someone (someone@somewhere.com) on 2/5/09 wrote:
>
>No. HP only sells the 24 MB SKU in bigger machines with
>slower memory systems and/or slower FSBs.

Ok. That's unexpected (I'd expect them to want to have
a "high end" sku for people who want to pay for them), but
Itanium never made much sense to me as a market so I guess
"unexpected" is just par for the course for me ;)

>Same compiler, 1.6 GHz:
>
>r6600, 24 MB L3, 533 MHz FSB - 14.5 SPECint_base2006
>r3600, 18 MB L3, 533 MHz FSB - 13.9 "
>r2620, 18 MB L3, 400 MHz FSB - 13.4 "

I was too lazy to wade through the details, so I just went
by "same vendor, same date" (as an approximation for "same
compiler"), and looked at the frequency and cache size.

And I then expected a larger cache to go with the more
expensive product that would presumably have had a better
FSB, rather than the other way around.

So I did not really expect to see the "oh, more cache but
crappier FSB" case. Yes, I mentioned it as a possibility,
but I didn't really think anybody would do that.

Thanks for lining them up like that.

>33% more BW = +4%
>33% more L3 = +4%
>33% more BW+L3 = +8%

And I'll happily admit that 4% is actually a bigger change
from just the higher FSB frequency than I expected. I
guess I'm still mentally stuck with Spec2000.

>Changes - Mont 9xxx -> Tukwila:
>
>smaller L3 per core (12/9 MB -> 6 MB per core)
>lower memory latency (4s: 10s of ns, 16s+: 100s of ns)
>higher mem BW per core (0.8-2.6 GB/s -> 8.5 GB/s)
>higher CPU frequency (1.60-1.66 -> 2 GHz = 20-25%)
>
>so
>
>- SPECint likely won't improve too much (<10% ?) above
>GHz increase (20-25%) and compiler improvements (???).

Agreed. And I really think that the frequency thing is
what matters. 2GHz is going to help. But 2.4GHz or
something like that would have been a much bigger deal.

As it is, it may drag the numbers from "pitiful" to "not
nearly competitive with x86, but at least with POWER".

>- bigger improvements in SPECfp, huge improvements in
>SPEC_rate.
>
>- bigger system = bigger relative performance boost.

.. except x86 grew bigger much faster with Core i7
(assuming Intel does release the 8-core version, of course,
and I don't think there's any reason to really doubt that).

So all the Specrate and parallel performance metrics will
actually be much smaller compared to x86.

Linus