Latency and HPC Workloads

Article: Intel's Near-Threshold Voltage Computing and Applications
By: forestlaughing (forestlaughing.delete@this.yahoo.com), October 15, 2012 7:41 am
Room: Moderated Discussions
Michael S (already5chosen.delete@this.yahoo.com) on October 8, 2012 5:12 pm wrote:
> > Sure, but i was talking about the
> > higher end
> of the performance scale, not commodity systems. Something like the
> > big
> POWER systems,
>
> Big POWER systems use fully-buffered memories. So, latency
> wouldn't be as good as commodity boxen regardless of interface you use between
> buffer and memory device.
> Besides, I don't know we they do the scheduling, in
> the controller, i.e in Power chip or in the buffer itself. If the former, then
> you face the same problem as with Intel/AMD.

The POWER solution to this problem is to put an enormous L3 cache on the processor. 80MB / 8-cores in power7+. This is essentially the place we are at with low latency memories. It is cost prohibitive to build an entire system memory from high cost / low latency memory. Also, as capacity goes up, that latency becomes less low. The alternative is to have a small low-latency memory paired with a larger memory to fill that cache.

IBM uses that small/fast memory for hardware controlled cache, on the power7. So too have the most recent generations of vector machines from cray and NEC. IBM cell, though a little long in the tooth now, was cool in that it gave the programmer control over that small/fast memory. This impacts bandwidth too. As core counts go up, it's not possible to put enough pins on a chip, to maintain the bandwidth/core that we're used to. The solution is to put more of that memory on package, or even on-chip. I think we're going to see another level in the memory hierarchy, for bandwidth reasons, if not latency. Hopefully latency comes along for the ride.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
New article: Intel's Near-Threshold ComputingDavid Kanter2012/09/18 11:26 AM
  Higher SRAM voltage and shared L1Paul A. Clayton2012/09/18 01:38 PM
    Higher SRAM voltage and shared L1David Kanter2012/09/18 04:20 PM
      Higher SRAM voltage and shared L1Eric2012/09/20 09:44 AM
        Higher SRAM voltage and shared L1David Kanter2012/09/20 11:24 AM
      Yes, that kind of asynchronousPaul A. Clayton2012/09/20 01:53 PM
    Higher SRAM voltage and shared L1somebody2012/09/19 08:27 AM
      So micro-turboboost is doubly impracticalPaul A. Clayton2012/09/20 01:53 PM
  Big littleDoug S2012/09/18 02:04 PM
    Big littleDavid Kanter2012/09/18 03:05 PM
    Big littleRicardo B2012/09/19 03:06 AM
  New article: Intel's Near-Threshold Computingdefderdar2012/09/18 08:39 PM
    New article: Intel's Near-Threshold Computingtarlinian2012/09/19 07:32 AM
      New article: Intel's Near-Threshold ComputingDavid Kanter2012/09/19 09:44 AM
  New article: Intel's Near-Threshold ComputingMark Christiansen2012/09/19 10:31 AM
    New article: Intel's Near-Threshold ComputingChris Brodersen2012/09/19 11:54 AM
  New article: Intel's Near-Threshold ComputingEric2012/09/20 09:47 AM
  Latency and HPC WorkloadsRobert Myers2012/10/03 09:52 AM
    Latency and HPC Workloadsanon2012/10/03 05:50 PM
      Latency and HPC WorkloadsRobert Myers2012/10/04 09:24 AM
        Latency and HPC WorkloadsSHK2012/10/08 04:42 AM
          Latency and HPC WorkloadsMichael S2012/10/08 12:59 PM
            Latency and HPC WorkloadsSHK2012/10/08 01:42 PM
              Latency and HPC WorkloadsMichael S2012/10/08 04:12 PM
                Latency and HPC Workloadsforestlaughing2012/10/15 07:41 AM
                  The original context was Micron RLDRAM (NT)Michael S2012/10/15 07:55 AM
                    The original context was Micron RLDRAMforestlaughing2012/10/15 09:21 AM
              Latency and HPC Workloads - Why not SRAM?Kevin G2012/10/09 08:48 AM
                Latency and HPC Workloads - Why not SRAM?Michael S2012/10/09 09:33 AM
                  Latency and HPC Workloads - Why not SRAM?SHK2012/10/09 11:55 AM
                    Why not SRAM? - CapacityRohit2012/10/09 08:13 PM
                  Latency and HPC Workloads - Why not SRAM?Kevin G2012/10/09 02:04 PM
                    Latency and HPC Workloads - Why not SRAM?Michael S2012/10/09 03:52 PM
                      Latency and HPC Workloads - Why not SRAM?Robert Myers2012/10/10 09:11 AM
                        Latency and HPC Workloads - Why not SRAM?forestlaughing2012/10/15 07:02 AM
                          Latency and HPC Workloads - Why not SRAM?Robert Myers2012/10/15 08:04 AM
                            Latency and HPC Workloads - Why not SRAM?forestlaughing2012/10/16 08:13 AM
                          Latency and HPC Workloads - Why not SRAM?SHK2012/10/16 07:12 AM
                    Latency and HPC Workloads - Why not SRAM?slacker2012/10/11 12:35 PM
                      SRAM leakageDavid Kanter2012/10/11 02:00 PM
          Latency and HPC Workloadsforestlaughing2012/10/15 07:57 AM
            Latency and HPC WorkloadsRobert Myers2012/10/16 06:28 AM
              Latency and HPC WorkloadsMichael S2012/10/16 06:35 AM
              Latency and HPC Workloadsanon2012/10/16 07:17 AM
                Latency and HPC WorkloadsRobert Myers2012/10/16 08:56 AM
                  Supercomputer variant of Kahan quotePaul A. Clayton2012/10/16 10:09 AM
                    Supercomputer variant of Kahan quoteanon2012/10/17 12:17 AM
                      Supercomputer variant of Kahan quoteRobert Myers2012/10/17 03:34 AM
                        Supercomputer variant of Kahan quoteanon2012/10/17 04:12 AM
                          Supercomputer variant of Kahan quoteRobert Myers2012/10/17 01:38 PM
                            Supercomputer variant of Kahan quoteanon2012/10/17 04:24 PM
                              Supercomputer variant of Kahan quoteRobert Myers2012/10/17 04:45 PM
                                Supercomputer variant of Kahan quoteanon2012/10/17 04:58 PM
                                Supercomputer variant of Kahan quoteanon2012/10/17 04:58 PM
                                  Supercomputer variant of Kahan quoteRobert Myers2012/10/17 06:14 PM
                                    Supercomputer variant of Kahan quoteanon2012/10/17 07:36 PM
                                      Supercomputer variant of Kahan quoteRobert Myers2012/10/18 08:47 AM
                                        Supercomputer variant of Kahan quoteanon2012/10/19 01:34 AM
                                          Supercomputer variant of Kahan quoteanon2012/10/19 03:47 AM
                                          Supercomputer variant of Kahan quoteRobert Myers2012/10/19 02:14 PM
                        Supercomputer variant of Kahan quoteMichael S2012/10/17 05:56 PM
                          Supercomputer variant of Kahan quoteanon2012/10/17 08:02 PM
                            Supercomputer variant of Kahan quoteRobert Myers2012/10/18 12:29 PM
                              Supercomputer variant of Kahan quoteanon2012/10/19 01:27 AM
                                Supercomputer variant of Kahan quoteRobert Myers2012/10/19 06:24 AM
                                  Supercomputer variant of Kahan quoteanon2012/10/19 07:00 AM
                                    Supercomputer variant of Kahan quoteRobert Myers2012/10/19 08:28 AM
                                      Supercomputer variant of Kahan quoteanon2012/10/19 09:27 AM
                              Supercomputer variant of Kahan quoteforestlaughing2012/10/19 09:26 AM
                                Supercomputer variant of Kahan quoteRobert Myers2012/10/19 06:04 PM
                                  Supercomputer variant of Kahan quoteEmil Briggs2012/10/20 03:52 AM
                                    Supercomputer variant of Kahan quoteRobert Myers2012/10/20 06:51 AM
                                      Supercomputer variant of Kahan quoteEmil Briggs2012/10/20 07:33 AM
                                        Supercomputer variant of Kahan quoteEmil Briggs2012/10/20 07:34 AM
                                          Supercomputer variant of Kahan quoteRobert Myers2012/10/20 08:35 AM
                                            Supercomputer variant of Kahan quoteEmil Briggs2012/10/20 09:04 AM
                                              Supercomputer variant of Kahan quoteRobert Myers2012/10/20 10:23 AM
                  Latency and HPC Workloadsanon2012/10/16 05:48 PM
                    Latency and HPC Workloadsforestlaughing2012/10/19 10:43 AM
              Latency and HPC Workloadsforestlaughing2012/10/19 08:38 AM
                Latency and HPC WorkloadsRobert Myers2012/10/19 10:40 AM
                Potential false economics in researchPaul A. Clayton2012/10/19 11:54 AM
                  Potential false economics in researchVincent Diepeveen2012/10/20 07:59 AM
                  Potential false economics in researchforestlaughing2012/10/23 09:56 AM
                    Potential false economics in researchRobert Myers2012/10/23 06:16 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?