Some 'reverse-engineering' of Zen

By: dmcq (dmcq.delete@this.fano.co.uk), October 31, 2015 8:19 am
Room: Moderated Discussions
bakaneko (nyan.delete@this.hyan.wan) on October 31, 2015 7:28 am wrote:
> dmcq (dmcq.delete@this.fano.co.uk) on October 30, 2015 5:12 am wrote:
> > lurker (lurker9000.delete@this.realemail.mail) on October 30, 2015 2:39 am wrote:
> > > > First of all - welcome to RWT, glad to hear your perspective.
> > >
> > > Thanks.
> > >
> > > > My guess is that everything you say is true...
> > >
> > > Eh, I just thought I'd post what I heard from a guy who supposedly worked on Zen.
> > >
> > > > and that AMD isn't intending to hit the HPC
> > > > market. They have 128b vectors (since that's all ARM supports), which simply isn't wide
> > > > enough to be competitive with Skylake. So giving up on a third AGU makes sense. The third
> > > > AGU is probably most helpful for HPC (where they cannot compete anyway) and isn't a particularly
> > > > small unit in terms of design complexity and impact on the load/store buffer.
> > > >
> > > > David
> > >
> > > 128bit FP pipes seem optimal for most desktop and server software. HPC is pretty
> > > much the only place where latest instructions are used and even if zen was competitive
> > > here I don't think anyone would want to switch from Intel.
> > > Personally I just hope the lack of 3rd AGU won't cause problems in SMT. I don't think
> > > normal workloads have that many operations that access memory, but SMT aims to maximize
> > > utilization of all available resources and only 2 AGUs might be a problem there.
> >
> > I think 256 bits would be better as you can do four double precision operations at once and that is quite
> > common. On the other hand with four SIMD units instead one could merge two operations to give an effective
> > two by 256 bit units except for some special operations. For anything larger they'd probably be better
> > off relying on GPUs I think if they can get the coherence and message passing working well. I can see
> > how to save larger register sets without impacting interrupt handling too badly but it seems a lot of
> > work when ARM is probably hoping to move 64 bit ARM into the embedded processor market.
>
> Except nobody sane would work on large amounts of
> doubles for most normal applications. It's really
> only useful for HPC, and there some other things
> probably matter even more, as floating point can be
> the wrong hammer.

Well I know games often just use floats in the GPUs an some AI people say 8-bit integers are enough for any useful AI problem - but it is amazing how fast a sequence of float operations can start to give obviously wrong results. If one wants half a chance of something approximating a reasonable result and aren't an expert at error analysis there's nothing to beat just doing the work using doubles.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Some 'reverse-engineering' of Zenjuanrga2015/10/03 04:48 AM
  Some 'reverse-engineering' of ZenHeikki Kultala2015/10/03 08:01 AM
    Some 'reverse-engineering' of Zenitsmydamnation2015/10/03 03:47 PM
      Some 'reverse-engineering' of ZenAnders2015/10/03 11:04 PM
        Some 'reverse-engineering' of ZenContrarian2015/10/04 11:53 AM
          Some 'reverse-engineering' of ZenJukka Larja2015/10/05 09:15 AM
            Some 'reverse-engineering' of Zenmpx2015/10/05 09:37 AM
              Some 'reverse-engineering' of ZenContrarian2015/10/25 12:25 PM
                Some 'reverse-engineering' of Zenitsmydamnation2015/10/26 01:49 AM
                Some 'reverse-engineering' of Zenjuanrga2015/10/29 02:16 AM
                  Some 'reverse-engineering' of Zenanon2015/10/30 06:54 AM
                    Some 'reverse-engineering' of Zenjuanrga2015/10/30 10:37 AM
                    POWER8 load/store unitsGabriele Svelto2015/10/30 06:34 PM
                Some 'reverse-engineering' of Zenlurker2015/10/29 03:12 PM
                  Some 'reverse-engineering' of ZenDavid Kanter2015/10/30 12:06 AM
                    Some 'reverse-engineering' of Zenlurker2015/10/30 02:39 AM
                      Some 'reverse-engineering' of ZenGabriele Svelto2015/10/30 03:14 AM
                        Some 'reverse-engineering' of Zenlurker2015/10/30 03:41 AM
                      Some 'reverse-engineering' of Zendmcq2015/10/30 05:12 AM
                        Some 'reverse-engineering' of Zenlurker2015/10/30 05:56 AM
                          Some 'reverse-engineering' of Zendmcq2015/10/30 06:09 AM
                            Some 'reverse-engineering' of Zennone2015/10/30 06:16 AM
                          Some 'reverse-engineering' of ZenDavid Kanter2015/10/30 08:16 AM
                            Some 'reverse-engineering' of Zenlurker2015/10/30 08:44 AM
                          Some 'reverse-engineering' of Zenjuanrga2015/10/30 01:00 PM
                            AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)Heikki Kultala2015/10/30 03:45 PM
                              AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)juanrga2015/10/31 06:20 AM
                                AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)Heikki Kultala2015/10/31 02:19 PM
                                  AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)juanrga2015/11/01 07:41 AM
                                    AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)Anne O. Nymous2015/11/01 09:39 AM
                                AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)Matthias Waldhauer2015/11/02 01:26 PM
                                  AVX and 128-bit FPUs in Zen(Re:Some 'reverse-engineering' of Zen)juanrga2015/11/03 03:42 AM
                            Some 'reverse-engineering' of ZenMatthias Waldhauer2015/11/02 11:39 AM
                        Some 'reverse-engineering' of Zenbakaneko2015/10/31 07:28 AM
                          Some 'reverse-engineering' of Zendmcq2015/10/31 08:19 AM
                            Some 'reverse-engineering' of ZenJukka Larja2015/10/31 09:21 AM
                            Some 'reverse-engineering' of Zenbakaneko2015/10/31 10:23 AM
                              Some 'reverse-engineering' of Zendmcq2015/10/31 04:12 PM
                                Some 'reverse-engineering' of Zenbakaneko2015/10/31 05:25 PM
                                  Some 'reverse-engineering' of Zendmcq2015/11/01 07:36 AM
                                    Some 'reverse-engineering' of Zenbakaneko2015/11/01 10:11 AM
                                      Some 'reverse-engineering' of Zendmcq2015/11/01 10:27 AM
                                        Some 'reverse-engineering' of Zenbakaneko2015/11/01 03:35 PM
                                          Some 'reverse-engineering' of Zendmcq2015/11/01 04:52 PM
                                            Some 'reverse-engineering' of Zenbakaneko2015/11/03 03:17 AM
                                              Some 'reverse-engineering' of Zendmcq2015/11/03 04:17 AM
                  Some 'reverse-engineering' of Zenjuanrga2015/10/30 12:43 PM
                    Some 'reverse-engineering' of Zenlurker2015/10/30 02:09 PM
                      Some 'reverse-engineering' of Zenjuanrga2015/10/31 06:58 AM
                        Some 'reverse-engineering' of Zenlurker2015/10/31 08:07 AM
                          Some 'reverse-engineering' of Zenjuanrga2015/10/31 12:50 PM
                            Some 'reverse-engineering' of ZenPoindexter2015/10/31 02:47 PM
                              Some 'reverse-engineering' of Zenlurker2015/10/31 04:06 PM
                                Some 'reverse-engineering' of ZenPoindexter2015/10/31 05:37 PM
                                  Some 'reverse-engineering' of Zenlurker2015/11/01 03:46 AM
                                  Some 'reverse-engineering' of Zenjuanrga2015/11/01 08:16 AM
                                    Some 'reverse-engineering' of ZenMaynard Handley2015/11/01 06:33 PM
                                      Some 'reverse-engineering' of Zenjuanrga2015/11/02 05:06 AM
                                        Zen transistor countHeikki Kultala2015/11/04 01:30 AM
                                          Zen transistor countjuanrga2015/11/05 05:34 AM
                                    Some 'reverse-engineering' of ZenSymmetry2015/11/02 06:56 AM
                                      Some 'reverse-engineering' of ZenDavid Hess2015/11/02 07:16 AM
                                      Some 'reverse-engineering' of Zennobody2015/11/02 08:19 AM
                                        Some 'reverse-engineering' of ZenJukka Larja2015/11/02 09:34 PM
                                          Some 'reverse-engineering' of Zennobody2015/11/03 01:35 AM
                                            Some 'reverse-engineering' of ZenJukka Larja2015/11/03 02:41 AM
                                              Some 'reverse-engineering' of Zennobody2015/11/03 05:10 AM
                                                Some 'reverse-engineering' of ZenJukka Larja2015/11/04 02:52 AM
                                            Some 'reverse-engineering' of ZenKlimax2015/11/03 02:47 AM
                                              Some 'reverse-engineering' of Zennobody2015/11/03 04:19 AM
                                                Some 'reverse-engineering' of ZenKlimax2015/11/03 12:09 PM
                                          Some 'reverse-engineering' of ZenBanana_Comedown2015/11/04 11:17 AM
                                            Some 'reverse-engineering' of ZenJukka Larja2015/11/05 03:51 AM
                                              Some 'reverse-engineering' of ZenBanana_Comedown2015/11/05 01:23 PM
                                      Some 'reverse-engineering' of Zenjuanrga2015/11/03 04:13 AM
                                Some 'reverse-engineering' of Zennobody2015/10/31 08:16 PM
                                  Some 'reverse-engineering' of Zenlurker2015/11/01 03:48 AM
                                    Some 'reverse-engineering' of Zensylt2015/11/01 04:07 AM
                                    Some 'reverse-engineering' of Zennobody2015/11/01 11:49 AM
                                  Some 'reverse-engineering' of Zenmd2015/11/01 01:06 PM
                                    Some 'reverse-engineering' of Zenblu2015/11/01 01:59 PM
                                      Some 'reverse-engineering' of Zenjuanrga2015/11/02 05:10 AM
                              Some 'reverse-engineering' of Zenjuanrga2015/11/01 07:00 AM
                                Some 'reverse-engineering' of ZenPoindexter2015/11/01 09:25 AM
                                  Some 'reverse-engineering' of Zenjuanrga2015/11/02 04:47 AM
                                    Some 'reverse-engineering' of ZenPoindexter2015/11/02 05:39 AM
                                      Some 'reverse-engineering' of Zenjuanrga2015/11/02 01:26 PM
                                        Some 'reverse-engineering' of ZenDan Downs2015/11/02 03:12 PM
                                          Some 'reverse-engineering' of Zenjuanrga2015/11/03 03:51 AM
                            Some 'reverse-engineering' of Zenlurker2015/10/31 03:57 PM
                              Some 'reverse-engineering' of Zenjuanrga2015/11/01 07:26 AM
                                Some 'reverse-engineering' of ZenHeikki Kultala2015/11/01 09:57 AM
                                  Some 'reverse-engineering' of ZenMaynard Handley2015/11/01 06:39 PM
                              Some 'reverse-engineering' of ZenUngo2015/11/01 02:05 PM
                  Some 'reverse-engineering' of Zenquzhujian2015/10/31 10:39 AM
                  Some 'reverse-engineering' of ZenPoindexter2015/11/04 06:10 AM
                    Some 'reverse-engineering' of Zenlurker2015/11/05 01:54 PM
                      Some 'reverse-engineering' of Zennobody2015/11/05 03:28 PM
                      The sad status of tech mediajuanrga2015/11/06 05:19 AM
                        The sad status of tech mediaJoel2015/11/06 06:45 PM
                        The sad status of tech mediaMatthias Waldhauer2015/11/07 03:12 PM
                          The sad status of tech mediaJoel2015/11/07 05:40 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?