Re: Portents of Itanium death

From: Bill Todd (billtodd_at_metrocast.net)
Date: 06/17/03 

Date: Tue, 17 Jun 2003 16:41:51 -0400

"Rob Young" <young_r@encompasserve.org> wrote in message
news:jomTkTPmcupf@eisner.encompasserve.org...
> In article <bcno9701tlr@enews2.newsguy.com>, "Dann Corbit"
<dcorbit@connx.com> writes:
> > "Rob Young" <young_r@encompasserve.org> wrote in message
>
> >
> > Depends on what you are doing. For database benchmarks and for things
that
> > need huge memory or disk, 32 bits craps out early.
> >
>
> Oh really?

Actually, by definition in the case of 'things that need huge memory'. And
that includes current TPC-C configurations (at least those using > 4 GB of
physical memory): IA32 *can* address it, but nowhere nearly as efficiently
(let alone easily) as a flat 64-bit model can.

There's no obvious reason that the architecture word size has anything
whatsoever to do with heavy disk activity, though: it's Opteron's
significant bandwidth advantage that makes it superior to IA32 there.

>
> > Here is a favorable TPC report, even using a database compiled for 32
bits:
> > http://www.tpc.org/results/FDR/TPCC/RackSaverQuatreX-64FDR.pdf
>
> Good thing to bring up I suppose. Bill has chided me, that
> is a 82000 tpmC result if memory serves. You should be able
> to find its corresponding Xeon equivalent competitor
> - a 4-CPU 2.0 GHz Xeon Dell - at 78000 tpmC.
>
> When the Xeon goes from 2.0 GHz to 2.8 GHz, the Opteron will
> be trailing the Xeon.

That's probably true in the specific case of TPC-C (Opteron's lead over Xeon
in other benchmarks probably won't be threatened, because other benchmarks
don't seem nearly as cache-size-sensitive), though it's not clear from
analysis of the patterns of TPC-C scores just how much better TPC-C
performance that 40% boost in clock rate will translate to (since even
though they'll supposedly be available within weeks no one seems to have
posted TPC-C results for them yet: do you still consider that
irresponsible?) - nor how long they'll retain whatever lead they manage to
eke out (since the 2 GHz Opterons are also imminently due, followed by 2.2
and 2.4 GHz versions before year's end).

What *is* clear is that those Xeon MPs will still cost close to 5 times as
much as their Opteron counterparts, and that the systems they're available
in will still cost close to twice as much (i.e., far costlier $/tpmC). So
the only reason you'd buy one would be if your performance needs were
precisely in the (SWAG) 85K - 100K tpmC range: lower, and Opteron will
satisfy your needs far more cost-effectively; higher, and Xeon won't satisfy
them at all.

...

  But my point was/is Xeon gets a huge
> speedup in the 2 processor space - going to 3.06 GHz

>From its current 2.8 GHz: do you consider a 9% speed increase 'huge'?

 and the
> Xeon MP gets a big speed bump going from 2.0 to 2.8 GHz. June 30th.
>
> > Someday in the near future, due to active development by AMD, it may be
a
> > better than the Alpha chip.
>
> Sure. Athens - the next spin on Opteron - is "the great hope" now

No: current Opterons will be more than adequate as long as they meet the
dates for 2, 2.2, and 2.4 GHz versions that AMD has set forth.

> as Xeon will be faster than Opteron for at least 6 months.

Not if AMD keeps to its schedule: 2.2 GHz Opterons in 2 - 3 months should
about equal the $3692 2.8 GHz Xeon MPs in TPC-C and 2.4 GHz Opterons in late
fall should surpass them - still at far lower prices. The late-fall 90 nm
3.4 GHz Xeons with 1 MB of on-chip cache should be considerably more
cost-effective for most uses than the Xeon MP dinosaurs and just about equal
the 2.4 GHz Opterons in both performance and price (though only in
applications where multi-processor bandwidth is not critical, since Opteron
promises to lead there for the foreseeable future) - though since they'll be
a full process generation ahead of Opteron that's hardly impressive, and
Opteron is scheduled to move to 90 nm before next summer.

- bill

Relevant Pages

  • Re: Portents of Itanium death
    ... > 2.0 GHz certainly isn't going to do much regarding an increase. ... How much a 40% clock-rate boost will help the Xeon MP ... One was a 2.4 GHz system with 8 GB of RAM, ... > Where is Opteron in that space? ...
    (comp.os.vms)
  • Re: Do you own a Mac Pro?
    ... I was the one giving these prices and I said that the cpu's were ... ghz dual core $328 ... If you're trying to prove that a mac pro costs too much as you appear to ... ghz xeon to a do-it-yourself version with a 2.13ghz xeon. ...
    (comp.sys.mac.system)
  • Re: Using FPGAs on AMDS Opteron to accelerate servers 20--100 times
    ... GHz. ... combined with AMD's on-chip memory controller eliminate latency ... It also makes for the "scaling" of multiple CPUs much more linear. ... i.e. if you have 8 OPterons, it's going to be "nearly" as fast as the ...
    (comp.lang.apl)
  • Re: Comparing Opteron to Xeon
    ... >> To make matters more complex, BTW, the Opterons are dual core. ... > The Xeon won't see the light of day in a direkt comparison. ... more for running Oracle on a multicore CPU. ... This means that running Oracle on a server with 2 x dual-core Opterons means ...
    (comp.os.linux.hardware)
  • Re: Comparing Opteron to Xeon
    ... Andrew Gideon wrote: ... > I'm looking at two different motherboards, one of which has dual Xeons ... > To make matters more complex, BTW, the Opterons are dual core. ... The Xeon won't see the light of day in a direkt comparison. ...
    (comp.os.linux.hardware)