Yousuf Khan said:
Itanium sales have finally surpassed Alpha sales at HP. Looks like it's
mostly in the OpenVMS market though. Most OpenVMS customers are entrenching
around Itanium now. The Alpha-Tru64 market still seems to be volatile.
http://www.techweb.com/wire/story/TWB20040820S0005
Yousuf Khan
QUOTE:
"The performance crossover point -- the point at which IPF will meet,
and begin to exceed by a widening margin, the performance of Alpha --
is expected to occur in the EV7z/Madison9M timeframe," Shannon said
referring to the final iteration of the Alpha family (EV7z) and a new
Itanium configuration (Madison9M.)"
(End of quote)
Well, this is interesting, isn't it? The Madison9M can finally beat
the lower clocked Alpha. That makes me wonder what the promise of
higher IPC rates for EPIC compared to other ISAs are really worth.
Especially when you take the fact into account that the Madison has
huge 9 MiB L3 Cache on die compared to tiny 1.75 MiB L2 of the EV7z.
On-die cache mostly helps a lot for performance.
That's what we have seen regarding single CPU performance as i.e. SPEC
INT/FP BASE 2000 results are higher for Itanium than for Alpha. So if
HP (ok not really HP but Terry Shannon) is still saying that with
Madison9M the performance level of Alpha is reached/overtaken they/he
must refer to SMP systems. Here Alpha has an advantage with its four
highspeed interconnects for direct CPU communication compared to the
shared bus of Itanium which seems to be a bottleneck.
So if performance for larger system is more dependent on interconnect
choice and its implementation I ask myself why a new ISA is neccessary
and why other ISAs had to die in favor of this new one. The point
seems to me not to choose an ISA with a theoretical advantage in ILP
rate but to implement good (low latency) CPU interconnects.
Call me an Alpha fanboy but I find it simply amusing that a 0.18 µm
Alpha EV7z finally gets overtaken in terms of performance from a not
yet released 0.09 µm Madison with 5 times the amount of on-die cache.
Well, I wonder when the point of "overtaken" would have been reached
if the original plans for an EV79 (0.13µm, 3 MB L2, 1.8 GHz+) had been
realized and not delayed and finally canceled by HP.
Sorry for OT, but when hearing those statements from the above quote I
really ask myself why Alpha is being replaced by Itanium.
Regards,
Matt