Performance Comparison rx7620 vs GS1280

[EWL]

We are searching for any kind of information that can give us an idea of the performance difference in an rx7620 vs GS1280. We've asked and searched the web, but either not recognizing the data or it's not available. Anyone have suggestions for SpecInt for VUPS or flops or whatever?

[Hoff]

Best guess: the rx7620 will be slower.

VUPS? That benchmark died with VAX.

Here are the SPEC 2000 submissions (Google is my friend) for the two boxes:

http://www.spec.org/cpu/results/res2003q1/cpu2000-20030113-01918.html
http://www.spec.org/cpu/results/res2003q4/cpu2000-20031020-02595.html

All current Itanium boxes with current or recent processors are faster than the corresponding Alpha boxes, as per OpenVMS engineering representatives.

The rx7620 isn't current gear.

And unfortunately, it is very possible for even a brand-new Itanium box to run slower than the benchmarks would imply for specific degenerate applications. Non-optimal or non-optimized code runs slowly. No one here can tell you your application will be faster without a look at the application; it'll *probably* be faster on a current-generation Itanium, but then the AlphaServer GS1280 and EV7z was pretty speedy.

For comparison purposes, the recent QuickPath Interconnect (QPI) design in use over on the latest Nehalem-class x86 benchmarks faster than the EV7z torus bandwidth, and QPI is reportedly coming to Itanium with its next generation. EV7z chip and the Marvel-class bus latency and bandwidth was (and still is) impressive.

What to do here? Benchmark real examples of real code on the target system, and see where you end up. That is, unless you're planning on running the actual SPEC benchmarks as your production load. (And I'm guessing you're not planning to do that.)

If you're looking to purchase this rx7620 from HP, ask for access into the OpenVMS customer lab, and run some (real) benchmarks from your (real) application code on a machine that HP reports is equivalent. Details on the lab are here:

http://h71000.www7.hp.com/custlab/index.html

As for comparisons, the rx7620 looks to be a generation or two back both in platform and in the processor; the current matching gear here is probably the 1.6 GHz rx6600 or the Superdome, depending on the application and the local requirements.

http://h18000.www1.hp.com/products/quickspecs/12596_div/12596_div.HTML
http://www.hp.com/go/superdome

And as should be obvious, the SPECint2000 from that benchmark cited earlier is substantially better on AlphaServer GS1280 and EV7z than on the rx7620 box.

[Martin P.J. Zinser]

I second the suggestion to look at an rx6600. Given a "normal" behaved application these should be somewhat faster than a GS1280 (but much more compact). An the rx6600 is still a standard SMP box, while the rx8xxx and Superdomes get you in the area of hard partitions, which can have interesting effects concerning memory latency.

[Ian Miller.]

Do follow up with the suggestion to use HP Customer Labs. Performance comparisons like this is one of the things that they can do.

[Ian Miller.]

link to customer lab info
http://h71000.www7.hp.com/custlab/

[EWL]

Thanks for all the suggestions/responses.