Ashraf Eassa (aeassa.delete@this.gmail.com) on May 18, 2013 9:06 pm wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on May 17, 2013 8:12 am wrote:
> > Ashraf Eassa (aeassa.delete@this.gmail.com) on May 15, 2013 11:59 am wrote:
> > > Hi everybody,
> > >
> > > I've been lurking for years, but the time has come when I would really love to pick the brains of
> > > the experts we have here. From my understanding, Atom is a much narrower design than Krait, Cortex
> > > A15 and others, and yet, in many benchmarks the older Saltwell core holds its own against even Krait
> > > in both FPU/INT, and against A15 in Linux integer benchmarks (but it gets decimated in FPU).
> >
> > > So, my question is, how do I think about "Silvermont" competitive position against a fairly
> > > beefy modern ARM design such as the Cortex A15? From a high level perspective, it looks
> > > like on a per-clock basis it should be no contest - A15 is wider and more aggressive.
> > > But Intel is claiming that Silvermont is as fast as A15 on a per-clock basis.
> >
> > I don't think anyone at Intel has ever made claims on a per-clock
> > basis. They have made comparisons on a per-unit-of-power basis.
> >
> > > A couple of questions then:
> > >
> > > 1. How can a narrower design pull this off?
> >
> > First, issue width and actual IPC are only loosely related.
> >
> > I'd suggest you look at my article on silvermont and other architectures. I spend a tremendous
> > amount of time discussing all the details including techniques which boost IPC without increasing
> > issue width. My Haswell article is probably also quite informative in that regard.
> >
> >
> >
> > > 2. Is this likely to be integer only as A15 includes FMAC instructions,
> > > which in the right cases double FPU performance?
> >
> > Intel isn't really aiming for FP performance with Saltwell or Silvermont,
> > so it's unlikely that they will do that well compared to the A15. Although,
> > Intel's vector implementation seems much more robust than the A15 Neon.
> >
> > DK
>
> Hi DK,
>
> Thanks for the responses (and to everyone else in the thread too). I shot off the following questions to
> Intel architect Jon Tyler, and got the following responses. They may be helpful for this discussion:
>
> 1. Could you please help me understand the competitive positioning of Silvermont to A15 and,
> perhaps, A57? Intel claimed that Silvermont IPC will be about on-par with A15, despite having a
> narrower front end/back end, smaller OoO structures, lack of FMAC capability, etc. So, my real
> question is: what gives you confidence that Silvermont stays this competitive with A15 on integer
> workloads (this is what the competitive comparison at the uArch disclosure talked about) despite
> the narrower design? Further, should we expect that Silvermont is as competitive on floating
> point with the ARM design as it is on integer (this is what was highlighted at the talk)?
>
>
>
> We have A15 measurements, and Silvermont measurements, so confidence here is based on actual
> measurements we’ve seen. (We cross-check these results with what ARM quotes themselves)
>
> We’ve done a lot of analysis and there are many areas where we have an advantage: our memory latency
> (at each level of cache hierarchy) and memory BW, our superior branch handling, our complex-instruction
> support, our smart resources management, and many other things, allow us to outperform.
>
>
>
> From Silverthorne to Silvermont, gains on FP workload
> should be even better than gains on integer workloads.
> Again, we have A15 measurements, and Silvermont measurements,
> and are confident in our FP scores are similarly impressive.
>
>
> 2. Silvermont products hit the shelves in late 2013, which implies that the competition may not be the
> A15 for too long and instead will be the A57 (and updated micro-architectures from Qualcomm, Apple, etc.).
> Do you guys believe that Silvermont, built on the 22nm process, stays competitive against these new parts
> on performance and performance/watt, or do we need Airmont to get that lead back? I guess broadly, my question
> is, do you see Intel maintaining a continuous lead from here on out, or should we expect "leap-frogging"
> until the Atom schedule is more in-line with the Core parts in terms of process technology?
>
> Silvermont should continue to be very competitive against even what we project as 2014 offerings.
>
>
> 3. I saw very interesting comparisons of Silvermont with Saltwell in the disclosure. What puzzles me,
> though, is that it is very difficult to get a read on CPU-limited performance of these low power micro-architectures.
> For example, a benchmark like "Geekbench" paints "Saltwell" in a rather unflattering light compared
> to the ARM contemporaries, but then you see benchmarks such as AnTuTu showing a 2C/4T Saltwell taking
> leadership positions againt a 4C/4T "Krait" or even Cortex A15 in integer and memory bandwidth, while
> even staying competitive on floating point! Could you help me to understand how Saltwell compares to
> the competition from what you have seen with more sophisticated measurements, and then from there I
> have a lot better context to think about Silvermont performance?
>
> Geekbench is interesting: you look at the results, and the main “unflattering” results
> are in a few sub-benchmarks, where the IA version is set up to handle denorms precisely, and
> the input dataset is configured to be 100% denorms. This is not normal FP code, not a normal
> setup, and possibly not even a apples-to-apples comparison to how ARM is handling these numbers.
> So we view this as an anomaly. (The Geekbench developer agrees with us)
> Saltwell trails A15 in raw IPC, but its higher frequency and threads are able to help compensate.
> Saltwell trails Krait on very basic workloads like DMIPS, but on more complicated
> workloads Saltwell’s robust architecture will pull ahead.

Forgot to bold this part:

From Silverthorne to Silvermont, gains on FP workload should be even better than gains on integer workloads.
Again, we have A15 measurements, and Silvermont measurements, and are confident in our FP scores are similarly impressive.