By: anon (anon.delete@this.anon.com), June 4, 2013 5:53 pm
Room: Moderated Discussions
David Kanter (dkanter.delete@this.realworldtech.com) on June 4, 2013 4:51 pm wrote:
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on June 4, 2013 3:56 pm wrote:
> > anon (anon.delete@this.anon.com) on June 4, 2013 1:36 pm wrote:
> > > > in this segment Arm has
> > > > not a future at all. This is the reason of this orrible presentation :)
> > >
> > > This just shows you haven't paid attention to ARM's marketing before. It has always made overblown claims,
> > > even when there wasn't much reason. The rest of your post reads like fanboy speculation about silvermont.
> >
> > Can you give some examples of overblown claims from ARM? That shouldn't be difficult.
>
> I've found a number of disturbing marketing practices from ARM:
>
> 1. The use of Dhrystone and other CPU-core only benchmarks
> 2. Prohibition on benchmarking with their compiler
> 3. Making performance comparisons between cores on N and N+1 process tech without mentioning that difference
>
> Lack of footnotes on comparisons to provide supplemental detail (e.g., what process, compiler,
> all the subtle conditions; Intel typically has massive backup slides with those details).
>
> But that's not surprising. Marketing tends to be optimistic at most
> companies. No companies are truly honest in terms of marketing.
>
> Incidentally, pretty much every frequency number from an ARM phone or
> tablet vendor is rubbish. Intel is better, but still has issues.
>
> > However Intel's claims are often misleading to the point of lying. I recently posted an example
> > of how Intel typically compares Atom with ARM: use a low-clocked previous generation ARM on
> > an old process to make your own core look significantly faster and more power efficient.
>
> I found Intel's comparisons in the Silvermont presentation to be plausible, although
> it's hard to tell whether their competitive projections are likely to be correct.
> They certainly weren't comparing against ancient and uncompetitive parts.
>
> >And
> > when will Intel finally release a CPU that really shows off those amazing 22nm FinFETs?
>
> I would expect that the ultra-mobile flavors of Haswell should be quite interesting.
>
> David
David mostly hit it. Misrepresented comparisons across process nodes when they show how much better core X is than core Y (and both are ARM cores). Probably to convince partners to relicense. I haven't visited the ARM product pages in a while (and don't really want to), but this used to be on their primary pages.
Marketing sucks everywhere.
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on June 4, 2013 3:56 pm wrote:
> > anon (anon.delete@this.anon.com) on June 4, 2013 1:36 pm wrote:
> > > > in this segment Arm has
> > > > not a future at all. This is the reason of this orrible presentation :)
> > >
> > > This just shows you haven't paid attention to ARM's marketing before. It has always made overblown claims,
> > > even when there wasn't much reason. The rest of your post reads like fanboy speculation about silvermont.
> >
> > Can you give some examples of overblown claims from ARM? That shouldn't be difficult.
>
> I've found a number of disturbing marketing practices from ARM:
>
> 1. The use of Dhrystone and other CPU-core only benchmarks
> 2. Prohibition on benchmarking with their compiler
> 3. Making performance comparisons between cores on N and N+1 process tech without mentioning that difference
>
> Lack of footnotes on comparisons to provide supplemental detail (e.g., what process, compiler,
> all the subtle conditions; Intel typically has massive backup slides with those details).
>
> But that's not surprising. Marketing tends to be optimistic at most
> companies. No companies are truly honest in terms of marketing.
>
> Incidentally, pretty much every frequency number from an ARM phone or
> tablet vendor is rubbish. Intel is better, but still has issues.
>
> > However Intel's claims are often misleading to the point of lying. I recently posted an example
> > of how Intel typically compares Atom with ARM: use a low-clocked previous generation ARM on
> > an old process to make your own core look significantly faster and more power efficient.
>
> I found Intel's comparisons in the Silvermont presentation to be plausible, although
> it's hard to tell whether their competitive projections are likely to be correct.
> They certainly weren't comparing against ancient and uncompetitive parts.
>
> >And
> > when will Intel finally release a CPU that really shows off those amazing 22nm FinFETs?
>
> I would expect that the ultra-mobile flavors of Haswell should be quite interesting.
>
> David
David mostly hit it. Misrepresented comparisons across process nodes when they show how much better core X is than core Y (and both are ARM cores). Probably to convince partners to relicense. I haven't visited the ARM product pages in a while (and don't really want to), but this used to be on their primary pages.
Marketing sucks everywhere.