SM[567]1/HyperSPARC comparison (was RE: [rescue] SS10/20 death)

[rescue at sunhelp.org]

If I had a quad SM51 system, then my upgrade path would
be to an Ultra 2.  Seriously--a dual SM71 has real-world
performance on the order of a U1/170 based on personal
experience.  (The SPEC ratings[0] I can find seem to confirm
that--but I can't find SPEC ratings for the HyperSPARC
chips.)  A quad SM51 should be similar or perhaps
somewhat better.  The advantages of going to an U1/U2 are
things like more memory capacity (128MB DSIMMs) and support
for recent 24-bit color frame buffers, IMHO.

Besides Dave's point about clock efficiency[1], the other
concern I have with the HyperSPARCs is the heat they
generate and resulting need for more fans.  It's not until
the 150+ MHz HyperSPARCs come into play that the processor
really has that much better performance than the SM51/61/71/81,
and those suckers run *really* hot.  It's also necessary
to start becoming concerned with RFI at that point, from
discussions with other folks.

As for me, I love having unbalanced systems--my Axil has
one SM51 and one SM61 and it's always fun to see the look
on PC-type techies' faces when I explain that.

[0] SPARC SPEC ratings:
I have an evil table of MIPS/SPEC info assembled from a
grunch of different sources that someone could re-format
and finish populating if they care--I ran out of
interest.  The key comparisons from the populated portions
of that SPARC SPEC table are as follows:

SPARC LX:
  SPECint92		26.4
  SPEC-CINT95	NR

SPARC 10, single SM40:
  SPECint92		50.2
  SPEC-CINT95	1.13

SPARC 20, single SM51:
  SPECint92		73.6
  SPEC-CINT95	NR

SPARC 20, single SM71:
  SPECint92		NR
  SPEC-CINT95	3.11

SPARC 5/170:
  SPECint92		NR
  SPEC-CINT95	3.53

Ultra 1/170E:
  SPECint92		NR
  SPEC-CINT95	6.26

Ultra 2/1300 (single 300MHz)
  SPECint92		NR
  SPEC-CINT95	12.1

The SPEC tests aren't perfectly representative of real-world
performance, but they're at least some reference point.  If
things scaled perfectly/linearly (and throwing in some voodoo
math to convert '92 results to '95), a quad SM51 would be
about the same as a dual SM71, or about the same as a U1/170E.
Anyone know where the HyperSPARCs fit?  

> > > In fact, I would rather have a dual SM51 system than
> > > a quad 100MHz Hypersparc system--but that's just me.
> > 
> > Why? I happen to have a quad SM51 SS20, and the one thing that keeps
> > nagging at me is the possibility of upgrading it to quad 100MHz
> > HyperSPARCs. The biggest advantage in my book would be the 
> recovery of
> > those two SBUS slots, but I would think performance would improve in
> > some way, no?
> 
>   SuperSPARCs are much more clock-efficient than HyperSPARCs.  The one
> datapoint that I can personally verify is that a single SM61 will
> *trounce* a single 125MHz HyperSPARC.

[1] SM-61 should be SuperSPARC architecture, while SM71/81 are actually
    SuperSPARC-II.  *Supposedly* SuperSPARC-II is significantly faster--
    which is why an SM-71 might actually be worth half again as much
    as an SM-61.  I've never done any benchmarking on the systems I
    have, but if anyone cares let me know a good freely-available
    benchmark suite to run.  I do know that in limited testing a single
    SM71 was *way* faster than a single 90MHz HyperSPARC, before I
    sold the HyperSPARC to someone concerned only about MHz.

Let's see how long it takes for someone to find a grievous
error in my babbling above.  </PEDANTIC>

  --Rip