019/01/01 11:36 AM
Independent micro threadsTravis Downs2018/12/30 07:38 PM
Independent micro threadsTravis Downs2018/12/30 09:48 PM
Mill and Independent micro threadsBrett2019/01/01 07:39 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Heikki Kultala2019/01/02 12:29 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Brett2019/01/02 01:15 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Heikki Kultala2019/01/02 02:22 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Brett2019/01/03 01:13 AM
"Leaf branch" is not a commonly used termHeikki Kultala2019/01/03 03:48 AM
"Leaf branch" is not a commonly used termBrett2019/01/03 04:35 AM
You lack parallelism than OoOE givesHeikki Kultala2019/01/03 07:38 AM
You lack parallelism than OoOE givesBrett2019/01/04 02:41 AM
You lack parallelism than OoOE givesBrett2019/01/04 04:10 PM
You lack parallelism than OoOE givesBrett2019/01/05 08:29 PM
Mill speculates, more parallelism than OoOE givesBrett2019/01/05 08:31 PM
Mill *is* a speculationEric Bron2019/01/06 05:04 AM
Mill *is* a speculationMichael S2019/01/06 05:53 AM
Mill *is* a speculationBrett2019/01/06 09:03 PM
Mill *is* a speculationjuanrga2019/01/06 06:10 AM
probably ~2 (NT)Michael S2019/01/06 06:51 AM
Mill *is* a speculationBrett2019/01/06 01:18 PM
Mill *is* a speculationEric Bron2019/01/06 03:36 PM
Mill *is* a speculationBrett2019/01/06 08:47 PM
Mill *is* a speculationJacob Marley2019/01/06 10:29 PM
Mill *is* a speculationBrett2019/01/07 04:24 AM
Mill *is* a speculationMichael S2019/01/07 05:23 AM
Mill *is* a speculationEric Bron2019/01/07 05:36 AM
Mill *is* a speculationBrett2019/01/07 03:40 PM
Mill *is* a speculationEric Bron2019/01/07 05:32 PM
Mill is something you don't understandHeikki Kultala2019/01/08 04:19 AM
Mill is something you don't understandMichael S2019/01/08 07:44 AM
Itanium and static vs dynamicHeikki Kultala2019/01/09 03:14 AM
Itanium and static vs dynamicPaul A. Clayton2019/01/09 08:51 AM
Mill *is* a speculationEric Bron2019/01/07 05:27 AM
Mill *is* a speculationEric Bron2019/01/07 06:23 AM
Mill *is* a speculationanon2019/01/07 06:24 AM
Mill *is* a speculationEric Bron2019/01/07 06:52 AM
Mill *is* a speculationanon2019/01/07 08:36 AM
Mill *is* a speculationEric Bron2019/01/07 09:20 AM
Mill *is* a speculationjuanrga2019/01/07 10:22 AM
Mill *is* a speculationanon2019/01/07 01:16 PM
Mill *is* a speculationanon2019/01/07 09:46 PM
Mill *is* a speculationanon2019/01/08 01:56 AM
Mill *is* a speculationanon2019/01/08 03:39 AM
Mill *is* a speculationMichael S2019/01/08 03:52 AM
Mill *is* a speculationanon2019/01/08 10:10 PM
Wasted width not wasted work.Brett2019/01/09 11:44 AM
No such thing was declared. (NT)anon2019/01/09 03:41 PM
Very simple test for new uarch ideassomeone2019/01/10 07:03 AM
Very simple test for new uarch ideasdmcq2019/01/10 07:21 AM
Very simple test for new uarch ideasDoug S2019/01/10 10:01 AM
Very simple test for new uarch ideasDan Fay2019/01/10 01:13 PM
Very simple test for new uarch ideasanonymous22019/01/10 11:03 AM
Very simple test for new uarch ideasAlberto2019/01/10 11:32 AM
Very simple test for new uarch ideasEtienne2019/01/11 03:03 AM
Very simple test for new uarch ideasFoo_2019/01/11 04:31 AM
Very simple test for new uarch ideasEtienne2019/01/11 05:51 AM
Very simple test for new uarch ideasFoo_2019/01/11 05:53 AM
Very simple test for new uarch ideasdmcq2019/01/11 06:08 AM
Very simple test for new uarch ideasEtienne2019/01/11 06:13 AM
Very simple test for new uarch ideasFoo_2019/01/11 06:54 AM
Very simple test for new uarch ideasEtienne2019/01/11 07:32 AM
Very simple test for new uarch ideasBrett2019/01/11 10:25 AM
Very simple test for new uarch ideasMegol2019/01/12 06:29 AM
Very simple test for new uarch ideasMichael S2019/01/12 09:21 AM
Word salad AI fundamentaliy brokenBrett2019/01/12 01:59 PM
Very simple test for new uarch ideasMegol2019/01/13 11:51 AM
Mill *is* a speculationanon2019/01/08 08:50 AM
Mill *is* a speculationEric Bron2019/01/08 09:03 AM
Mill *is* a speculationanon2019/01/08 09:21 AM
"Leaf branch" is not a commonly used termMichael S2019/01/03 07:57 AM
"Leaf branch" is not a commonly used termBrett2019/01/04 03:29 AM
Calls are not needed for speculation for mill if there are no side effect,and dont help if there areHeikki Kultala2019/01/08 04:28 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.anon2019/01/02 03:05 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Doug S2019/01/02 11:38 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.rwessel2019/01/02 05:53 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.anon2019/01/02 08:56 PM
itanicBrett2019/01/03 12:41 AM
itanicanon2019/01/03 03:12 AM
itanicDavid Hess2019/01/03 08:06 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.rwessel2019/01/03 09:18 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.anon2019/01/04 05:25 AM
Itanium could have been RISC or CISC - same outcomesomeone2019/01/04 07:45 AM
Itanium could have been RISC or CISC - same outcomeDoug S2019/01/04 12:39 PM
Itanium could have been RISC or CISC - same outcomeJan Olšan2019/01/04 01:58 PM
"fluffyRISC" has a namevvid2019/01/04 03:48 PM
Itanium could have been RISC or CISC - same outcomeBrett2019/01/04 03:43 PM
Itanium could have been RISC or CISC - same outcomeanonymou52019/01/04 12:41 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.David Hess2019/01/03 08:15 AM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Maynard Handley2019/01/03 12:24 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.Maynard Handley2019/01/03 12:27 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.dmcq2019/01/04 01:59 AM
EPIC target marketsFoo_2019/01/04 06:29 AM
EPIC target marketsDoug S2019/01/04 12:42 PM
Lack of future visionDoug S2019/01/04 12:57 PM
Lack of future visionBrett2019/01/04 02:59 PM
Lack of future visionDoug S2019/01/04 04:25 PM
Lack of future visionBrett2019/01/04 05:18 PM
Lack of future visionDoug S2019/01/05 12:47 AM
Lack of future visionBrett2019/01/05 02:06 PM
Lack of future visiondmcq2019/01/05 02:22 PM
Lack of future visionanon2019/01/05 03:01 PM
Lack of future visionMichael S2019/01/05 04:18 PM
Lack of future visionanon2019/01/05 06:14 PM
Lack of future visionMichael S2019/01/06 02:01 AM
Lack of future visionanon2019/01/06 03:23 AM
Mitch Alsup's MY66000 uses IF-like predication (I think) (NT)Paul A. Clayton2019/01/06 04:54 PM
??? (NT)Michael S2019/01/07 05:25 AM
88K ? (NT)anonymous22019/01/07 04:20 PM
Modestly expanded response: MY66000 predicate shadowPaul A. Clayton2019/01/07 11:53 AM
Thanks for the correction (NT)Foo_2019/01/04 04:31 PM
No. Mill does not get the hit because it does not get the benefit even when correctly predicted.sdrc2019/01/04 07:36 AM
Mill and Independent micro threadsMichael S2019/01/02 02:32 AM