charl...@mcn.org (Charles Mills) writes:
> Why is that useful? Because the speed gains in the last several generations
> of mainframe are not in clock/cycle speed. System 370 object code does not
> run any faster on a z13 than on a z10. The gains are in new instructions.
> The same functionality as that S/370 object code expressed in z13 object
> code runs a lot faster.
count of latency to memory (& cache miss), when measured in count of
processor cycles is comparable to 60s latency to disk when measured in
count of 60s processor cycles.
claim is that over half of per processor improvement from z10 to z196
was introduction of latency masking technology (that have been in other
platforms for decades), out-of-order execution (proceed with other
instructions while waiting for cache miss), speculative execution and
branch prediction (start executing instructions before condition for
branch is available), etc. (aka basically do other things while waiting
z196->ec12 is supposedly further refinements in memory latency masking
features (again have been in other platforms for decades).
z900, 16 processors, 2.5BIPS (156MIPS/proc), Dec2000
z990, 32 processors, 9BIPS, (281MIPS/proc), 2003
z9, 54 processors, 18BIPS (333MIPS/proc), July2005
z10, 64 processors, 30BIPS (469MIPS/proc), Feb2008
z196, 80 processors, 50BIPS (625MIPS/proc), Jul2010
EC12, 101 processors, 75BIPS (743MIPS/proc), Aug2012
z13 published refs is 30% more throughput than EC12 (or about 100BIPS)
with 40% more processors ... or about 710MIPS/proc
I've told story before about after FS imploded there was mad rush
to get stuff back into product pipelines ... some past posts
303x and 3081 were kicked off in parallel, 3031&3032 were 158&168
repackaged to work with channel director ... and 3033 started out 168
logic mapped to some warmed over FS chips that were 20% faster.
we had a project to do 16-way SMP multiprocessor and had con'ed the 3033
processor engineers to work on it in their spare time (lot more
interesting than 3033), many in POK thot it was really neat ... until
somebody told the head of POK that it could be decades before the POK
favorite son operating system had effective 16-way support ... then some
of us were instructed to never visit POK again (and the 3033 processor
engineers were told to stop getting distracted). 16-way finally ships
almost 25yrs later (Dec2000).
virtualization experience starting Jan1968, online at home since Mar1970
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN