joel (no.delete@this.way.com) on August 8, 2014 2:26 pm wrote:
> juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 12:09 pm wrote:
> > Megol (golem960.delete@this.gmail.com) on August 8, 2014 11:23 am wrote:
> > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 10:49 am wrote:
> > > >
> > > > They compared older hardware. Migrating from SB-i7 to HW-i7 introduces little benefits
> > > > in performance (except when using new AVX2 extensions to x86) but in ARM each gen is
> > > > not a mere 5-10% faster than former gen but much more. Their choice favored x86.
> > >
> > > 5-10% is a huge difference given that it results from slight
> > > polishing. I have to say this point is nonsense.
> >
> > My point has been ignored.
> >
> > > > They used only Intel designs. Using only AMD or a mixture of AMD and Intel
> > > > would change both performance and efficiency. Their choice favored x86.
> > >
> > > But would that be relevant? If Intel processors are the most efficient
> > > x86 ones available shouldn't that be a wise choice?
> >
> > If instead comparing Intel designs against ARM Standard cores they had compared AMD designs
> > against best ARM cores, their performance and efficiency values had changed completely.
> >
> > If they were really comparing ISAs, they would have used a mixture of different chips from
> > different manufacturers to minimize the impact due to concrete archs. and process nodes.
> >
> > > > They computed the power consumption incorrectly. Their methodology choice favored x86.
> > > >
> > > > They used an old compiler that didn't optimize the code for ARM all that would do. Their choice favored x86.
> > >
> > > Most code in the wild isn't optimized much.
> >
> > The software that I use wasn't compiled with their ancient compiler.
> >
> > > > Also the x86 ISA is full of legacy instructions, which have to be implemented
> > > > in hardware and then verified/tested which increases development costs and time of development.
> > >
> > > Wrong. Legacy instructions need some hardware, true. But most of the functionality
> > > is implemented in microcode instead of adding complex hardware.
> > > Now there are some quirks in the x86 ISA that does waste power like handling
> > > of shift by zero, calculating the auxilary flag (nibble carry) etc. But those
> > > are far from the most power consuming parts of an OoO processor core.
> >
> > I gave typical costs and time of implementation of ARM vs x86
> > before. Resume: 10x more $ and 3x more months for x86.
> >
> > Some studies estimate that microcode ROM takes about a 20% of die for small cores.
> >
> > We have independent data showing Intel x86 designs loosing against
> > ARM despite Intel having a clear process node advantage.
> >
> > The ISA doesn't matter myth has been debunked lots of times.
> >
> > It is interesting that Intel claims that the ISA doesn't matter, x86 is enough, but then is introducing
> > most of its performance/efficiency gains from new ISA extensions to x86: AVX, AVX2, AVX512, TSX...
> They alleviated the historical complexity of x86 by... adding new crap?
>

Most of those extensions are targeted at specific niche of numeric computing (multimedia nd HPC) and have approximately no effect on general-purpose code.
As to numerics, nobody says that difference between x87 and good scalar FP ISA (e.g. Alpha) does not matter. My guess would be that x87 tax = 1.6x with ideal compiler and ~2x with real-world compilers. And that's even before we started talking SIMD.

And I don't understand why one would want to call AVX crap. It's mostly very good stuff.