Multi-core vs. Single thread performance

By: IntelUser2000 (Intel_user2000.delete@this.yahoo.ca), November 10, 2008 6:34 am
Room: Moderated Discussions
It seems everyone is around equally split on having a fast single core vs. multiple slower cores.

How many people want to continue seeing single core only development?? Say a Core 2 at 5GHz.

How many people want to see the current development?? Like now, the Core 2 Quad at 3GHz.


The argument for fast single core is that unoptimized, the programs run faster. But we are really starting to reach the limit on single thread performance. If the Pentium 4 went so far to reach such high clock speeds, do you really think it'll be possible to have a much higher clocked single core CPU?? We know that power isn't the only limit for clock speed.

Likely, we won't see single core variants that might or might not improve clock speeds.

Same with cache. Caches have diminishing returns. Doubling caches from 2MB to 4MB brings at best 10% performance.

The argument for multiple cores is that the potential is far greater with optimization. Well threaded application on a 3GHz 4 core CPU is far faster than a 5GHz single core(not to mention "5GHz" is reachable is up in the air).

You guys say people care about single cores in office. A well-designed SSD will offer significantly better improvements than miniscule CPU improvements will.

I like what glimpses Intel gave of the future CPU. At Haswell, we'll see the pinnacle of large multiple core CPUs with 8-10 cores. By then the software world should be better suited to multi-threaded.

The step after Haswell is probably heterogenous cores with few big cores and many small cores. There will be a transition period I assume. We won't suddenly go from 8-10 large cores to 1 huge core + 30 mini cores.

The transition CPU will probably has 2-3 large core with 10 small cores so it'll satisfy single threaded apps that is really hard to scale(hence the large cores), multi-threaded apps that scales moderately needing a balance of multi and single thread performance(hence the 2-3 large cores), and multi-threaded apps that scales really really well(hence the many small cores).

They aren't stupid and the multi-cores won't be just multiple cores just dumbed together. There will probably be better core to core interconnects, more bandwidth better synchronization to mitigate the effect of diminishing returns. Sandy Bridge is said to focus on core interconnects.

They'll slowly find a way.
 Next Post in Thread >
TopicPosted ByDate
Multi-core vs. Single thread performanceIntelUser20002008/11/10 06:34 AM
  Multi-core vs. Single thread performanceLinus Torvalds2008/11/10 08:25 AM
    Multi-core vs. Single thread performanceanon2008/11/10 09:08 AM
      Multi-core vs. Single thread performanceJasonB2008/11/10 09:50 AM
        Multi-core vs. Single thread performanceIntelUser20002008/11/10 06:54 PM
          Multi-core vs. Single thread performanceJasonB2008/11/10 09:17 PM
            What does your app do?David Kanter2008/11/11 12:20 AM
              What does your app do?JasonB2008/11/11 01:05 AM
        Multi-core vs. Single thread performanceRagingDragon2008/11/14 11:02 PM
          Multi-core vs. Single thread performanceJasonB2008/11/15 12:12 AM
            Multi-core vs. Single thread performanceRagingDragon2008/11/15 12:37 PM
              Multi-core vs. Single thread performanceJasonB2008/11/15 07:03 PM
      Multi-core vs. Single thread performanceHenrik S2008/11/10 09:59 AM
      Multi-core vs. Single thread performancePaul2008/11/10 10:06 AM
        Multi-core vs. Single thread performanceslacker2008/11/10 11:08 PM
          Multi-core vs. Single thread performanceanonymous2008/11/10 11:44 PM
          Multi-core vs. Single thread performancePaul2008/11/11 01:11 PM
            Red AnnouncementsPaul2008/11/13 12:52 PM
            Multi-core vs. Single thread performanceslacker2008/11/14 02:33 AM
    Multi-core vs. Single thread performanceanon2008/11/10 09:10 AM
      Bob Colwell ?anon2008/11/11 12:04 AM
      Multi-core vs. Single thread performanceRagingDragon2008/11/14 08:56 PM
    Multi-core vs. Single thread performanceIntelUser20002008/11/10 07:00 PM
      Multi-core vs. Single thread performanceLinus Torvalds2008/11/11 01:23 PM
        Multi-core vs. Single thread performanceDoug Siebert2008/11/11 09:27 PM
          Multi-core vs. Single thread performanceLinus Torvalds2008/11/12 12:49 PM
            Multi-core vs. Single thread performanceDoug Siebert2008/11/12 07:16 PM
            Multi-core vs. Single thread performanceGabriele Svelto2008/11/12 11:49 PM
            Multi-core vs. Single thread performanceJasonB2008/11/13 01:00 AM
              Multi-core vs. Single thread performancegolfbum2008/11/13 07:30 PM
              Multi-core vs. Single thread performanceMatt Craighead2008/11/17 04:08 PM
                Multi-core vs. Single thread performanceLinus Torvalds2008/11/17 06:54 PM
                  Multi-core vs. Single thread performancea reader2008/11/17 08:55 PM
                    Multi-core vs. Single thread performanceJasonB2008/11/17 09:12 PM
                      Multi-core vs. Single thread performanceMatt Craighead2008/11/17 10:36 PM
                        Multi-core vs. Single thread performanceJasonB2008/11/18 01:46 AM
                          Multi-core vs. Single thread performanceIan Ameline2008/11/18 11:08 AM
                            Sounds like extreme use casesDavid Kanter2008/11/18 12:20 PM
                              Sounds like extreme use casesMatt Craighead2008/11/18 12:48 PM
                                SLI/XFire statsMatt Sayler2008/11/19 01:17 PM
                              Sounds like extreme use casesanon2008/11/18 06:38 PM
                              Sounds like extreme use casesJasonB2008/11/18 11:19 PM
                          Multi-core vs. Single thread performanceMatt Craighead2008/11/18 12:33 PM
                            Multi-core vs. Single thread performanceJasonB2008/11/18 10:34 PM
                              Multi-core vs. Single thread performanceAnonymous2008/11/19 12:01 PM
                                AgreedJasonB2008/11/19 05:07 PM
                                Multi-core vs. Single thread performanceanon2008/11/19 09:56 PM
                  Multi-core vs. Single thread performanceMatt Craighead2008/11/17 09:37 PM
                    Multi-core vs. Single thread performanceGabriele Svelto2008/11/18 05:05 AM
                      Multi-core vs. Single thread performanceAnonymouse2008/11/18 11:15 AM
                        Multi-core vs. Single thread performanceMatt Craighead2008/11/18 12:25 PM
                          Multi-core vs. Single thread performancepgerassi2008/11/18 06:54 PM
                            GPUs and VMDavid Kanter2008/11/18 11:15 PM
                          Multi-core vs. Single thread performanceJasonB2008/11/19 01:03 AM
                Multi-core vs. Single thread performanceJasonB2008/11/17 08:58 PM
                  Multi-core vs. Single thread performanceMatt Craighead2008/11/17 10:11 PM
                    Multi-core vs. Single thread performanceJasonB2008/11/18 12:33 AM
                      Multi-core vs. Single thread performanceMatt Craighead2008/11/18 01:18 AM
                        Multi-core vs. Single thread performanceJasonB2008/11/18 02:42 AM
                          Multi-core vs. Single thread performancepgerassi2008/11/18 07:51 PM
                            Multi-core vs. Single thread performanceUngo2008/11/19 05:14 PM
        idiots and performanceCarlie Coats2008/11/12 05:58 AM
          most are to dumb for performance?Moritz2008/11/13 11:11 AM
            most are to dumb for performance?Carlie Coats2008/11/13 12:00 PM
    Multi-core vs. Single thread performanceAnonymous2008/11/10 08:40 PM
      Multi-core vs. Single thread performanceJasonB2008/11/11 12:11 AM
        Multi-core vs. Single thread performanceJS2008/11/11 03:59 AM
          Multi-core vs. Single thread performanceIan Ollmann2008/11/14 06:20 PM
      Multi-core vs. Single thread performanceJesper Frimann2008/11/11 04:27 AM
        Multi-core vs. Single thread performanceChris2008/11/11 05:55 AM
          Multi-core vs. Single thread performanceMichael S2008/11/11 06:19 AM
            Multi-core vs. Single thread performanceChris2008/11/11 06:40 AM
              Multi-core vs. Single thread performanceHoward Chu2008/11/11 02:53 PM
    Ahmdahls law works for hardware also.Jouni Osmala2008/11/11 03:35 AM
      Ahmdahls law works for hardware also.Gabriele Svelto2008/11/11 05:54 AM
        Exactly (NT)Michael S2008/11/11 06:22 AM
        And so Svelto's law was born (NT)Rob Thorpe2008/11/11 12:11 PM
        Quad core CPUs are nowhere near power limitsDoug Siebert2008/11/11 10:22 PM
          Quad core CPUs are nowhere near power limitsanonymous2008/11/12 02:04 AM
            Quad core CPUs are nowhere near power limitsDoug Siebert2008/11/12 07:18 PM
    Multi-core vs. Single thread performanceMoritz2008/11/13 09:01 AM
      Multi-core vs. Single thread performanceJasonB2008/11/13 08:05 PM
      What are we going to do with all that silicon? (Paging Groo)James2008/11/14 05:47 AM
        What are we going to do with all that silicon? (Paging Groo)Groo2008/11/14 08:47 AM
    Multi-core vs. Single thread performanceIan Ollmann2008/11/14 05:38 PM
      Multi-core vs. Single thread performanceJasonB2008/11/14 08:59 PM
        Multi-core vs. Single thread performanceslacker2008/11/15 12:28 AM
          Multi-core vs. Single thread performanceJasonB2008/11/15 04:11 AM
        Multi-core vs. Single thread performanceLinus Torvalds2008/11/15 12:50 PM
          Multi-core vs. Single thread performanceRagingDragon2008/11/15 01:41 PM
            Multi-core vs. Single thread performanceJasonB2008/11/15 06:49 PM
              Multi-core vs. Single thread performanceAnonymous2008/11/16 12:36 AM
                Multi-core vs. Single thread performanceJasonB2008/11/16 03:08 AM
                Anonymous what became of your VLIW-based x86 startup company ?anon2008/11/16 09:18 AM
            Multi-core vs. Single thread performanceAnonymous2008/11/16 12:28 AM
              Multi-core vs. Single thread performanceRagingDragon2008/11/16 02:13 AM
                Floating point - a tool of the devilDavid Kanter2008/11/16 02:34 AM
                  Floating point - a tool of the devilFoo_2008/11/16 05:22 AM
                    Floating point - a tool of the devilMatt Sayler2008/11/16 08:04 AM
                      Floating point - a tool of the devilAnonymous2008/11/16 12:14 PM
                      Floating point - a tool of the devilFoo_2008/11/17 01:27 AM
                      Floating point - a tool of the devilFoo_2008/11/17 01:28 AM
                        Floating point - a tool of the devilHoward Chu2008/11/17 02:23 PM
            Multi-core vs. Single thread performanceAaron Spink2008/11/16 06:19 AM
  Multi-core vs. Single thread performancehobold2008/11/10 09:23 AM
  A third positionslacker2008/11/10 10:43 PM
    A third positionJasonB2008/11/11 12:56 AM
      A third positionpgerassi2008/11/11 09:07 AM
      A third positionRagingDragon2008/11/14 10:40 PM
    A third positionJukka Larja2008/11/11 05:26 AM
    Power 6Joe2008/11/11 07:13 AM
      Power 6slacker2008/11/11 07:48 AM
    A third positionWes Felter2008/11/11 12:31 PM
    A second position on physicsMoritz2008/11/13 09:10 AM
      A second position on physicsRagingDragon2008/11/14 10:53 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell purple?