It appears that I didn't spend enough time researching the differences
between SPEC*92 and SPEC*95. For a given hardware setup, SPECint92 values
are ~40-50x SPECint95 values. Or, as Peter Stern wrote, "It's like comparing
grams and ounces."
The LINPACK value is not meaningless, but it probably isn't a very useful
value for the apps we intend to run on the machine.
Thus the AlphaServer is ~2-4x faster than the 3000. That, along with the PCI
vs. Turbochannel issue, makes this a comparative no-brainer. Quoting from
another respondent, Robert L. McMillin, "... you'd be a fool not to take the
1000A. It's way faster, and has a substantially more robust SCSI
implementation. If DEC *really* wants back their old box, let 'em have it.
They'll be lucky to get ten cents on the dollar."
Thanks to Peter, Robert, and the other dozen or so respondents. I appreciate
your help very much.
--------------------------
ORIGINAL POST:
For reasons I won't go into here, my company has the opportunity to buy a 2
year old DEC 3000-700 Workstation that was loaned to us for some project
work. Since Digital has dropped their prices on most of their line, I wanted
to compare what I could get in a new Alpha.
I should mention that the 3000 was loaned to us to run as an NMS, and I
never really considered buying it until we got a helluva quote. Also,
Digital either wants us to buy the 3000 or return it. What we really need is
a Unix development machine that can double as a file/application server
(like many small companies, we're not exactly gushing with cash). With
discounts and all, I can get an AlphaServer 1000A 5/400 w/128MB for about
the same price.
I wanted to make sure that I wouldn't be taking too much of a hit on
performance, so I tried to check benchmarks between the two. Now I'm really
confused. Here's what I found:
3000-700
SPECint92 162.6
LINPACK 100 x 100 26.9 MFLOPS
Alpha 1000A 5/400
SPECint95 11.5
LINPACK 100 x 100 186.6 MFLOPS
Several questions come to mind. Is the 3000 really 14x faster than the
1000A? I realize the SPECint values are from different benchmarks, but
according to the SPEC FAQ there is a "certain correlation" between SPECint92
and SPECint95, though there is "no magic formula" for conversion.
Conversely, is the LINPACK value completely meaningless?
Finally, I would welcome any and all opinions/recommendations that you may have.
--
Casey Spangler spang_at_probita.com
Probita, Inc. 303-449-7665
"Go, and never darken my towels again!" -- Groucho Marx
Received on Tue Jan 21 1997 - 18:18:53 NZDT