Re: performance of 21164-300 vs P6-200

Arno PAHLER (paehler@atlas.rc.m-kagaku.co.jp)
Tue, 11 Jun 1996 01:45:12 +0900


The hardware was one of these PCI/ISA PC-style Alphas, with an EB164
motherboard - the compiler flags that I explicitely specified were
-O3 for the Alpha and -O for the P6 (my codes generally seem most
happy with just -O on the P6) - the test that failed definitely does
not compare equality of floats - it is a very straightforward loop -
1000 3D vectors rotated/translated 1000 times - it takes about 0.5
seconds (the 1000 times can be multiplied by a scale to get more
iterations) on a P5-90 - I aborted it on the Alpha after about a minute.
I have run these tests on P5, P6, IBM RS6K, SGIs without any problems
and any source code changes - the total source code size for tests 1
to 3 is 675 lines C. I am kind of suspecting that the code generated
by gcc (no f2c involved here) is plain wrong.

I am mainly interested in hearing how a 300 MHz 21164 does on floating
point specifically under Linux and with f2c and gcc. I would hope that
DEC Fortran and C under OSF/1, else I would wonder why a CPU rated at
512 SPECfp92 does at most 10 to 25 % better than one rated at 283.
The price of the machine I am interested is such that it is about 2x
more expensive than a P6-200 - I would accept if the Alpha was 1.5x
faster at that price, but not what I see - and if I were to run it
with OSF/1 and DEC compilers they'd cost me a fortune and make the
Alpha uncompetitive - and I want a linux environment, most of our
machines here are linux - don't even think of suggesting Windows NT.

What more troubles me are the tests marked as 'didn't run' because
that source code and those data files have been run on virtually any
computer known to mankind (well not quite but a lot - from MicroVaxII
to Crays and multiprocessor Convexes, SGIs and Alphas [under OSF/1
or VMS]).

My question simply is: is it worth shelling out two times the price
of a P6-200 for an Alpha running linux - I could live with sub-
standard performance for half a year or so until compilers got better -
I can't live with the fact that programs I know run on many machines
won't even run properly.

Arno