By: anon (anon.delete@this.anon.com),
Room: Moderated Discussions
Ricardo B (ricardo.b.delete@this.xxxxx.xx) on September 13, 2013 3:57 pm wrote:
> bakaneko (nyan.delete@this.hyan.wan) on September 13, 2013 2:04 am wrote:
> > Dan Fay (daniel.fay.delete@this.gmail.com) on September 12, 2013 7:50 am wrote:
> > > bakaneko (nyan.delete@this.hyan.wan) on September 11, 2013 4:46 pm wrote:
> > > > Dan Fay (daniel.fay.delete@this.gmail.com) on September 11, 2013 4:01 pm wrote:
> > > > > > The Quark is supposed to be really small, a Cortex M3 competitor (which, IIRC,
> > > > > > is thumb ISA only, no MMU, no FP). So, I would expect the Quark to be roughly
> > > > > > the same as an 486SX (internal micro architecture wise), but skip all of the
> > > > > > backward compatibility with 16 bit hassles.
> > > > >
> > > > > I seriously doubt that Quark is comparable to an M3 in size. Those
> > > > > things are really simple MCUs meant to replace 8-bit MCUs.
> > > > >
> > > > > Maybe one of the lower Cortex-A-series?
> > > >
> > > > They shouldn't limit it to one or the other.
> > > >
> > > > Otherwise there is the possibility that it
> > > > increases development time or that a
> > > > product based on it fails because it
> > > > doesn't have the right CPU.
> > > >
> > >
> > > That's kind of the problem with x86 -- IIRC, even a 386 is much bigger than a Cortex-M-series.
> > >
> >
> > Much bigger is an overstatement.
> >
> > The original 386 had to support
> > 3 different modes and was still designed
> > for performance, not size.
> >
> > So 20 years later a 376 could actually
> > work and it could be possible to drop
> > the gate count to maybe twice that of
> > an Mx with the same features.
> >
> > But I'm pretty negative here and wouldn't
> > be surprised if it could go lower.
> >
> > (gate count m4 with fpu: ~90k;
> > transistors i386 - 365k)
>
> A 365k transistor 386 was a non-pipelined CPU, without on chip cache or FPU.
>
> A 100k gates (~400k transistors) Cortex-M4 is a pipelined CPU,
> with 1-2 KB cache, DSP extensions and single precision FPU.
> It's more comparable to a 486DX, which had ~1M transistors.
It has no paged MMU, however.
> bakaneko (nyan.delete@this.hyan.wan) on September 13, 2013 2:04 am wrote:
> > Dan Fay (daniel.fay.delete@this.gmail.com) on September 12, 2013 7:50 am wrote:
> > > bakaneko (nyan.delete@this.hyan.wan) on September 11, 2013 4:46 pm wrote:
> > > > Dan Fay (daniel.fay.delete@this.gmail.com) on September 11, 2013 4:01 pm wrote:
> > > > > > The Quark is supposed to be really small, a Cortex M3 competitor (which, IIRC,
> > > > > > is thumb ISA only, no MMU, no FP). So, I would expect the Quark to be roughly
> > > > > > the same as an 486SX (internal micro architecture wise), but skip all of the
> > > > > > backward compatibility with 16 bit hassles.
> > > > >
> > > > > I seriously doubt that Quark is comparable to an M3 in size. Those
> > > > > things are really simple MCUs meant to replace 8-bit MCUs.
> > > > >
> > > > > Maybe one of the lower Cortex-A-series?
> > > >
> > > > They shouldn't limit it to one or the other.
> > > >
> > > > Otherwise there is the possibility that it
> > > > increases development time or that a
> > > > product based on it fails because it
> > > > doesn't have the right CPU.
> > > >
> > >
> > > That's kind of the problem with x86 -- IIRC, even a 386 is much bigger than a Cortex-M-series.
> > >
> >
> > Much bigger is an overstatement.
> >
> > The original 386 had to support
> > 3 different modes and was still designed
> > for performance, not size.
> >
> > So 20 years later a 376 could actually
> > work and it could be possible to drop
> > the gate count to maybe twice that of
> > an Mx with the same features.
> >
> > But I'm pretty negative here and wouldn't
> > be surprised if it could go lower.
> >
> > (gate count m4 with fpu: ~90k;
> > transistors i386 - 365k)
>
> A 365k transistor 386 was a non-pipelined CPU, without on chip cache or FPU.
>
> A 100k gates (~400k transistors) Cortex-M4 is a pipelined CPU,
> with 1-2 KB cache, DSP extensions and single precision FPU.
> It's more comparable to a 486DX, which had ~1M transistors.
It has no paged MMU, however.
