🚀 go-pugleaf

In article <BxMn1J.GB3.2@cs.cmu.edu>, lindsay+@cs.cmu.edu (Donald Lindsay) writes:
|> NetCmmnd@boole.uucp (System Administrator) writes:
|> >How can a non-superscalar implementation at 50MHz run 59.1 MIPS?
|> >I thought the holy grail of RISC was 1 instruction retired per cycle

|> However, "Dhrystone 1.1 MIPS" have ceased to be reasonable.  The
|> "MIPS" is calculated by assuming that the VAX 11/780 was 1 MIPS, and
|> then taking an execution-time ratio versus a historical VAX value.
YES.
It is interesting to note that the SUNFLASH introducing the low-end machines
emphasizes MIPS-ratings (and doesn't even say Dhrystone-mips....) and says:

"....
These new Sun(TM) workstations deliver 59.1 MIPS, with the SPARCclassic
offering industry-leading value at $63 per MIPS.
....
These new systems deliver more performance than `486 PCs, with 59.1
MIPS, 26.4 SPECint92 and 21.0 SPECfp92."

It is interesting to note the historical trend:

System	SPECint92	MIPS-rating	MIPS/SPECint92
SS IPC	13.8		16		1.2
SS2	21.8		28.5		1.3
-------------
SS LX	26.4		59.1		2.2
SS10/30	45.2		101.6		2.2
SS10/41	53.2		109.6		2.1

Yep:
1) Higher performance than `486 PCs: is certainly true, you can find 486 PCs
that are slower ... of course, you can find some that are faster,especially
with positioning in commercial market, and where 486DX2-66 certainly has a higher SPECint92.  The Sun's SPECfp92 is higher, I think.
2) There were no cost/SPECint or SPECfp numbers, just cost/(unspecified) MIPS;
obviously the former can be computed....
3) What's really fascinating is the combinations of compilers used to get
the various numbers (from other announcements):

CINT92: Sun C 3.0 Alpha
CFP92: Apogee C & FORTRAN 0.82 Alpha with integrated Kuck (for Classic & LX)
	Sun C & FORTRAN 3.0 Alpha + Kuck (for SS10s) 
MIPS: Dhrystone 1.1: Apogee C 0.82 Alpha
LINPACK: Apogee FORTRAN 0.82 Alpha

Apogee is listed as being available 1/93; the Sun compilers in 5/93.
Hence:
	If you want best Dhrystone- MIPS rating, use Apogee C
	If you want best SPECint92, use Sun C in 5/93
	If you want best SPECfp92, use Apogee (soon, for low-end), or
			Sun (in 5/93 for SS10s)
	If you want best LINPACK, use Apogee (soon)
It will be fascinating to see which combinations real people will choose:
maybe somebody from SMCC would care to post something about advice to users about which compilers to actually buy?

 
-- 
-john mashey    DISCLAIMER: <generic disclaimer, I speak for me only, etc>
UUCP:    mash@sgi.com 
DDD:    415-390-3090
USPS:   Silicon Graphics 7U-005, 2011 N. Shoreline Blvd, Mountain View, CA 94039-7311