Why the platform focus?

Article: Tukwila Update
By: Vincent Diepeveen (diep.delete@this.xs4all.nl), February 6, 2009 10:59 am
Room: Moderated Discussions
Default (structuredmayhem@hotmail.com) on 2/6/09 wrote:
---------------------------
>Since your post was long, I'll try to keep my response simple.
>
>128 cores with small caches one a chip is a pipe-dream. Yes, the logical core
>of the Itanium is small, but it's not that small.
>
>Take for example, Madison, the first "real" Itanium worth talking about, because
>it had a respectable amount of cache to keep the processor fed.

That is another pipedream from most HPC guys, the huge L3 caches.

From all the different software i know that runs at itaniums, ranging from CFD's, to prime number crunching, to financial data, a 1MB L3 cache is doing just as well as a bigger L3 cache.

A bigger L3 cache for itanium only did do really well at specint/specfp. Other than these 2 testsuites a real giant L3 cache is not much of a help for all real world applications i know.

I see many different types of applications and none of them has a need for a tad larger L3. That really speeds up under 0.1% for them.

At specfp/specint it's maybe factor 2 or so for a 1MB L3 versus 16MB L3?

For my own proggie Diep, if you google a bit you already can find objective measured results. Johan de Gelas measured a 512KB L2 was not slower than a 2048KB L2.

On the other hand, increasing L1 size from 64KB to 128KB
has a very healthy impact at my software. Again not all softwares around me.

>At 374mm^2 on a
>0.13 micron process, it's a big one, and at best, you'll fit two of these boys on
>a 90nm process (Montecito), four on a 65nm process (Tukwila), and eight on a 45nm
>process, and sixteen on a 32nm process.

How about improving the core and getting rid of that much L3 cache?

>Even if you nix some more cache, the best you can get is around 16/32 cores using
>your cutting-edge 45nm/32nm processes (this is ignoring the overhead of cache coherency
>support, which gets more complex as you add more cores).

In contradiction to some regular posters here, i have faith in engineers to find clever ways to solve problems. I'm really disappointed in most commissions in that sense. They are so paranoia and trusting so little people, they usually only go for 'proven technology'. No one can outclever a commission huh?

So 64 core multicore should be no problem i read from your estimate in 32 nm. That's great.

>In addition, cutting the
>cache has a LARGE detrimental effect on performance with EPIC, because the processor
>takes every branch until the test is evaluated (see my post above). You're just
>not going to get 128 cores on a chip with EPIC.

Question is: does EPIC need MORE or LESS transistors to get to the same IPC like out of order?

>As for getting rid of vectors...vectorization is a necessity today! Vectorization
>cuts-down on the instruction decode overhead of having individual instructions,
>placing the burden on software. Sure, it sucks for many tasks, but the reality
>is that the costs of adding it are less than the costs of adding another ALU/FPU
>and another dispatch pipe to keep it fed.

Well how about a chip design where you DEMAND that the same software code and program executes at n cores. So it is a multicore that reads instructions from the same instruction cache and you have garantuee that for all the n cores it is the same program executing (kind of similar to how the operating systems do it). However each core is a multicore and can be at a different spot in executing the code.

Vectorisation is evil as it requires you must make new software and that eats a lot of time which not many are going to do. Only for tiny programs you can get away usually. For experimental codes like game tree search it is very difficult to make a quick test that also vectorizes.

Most algorithms is easy to experiment with at a multicore level and near to impossible at vector level. Practical no one is doing experiments at a vector level in game tree search, yet there is millions of people involved all together there.

>And I wouldn't decrease the number of instructions in a bundle - the number of
>instructions within a bundle defines the upper-bound of IPC for Itanium. Changing
>this would also mean you'd need a new set of optimizing compilers.

Well this is a valid point, but my question would be how many of those 2 bundles of 3 can be at latest intel design integer instructions?

In that madison only 2 out of 6 could be integers. So it is executing code at a maximum efficiency of 33%. Or did i calculate that wrong? (maybe a bit better considering cache fetches, am not knowledgeable there)

So in that case from the core 66% is total wasted already if not more as it doesn't deal with integers?

Thanks,
Vincent
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Tukwila Update - article onlineDavid Kanter2009/02/05 12:03 AM
  Tukwila Update - article onlineDan2009/02/05 03:17 AM
    Tukwila Update - article onlineJoe Chang2009/02/05 09:16 AM
    Tukwila Update - article onlineTemp2009/02/05 09:25 AM
      Tukwila Update - article onlinePaul2009/02/05 12:29 PM
        Tukwila Update - article onlineDavid Kanter2009/02/05 06:32 PM
          Tukwila Update - article onlinePhil2009/02/06 01:24 AM
            Great. Finally hard numbersMichael S2009/02/06 04:46 AM
            Tukwila Update - article onlinelubemark2009/02/06 05:54 AM
              Tukwila Update - article onlinePhil2009/02/06 07:29 AM
                Tukwila Update - article onlineRagingDragon2009/02/07 03:39 PM
                  Tukwila Update - article onlineMichael S2009/02/07 04:09 PM
            Tukwila Update - article onlinesavantu2009/02/06 06:23 AM
              Tukwila Update - article onlineMichael S2009/02/06 07:13 AM
            Tukwila Update - article onlinesomeone2009/02/06 07:18 AM
              Tukwila Update - article onlinePhil2009/02/06 07:47 AM
                Tukwila Update - article onlinesomeone2009/02/06 08:17 AM
                Tukwila Update - article onlineRagingDragon2009/02/07 03:51 PM
              Tukwila Update - article onlineLinus Torvalds2009/02/06 08:37 AM
                Tukwila Update - article onlinesomeone2009/02/06 09:19 AM
                Tukwila Update - article onlinesavantu2009/02/06 10:19 AM
                  Tukwila Update - article onlineLinus Torvalds2009/02/06 10:40 AM
                    Tukwila Update - article onlinesavantu2009/02/06 11:00 AM
              Tukwila Update - article onlinePhil2009/02/09 04:54 AM
                Tukwila Update - article onlineDoug Siebert2009/02/09 10:40 PM
                  Tukwila Update - article onlineJouni Osmala2009/02/10 01:03 AM
                  Tukwila Update - article onlinesomeone2009/02/10 06:15 AM
                Tukwila Update - article onlineslacker2009/02/10 06:22 PM
    Tukwila Update - article onlineMichael S2009/02/05 03:56 PM
    Tukwila Update - article onlineDavid Kanter2009/02/05 04:55 PM
    Tukwila Update - article onlinesomeone2009/02/05 05:47 PM
      Tukwila Update - article onlineanon2009/02/05 10:16 PM
        Tukwila Update - article onlineRagingDragon2009/02/05 10:27 PM
        Tukwila Update - article onlinesomeone2009/02/06 07:32 AM
          Tukwila Update - article onlineanon2009/02/06 09:25 AM
            Tukwila Update - article onlinesomeone2009/02/06 09:40 AM
  POWER6 memory bandwidthMichael S2009/02/05 03:30 AM
    POWER6 memory bandwidthsomeone2009/02/05 07:00 AM
      POWER6 memory bandwidthMichael S2009/02/05 07:36 AM
      POWER6 interconnectconfused2009/02/05 10:50 AM
        POWER6 interconnectfoobar2009/02/05 02:12 PM
    POWER6 memory bandwidthWes Felter2009/02/05 12:57 PM
    POWER6 memory bandwidthJesper Frimann2009/02/09 11:54 PM
      POWER6 memory bandwidthMichael S2009/02/10 07:21 AM
  Why the platform focus?Linus Torvalds2009/02/05 08:40 AM
    Why the platform focus?savantu2009/02/05 08:50 AM
      Why the platform focus?Vincent Diepeveen2009/02/05 09:29 AM
        Why the platform focus?savantu2009/02/05 10:34 PM
          Why the platform focus?Vincent Diepeveen2009/02/05 11:09 PM
          Why the platform focus?Phil2009/02/06 01:10 AM
            Why the platform focus?savantu2009/02/06 01:50 AM
              Why the platform focus?Phil2009/02/06 07:09 AM
                Why the platform focus?savantu2009/02/06 10:08 AM
                Why the platform focus?someone2009/02/06 10:21 AM
                  Why the platform focus?mpx2009/02/06 02:04 PM
                    Why the platform focus?someone2009/02/06 02:16 PM
                Why the platform focus?RagingDragon2009/02/07 04:16 PM
                Why the platform focus?mas2009/02/25 08:28 AM
              itanium bigger than entire car industryVincent Diepeveen2009/02/06 07:12 AM
                itanium bigger than entire car industryDevon Welles2009/02/06 07:51 AM
                  itanium bigger than entire car industryVincent Diepeveen2009/02/06 10:41 AM
                  itanium bigger than entire car industryDean Kent2009/02/06 07:56 PM
                    Unit sales is meaningless when ASP grows fastersomeone2009/02/07 09:38 AM
                      Unit sales is meaningless when ASP grows fasterDean Kent2009/02/07 03:10 PM
                        Unit sales is meaningless when ASP grows fasterRagingDragon2009/02/07 04:34 PM
                    itanium bigger than entire car industryVincent Diepeveen2009/02/08 05:35 AM
                itanium bigger than entire car industryRagingDragon2009/02/07 04:40 PM
            Why the platform focus?Vincent Diepeveen2009/02/06 07:47 AM
      Yes it doesm performance mattersbob2009/02/05 10:51 AM
        Yes it doesm performance mattersVenki2009/02/05 11:06 AM
      Why the platform focus?Doug Siebert2009/02/06 01:07 AM
        Why the platform focus?savantu2009/02/06 02:00 AM
    Why the platform focus?someone2009/02/05 10:49 AM
      Why the platform focus?Linus Torvalds2009/02/05 12:03 PM
        Why the platform focus?Default2009/02/05 01:29 PM
          Why the platform focus?anon2009/02/05 02:08 PM
        Why the platform focus?someone2009/02/05 02:24 PM
          Why the platform focus?Linus Torvalds2009/02/05 03:30 PM
    Why the platform focus?Paradox2009/02/05 11:22 AM
    Why the platform focus?slacker2009/02/05 01:41 PM
      Why the platform focus?RagingDragon2009/02/05 10:57 PM
      Why the platform focus?Michael S2009/02/06 06:11 AM
        Why the platform focus?slacker2009/02/06 01:58 PM
          Why the platform focus?Michael S2009/02/08 02:24 AM
            Why the platform focus?someone2009/02/08 09:38 AM
              Why the platform focus?David Kanter2009/02/08 04:27 PM
                Why the platform focus?someone2009/02/08 07:26 PM
                  Why the platform focus?savantu2009/02/09 12:35 AM
          Why the platform focus?someone2009/02/08 09:53 AM
    All x86 SpecInt scores are useless due to autopar (NT)Michael S2009/02/05 03:15 PM
      Auto parallelizationDavid Kanter2009/02/05 06:17 PM
      All x86 SpecInt scores are useless due to autopar (NT)Paradox2009/02/06 08:47 AM
    Why the platform focus?David Kanter2009/02/05 04:49 PM
    Why the platform focus?David Kanter2009/02/06 01:09 AM
      Why the platform focus?Linus Torvalds2009/02/06 08:14 AM
        Why the platform focus?savantu2009/02/06 10:37 AM
          Why the platform focus?Linus Torvalds2009/02/06 12:49 PM
            Why the platform focus?Linus Torvalds2009/02/06 01:09 PM
            Intel puts its money where its mouth issomeone2009/02/06 02:08 PM
              Intel puts its money where its mouth isRagingDragon2009/02/07 06:01 PM
                Intel puts its money where its mouth issomeone2009/02/08 02:24 PM
                  mission-criticalMichael S2009/02/08 05:06 PM
                    mission-criticalmpx2009/02/09 02:30 AM
                      mission-criticalrwessel2009/02/09 03:23 PM
                    mission-criticalanon2009/02/09 03:55 AM
                    mission-criticalEduardoS2009/02/09 05:17 PM
                      mission-criticalDean Kent2009/02/09 08:11 PM
                        mission-criticalMichael S2009/02/10 05:20 AM
                          mission-criticalDean Kent2009/02/10 07:26 AM
                            mission-criticalMichael S2009/02/10 08:01 AM
                              mission-criticalDean Kent2009/02/10 01:36 PM
                            mission-criticalsomeone2009/02/10 09:05 AM
                              mission-criticalDean Kent2009/02/10 01:22 PM
                                mission-criticalZt2009/02/22 04:54 PM
                          mission-criticalanon2009/02/10 10:41 PM
                        mission-criticalEduardoS2009/02/10 01:46 PM
                          mission-criticalDean Kent2009/02/10 02:31 PM
                        mission-criticalslacker2009/02/10 07:30 PM
                    mission-criticalVincent Diepeveen2009/02/18 07:20 AM
                  Mission criticalmpx2009/02/09 01:00 AM
            Why the platform focus?savantu2009/02/07 01:15 AM
              Sun and x86 server differentiationDavid Kanter2009/02/07 01:34 AM
                Sun and x86 server differentiationmax2009/02/07 03:30 AM
                Sun and x86 server differentiationsomeone2009/02/07 10:19 AM
                  Sun and x86 server differentiationLinus Torvalds2009/02/07 10:44 AM
                Sun and x86 server differentiationRagingDragon2009/02/07 06:09 PM
                  Sun and x86 server differentiationMichael S2009/02/08 05:05 AM
                    Sun and x86 server differentiationRagingDragon2009/02/10 12:03 AM
                    Sun and x86 server differentiationJesper Frimann2009/02/10 12:51 AM
                      Sun and x86 server differentiationAlex Jones2009/02/10 01:43 PM
            Why the platform focus?bob2009/02/08 04:51 AM
              Why the platform focus?someone2009/02/08 09:23 AM
          missing the big pictureAM2009/02/18 06:43 AM
            missing the big pictureMichael S2009/02/18 08:42 AM
              missing the big pictureAM2009/02/18 09:03 AM
        Why the platform focus?mpx2009/02/06 12:47 PM
          Itanium - slowest and most obsolete server CPU family in the world, NOW.mpx2009/02/06 04:48 PM
            Itanium - slowest and most obsolete server CPU family in the world, NOW.Paul2009/02/07 02:56 PM
              z series?Michael S2009/02/07 03:12 PM
              Itanium - slowest and most obsolete server CPU family in the world, NOW.someone else2009/02/24 04:37 AM
                Itanium - slowest and most obsolete server CPU family in the world, NOW.EduardoS2009/02/24 06:55 AM
                  Itanium - slowest and most obsolete server CPU family in the world, NOW.someone else2009/02/25 01:55 AM
                    Itanium - slowest and most obsolete server CPU family in the world, NOW.Michael S2009/02/25 02:27 AM
          Why the platform focus?RagingDragon2009/02/07 06:18 PM
            Why the platform focus?Paul2009/02/08 01:10 PM
              Why the platform focus?Jukka Larja2009/02/08 11:04 PM
        Why the platform focus?slacker2009/02/06 02:10 PM
          Why the platform focus?Linus Torvalds2009/02/06 02:40 PM
            Why the platform focus?savantu2009/02/06 02:51 PM
            Why the platform focus?someone2009/02/06 02:58 PM
              Why the platform focus?Linus Torvalds2009/02/07 09:26 AM
                Why the platform focus?someone2009/02/07 10:10 AM
                  Why the platform focus?Linus Torvalds2009/02/07 10:40 AM
                    Why the platform focus?someone2009/02/07 12:24 PM
                      Why the platform focus?Doug Siebert2009/02/08 12:32 AM
                        Why the platform focus?max2009/02/08 04:57 AM
                          Why the platform focus?Michael S2009/02/08 05:20 AM
                        Why the platform focus?someone2009/02/08 09:15 AM
                          Why the platform focus?Doug Siebert2009/02/08 11:36 PM
                      Why the platform focus?hobold2009/02/09 05:49 AM
                Why the platform focus?anon2009/02/24 01:57 AM
                  Why the platform focus?Linus Torvalds2009/02/24 09:45 AM
                    Why the platform focus?savantu2009/02/24 12:30 PM
                      Why the platform focus?slacker2009/02/24 01:51 PM
                        Why the platform focus?savantu2009/02/25 12:04 AM
                          Why the platform focus?Michael S2009/02/25 02:34 AM
                          Why the platform focus?anon2009/02/25 10:17 AM
                      Why the platform focus?max2009/02/25 11:15 AM
                    Why the platform focus?someone2009/02/24 05:43 PM
                      Why the platform focus?Doug Siebert2009/02/24 08:26 PM
                        Why the platform focus?Howard Chu2009/02/25 03:07 AM
                          Why the platform focus?someone2009/02/25 06:48 AM
                        Why the platform focus?someone2009/02/25 06:41 AM
                          Why the platform focus?Linus Torvalds2009/02/25 09:17 AM
                            Why the platform focus?someone2009/02/25 09:55 AM
                              has anyone seen Tukwila silicon?anon2009/02/25 10:38 AM
                              Why the platform focus?Linus Torvalds2009/02/25 11:05 AM
                                Why the platform focus?slacker2009/02/25 01:11 PM
                                Why the platform focus?a reader2009/02/26 09:11 PM
                                  Why the platform focus?rcf2009/02/27 01:32 PM
                                    Why the platform focus?max2009/02/27 02:11 PM
                                      Why the platform focus?rcf2009/02/27 03:50 PM
                            Why the platform focus?Vincent Diepeveen2009/02/25 04:30 PM
                            $40M sale to $16M companybob2009/02/25 08:25 PM
                              $40M sale to $16M companyRichard Cownie2009/02/26 12:21 PM
                  Why the platform focus?anonymous2009/02/24 11:52 AM
                    Why the platform focus?savantu2009/02/24 12:20 PM
                      Why the platform focus?anonymous2009/02/24 03:31 PM
                        Why the platform focus?savantu2009/02/25 12:05 AM
                        Why the platform focus?someone else2009/02/25 01:04 AM
                          Why the platform focus?Michael S2009/02/25 01:42 AM
                            Put me down for $500 that Poulson doesn't arrive earlier than Q4/2011 (NT)slacker2009/02/25 12:39 PM
                        Why the platform focus?someone2009/02/25 06:54 AM
                          Why the platform focus?anonymous2009/02/25 09:46 AM
                            Why the platform focus?someone2009/02/25 10:22 AM
                            Why the platform focus?anon2009/02/25 11:01 AM
                              Why the platform focus?anonymous2009/02/25 11:54 AM
                  Why the platform focus?mpx2009/02/24 02:11 PM
                    Why the platform focus?anon2009/02/24 08:57 PM
                      Why the platform focus?Doug Siebert2009/02/24 10:04 PM
                        Why the platform focus?anon2009/02/24 10:46 PM
                          Why the platform focus?Doug Siebert2009/02/25 05:13 PM
                            Why the platform focus?anon2009/02/25 08:53 PM
                              Why the platform focus?bob2009/02/25 09:00 PM
                                Please try to keep up (NT)anon2009/02/25 09:49 PM
                              Why the platform focus?Doug Siebert2009/02/26 12:09 AM
                                Why the platform focus?anon2009/02/26 01:12 AM
                                Why the platform focus?Michael S2009/02/26 02:16 AM
                                  Why the platform focus?James2009/02/26 06:09 AM
                                    sufficiently intimate with the OSMichael S2009/02/26 06:29 AM
                                      sufficiently intimate with the OSanon2009/02/27 01:01 AM
                                      sufficiently intimate with the OSHoward Chu2009/02/27 01:37 AM
                      Why the platform focus?Michael S2009/02/25 02:02 AM
                        Why the platform focus?anon2009/02/25 03:07 AM
              Why the platform focus?anon2009/02/07 01:18 PM
                Why the platform focus?Vincent Diepeveen2009/02/08 10:16 AM
                Why the platform focus?anon2009/02/25 07:40 AM
                  Intels financial statusVincent Diepeveen2009/02/25 12:02 PM
                Why the platform focus?someone2009/02/25 07:54 AM
      Why the platform focus?Vincent Diepeveen2009/02/06 08:20 AM
        Why the platform focus?Default2009/02/06 09:57 AM
          Why the platform focus?Vincent Diepeveen2009/02/06 10:59 AM
          Why the platform focus?RagingDragon2009/02/07 06:43 PM
  Tukwila Update - article onlineVincent Diepeveen2009/02/05 09:11 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?