Heikki kultala (heik.kikultal.a.delete@this.gmail.com) on July 12, 2020 2:11 pm wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on July 12, 2020 1:37 pm wrote:
> > Heikki Kultala (heik.kikultal.a.delete@this.gmail.com) on July 12, 2020 1:30 pm wrote:
> > > Michael S (already5chosen.delete@this.yahoo.com) on July 12, 2020 1:18 pm wrote:
> > > > Linus Torvalds (torvalds.delete@this.linux-foundation.org) on July 12, 2020 11:39 am wrote:
> > > > > ⚛ (0xe2.0x9a.0x9b.delete@this.gmail.com) on July 12, 2020 11:07 am wrote:
> > > > > > >
> > > > > > > I've said this before, and I'll say it again: in the heyday of x86, when Intel was laughing all the
> > > > > > > way to the bank and killing all their competition, absolutely everybody else did better than Intel
> > > > > > > on FP loads. Intel's FP performance sucked (relatively speaking), and it matter not one iota.
> > > > > >
> > > > > > As far as I remember, in the days of the original Pentium and
> > > > > > Pentium MMX, Intel had the best FP32/FP64 performance
> > > > > > in x86 CPU space
> > > > >
> > > > > Oh, not denying that part.
> > > > >
> > > > > But that wasn't my point. The competition I was talking about was the non-x86 one.
> > > > >
> > > > > Every single RISC architecture out there made x86 look pitifully bad.
> > > > >
> > > > > And they all died and went down in flames, because absolutely nobody cared.
> > > > >
> > > > > Linus
> > > >
> > > >
> > > > If you look at RISC market leaders, i.e. Sun and HP, I don't think that they were really ahead of Pentium
> > > > in FP performance in 1991-1994. They had better caches,
> > > > better memory bus and better sustained-to-peak ratio,
> > > > but Pentium had higher peak FLOPs, so at the end at application level all three were pretty close.
> > > > And PPC601 was probably slightly behind, but cheaper.
> > >
> > > Peak flops are irrelevant and do not represent the real-world performance.
> > >
> > > x87 with the terribly broken stack system needed huge amount of extra fxch instructions to do anything.
> > >
> > >
> > >
> >
> > But the difference in peak flops in favor of Pentium was big enough to compensate the badness of
> > x87. Which is probably no worse than 1.5-1.7x relatively to flat FP register of the same size.
> > Except, when you have wide in-order core and long latencies. But that's not Pentium.
>
> Pentium was not even on market at 1991-1992. And afaik it
> did not have any "peak flops advantage" in 1993-1994
>
> And 486, which was in market in 1991-1992 had much weaker, non-pipelined fpu
>
>

I got 1991-92 wrong, sorry.

In '93 Sun still had the same 36-40 MHz SuperSparc from 1992. 40 MFLOPs vs Pentium's 50 to 66.
They also had HyperSPARC at around 100 MHz. I didn't find much info about it. I don't think it had on-chip FPU. Or, may be, it had something in i486 class?

In '94 Sun got SuperSparc-II. Without to much googling it seems it was at 75 MHz in H1 and 85 MHz closer to end of the year. Pentium was at 100 MHz by then. So the ratio of peak FLOPs remained about the same.

It's becoming late so I'd give up on PA-RISC googling.