[RC5] OGR: 32 vs 64bit

Chris Frost chris at frostnet.net
Fri Feb 18 17:07:11 EST 2000


On Fri, Feb 18, 2000 at 10:48:43AM -0600, Wren, Alan wrote:
> At least on the alpha NT code it does not seem to be taking advantage of 64
> bit processing.  On my PIII 500 I get about 2.7 Gnodes in benchmarking, and
> on my Alpha 433 running NT I get about 2.1 Gnodes.  I was hoping it would
> scream on the Alpha like the DES contest did, but it seems to run about the
> same as an Intel box.
> 
> Maybe there are more optimizations to come? :)
I think for the ogr core is has more to do with the code being designed to
take advantage of being 64bit (twice as wide a data path!) instead of
32bit. A 195MHz R10k Octane gets around 1.6Mnodes/s. Fwiw a 200MHz Pentium
Pro gets 1.1 and a 450 Pentium 3 gets 2.3. So it does seem much more
efficient (actually, take a look at http://www.frostnet.net/dnet/dnetc-bench
it is). You mentioend des, I tested the ogr and des cores on an ev6 alpha
(not sure of the speed) and got 2.7Mnodes a second. The des bench was
21.5Mkeys/s! (That same pentium pro gots 0.716Mkey/s.)

From reading about the ogr-23 and 22 effort the core from which dnet's was
derived has quite a bit of speedup for 64bit computers, which was why I
was wondering. I was finally able to test the client on a mips-4 based-box
and found that the n64 (mips4) binary was actualy about 10% slower than
the n32 (mips3) binary. Very odd, why does distributed.net have a
(slower!) n64 binary?

-- 
Chris Frost  |  <http://www.frostnet.net/chris/>
-------------+----------------------------------
Public PGP Key:
   Email chris at frostnet.net with the subject "retrieve pgp key"
   or visit <http://www.frostnet.net/chris/about/pgp_key.phtml>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
Url : http://lists.distributed.net/pipermail/rc5/attachments/20000218/3462d395/attachment-0001.bin


More information about the rc5 mailing list