N2, or V1?

By: dmcq (dmcq.delete@this.fano.co.uk), December 5, 2021 2:43 am
Room: Moderated Discussions
Adrian (a.delete@this.acm.org) on December 5, 2021 12:15 am wrote:
> Rayla (rayla.delete@this.example.com) on December 4, 2021 1:20 pm wrote:
> > Adrian (a.delete@this.acm.org) on December 4, 2021 2:33 am wrote:
> > > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on December 3, 2021 1:58 pm wrote:
> > > > Adrian (a.delete@this.acm.org) on December 3, 2021 10:51 am wrote:
> > > > > G (108106020.delete@this.nitt.edu) on December 3, 2021 9:25 am wrote:
> > > > > > Adrian (a.delete@this.acm.org) on December 3, 2021 8:52 am wrote:
> > > > > > > Adrian (a.delete@this.acm.org) on December 3, 2021 8:47 am wrote:
> > > > > > > > Anon (no.delete@this.thanks.com) on December 3, 2021 6:52 am wrote:
> > > > > > > > > Kara (karaardalan.delete@this.gmail.com) on December 3, 2021 6:36 am wrote:
> > > > > > > > > > none (none.delete@this.none.com) on December 3, 2021 6:19 am wrote:
> > > > > > > > > > > Kara (karaardalan.delete@this.gmail.com) on December 3, 2021 6:05 am wrote:
> > > > > > > > > > > > Adrian (a.delete@this.acm.org) on December 3, 2021 5:51 am wrote:
> > > > > > > > > > > > >
> > > > > > > > > > > > > https://www.nextplatform.com/2021/12/02/aws-goes-wide-and-deep-with-graviton3-server-chip/
> > > > > > > > > > > > >
> > > > > > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > The N2 is described much wider here than Arm has ever disclosed.
> > > > > > > > > > > > I hate it when companies do this, like give some info what's gonna happen?
> > > > > > > > > > >
> > > > > > > > > > > What description of N2 is there in that article? I only see guesses.
> > > > > > > > > >
> > > > > > > > > > I meant the guesses, it's so not ARM to design 8 wide decode and 15 wide issue
> > > > > > > > > > mArchs. 64kb l1i cache wouldn't be able to keep up Nor the punly ROBs, how
> > > > > > > > > > do you feed such a wide core? Without smt or legendry l1i and rob?
> > > > > > > > > >
> > > > > > > > > > Neoverses are targeted toward server processors, making them wider doesn't make any sense.
> > > > > > > > >
> > > > > > > > > I thought that this was the V1 core? Aimed at maximum perf per thread.
> > > > > > > >
> > > > > > > >
> > > > > > > > Amazon did not disclose which core they have used, but what was said matches better N2 than V1.
> > > > > > > >
> > > > > > > > Choosing N2 also makes more sense for Amazon, because they are more interested in
> > > > > > > > general-purpose performance within a given power consumption than on performance
> > > > > > > > when only few threads are active or performance for floating-point applications.
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > Moreover, Graviton 3 uses 100 W for 64 cores.
> > > > > > >
> > > > > > > After subtracting uncore power, the power limit per core will be 1.25 W at most.
> > > > > > >
> > > > > > > This power level is below what a core from the X1/X2/V1 series requires, which is in the 2 ... 3 W range.
> > > > > > >
> > > > > > >
> > > > > > > Power levels in the range 1 ... 1.5 W per core are the target for N1/N2.
> > > > > > >
> > > > > > > This is an additional reason to believe that Graviton 3 uses N2.
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > >
> > > > > > I think Graviton3 uarch matches exactly with cortex-V1.
> > > > > >
> > > > > > https://i2.wp.com/semianalysis.com/wp-content/uploads/2021/12/7.jpg?w=2048&ssl=1
> > > > > > https://images.anandtech.com/doci/16640/Neoverse_V1_13_575px.png
> > > > >
> > > > >
> > > > > Some parts of the slides do indeed match well.
> > > > >
> > > > > We do not know which are the customization options provided by ARM for V1.
> > > > >
> > > > > From these slides, it looks like the core used in Graviton 3 is indeed based on V1, but with
> > > > > some parts deleted to lower the power consumption, e.g. with only half of the SVE datapath.
> > > >
> > > > That doesn't make any sense. As the optimization guide clearly states, V1 can either run
> > > > 4 128-bit Neon FMAs or 2 256-bit SVE FMAs. Both use the same 4 128-bit pipes. Obviously.
> > > >
> > > > > In that case the Graviton 3 cores, being slightly cut and used at a relatively low clock
> > > > > frequency, to be able to work within the 100 W TDP, should have intermediate performance
> > > > > and energy efficiency characteristics between an N2 and a V1 in maximum configuration.
> > > >
> > > > +60% on all-core SPECFP_rate doesn't exactly look like a minor gain, neither does this. The +60%
> > > > means at just 2.6GHz it beats EPYC 7763 at 3.5GHz on fprate while using a fraction of the power.
> > > >
> > > > Wilco
> > > >
> > >
> > > Yes, but the Graviton 3 slides show 1 x 256 SVE, not 2 x 256 SVE.
> > >
> > > This was also mentioned on the NextPlatform site, by someone who presumably has seen the
> > > complete presentation, as the reason why he did not believe that the core is a V1.
> > >
> > > However most of the other parts of the slides match V1, including the ARMv8.5-A compliance.
> > >
> > >
> > >
> > >
> >
> > What slide shows that it only has one 256b SVE pipe? All they say is that they have
> > 256b SVE - which is in line with what the V1 has. It's not different from Intel saying
> > that ICL and SKL-SP have 512b AVX, despite having multiple 512b datapaths.
>
>
> I agree that the slide is ambiguous, but that area in the diagram matched
> the same area in the ARM diagram, except that the "2x" was removed.
>
> Also, this was not my own interpretation, but of Timothy Prickett Morgan from
> NextPlatform, who asked Amazon about this, but did not receive any reply yet.
>
> If Graviton 3 would have indeed a complete V1, running 2 x 256-bit FMA at 2.6 GHz
> with only a power consumption of 1.0 ... 1.25 W per core, that would be a huge improvement
> in energy efficiency over the existing CPUs, which does not seem likely.

It has four SIMD units so I think it is practically definitely a V1 2x256 bit SVE chip. There just wouldn't be any point in doing anything else. The cores would be run at 2.6GHz and with 5nm that would cut the power down greatly.

< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Some info about the Amazon Graviton 3Adrian2021/12/03 06:51 AM
  Some info about the Amazon Graviton 3Kara2021/12/03 07:01 AM
    Some info about the Amazon Graviton 3---2021/12/03 10:03 AM
      Some info about the Amazon Graviton 3Kara2021/12/03 10:45 AM
  Some info about the Amazon Graviton 3Kara2021/12/03 07:05 AM
    Some info about the Amazon Graviton 3none2021/12/03 07:19 AM
      Some info about the Amazon Graviton 3Kara2021/12/03 07:36 AM
        N2, or V1? Anon2021/12/03 07:52 AM
          N2, or V1? Adrian2021/12/03 09:47 AM
            N2, or V1? Adrian2021/12/03 09:52 AM
              N2, or V1? G2021/12/03 10:25 AM
                N2, or V1? Adrian2021/12/03 11:51 AM
                  N2, or V1? Wilco2021/12/03 02:58 PM
                    N2, or V1? Adrian2021/12/04 03:33 AM
                      N2, or V1? -.-2021/12/04 04:37 AM
                      N2, or V1? Rayla2021/12/04 02:20 PM
                        N2, or V1? Adrian2021/12/05 01:15 AM
                          N2, or V1? dmcq2021/12/05 02:43 AM
                            N2, or V1? Adrian2021/12/05 03:54 AM
                              N2, or V1? ---2021/12/05 11:45 AM
                                N2, or V1? Adrian2021/12/05 01:07 PM
                                  Other (minor) power factors?Paul A. Clayton2021/12/06 07:37 AM
                    N2, or V1? Anon2021/12/04 10:53 PM
                      N2, or V1? Andrei F2021/12/05 04:22 AM
                Only 4 ALUsJörn Engel2021/12/03 07:37 PM
                  Only 4 ALUsWilco2021/12/04 09:54 AM
          N2, or V1? -.-2022/05/24 06:34 AM
            Graviton3 on Chip &CheesePer Hesselgren2022/06/17 06:19 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?