[Beowulf] [gregory.brittelle@kirtland.af.mil: Re: Intel?]

Robert G. Brown rgb at phy.duke.edu
Wed Jun 8 18:25:53 PDT 2005


On Thu, 9 Jun 2005, Eugen Leitl wrote:

Said a bit more forcefully than I would have, and he ignores the advent
of L3 caches and the fact that multicore CPUs are still in their infancy
and will doubtless improve quickly no matter who makes them, but I tend
to agree about the hyperthreading vs hypertransport issue >>for HPC<<.
HA is a different matter -- maybe.  I remain unconvinced about the
virtue of constructing multiple virtual pipelines out of one that is, by
chance, too long to be efficient for most code.  Intel's HT could come
into its own with multicores, though.

I think it is safer to rant less and refer to the present more -- at the
MOMENT I think that AMD has an obviously superior HPC architecture, both
multicore and otherwise.  Hypertransport is also a real >>change<< in
the way PC-COTS motherboards are put together -- as big a change as I
can recall since the PC was invented.  It is sort of like a
supercomputer architecture being offered up in the COTS marketplace,
strange as that sounds.

IBM's CELL, though, I still am very uncertain about, partly because I
haven't been able to find any real details about just how it will talk
to main memory and peripherals.  Inside the multicore CPU, it is very
cool and could be very effective indeed at suitable vectorized code, but
I have a lot of unanswered questions, like how fast can it talk to
memory relative to its internal peak speeds, how fast can it interact
with peripherals ditto, can it function in an SMP configuration or is it
strictly 1 (multi)CPU per system.  For AMD the answers seem to be a bit
clearer, although if anybody knows the answers outside IBM they are
likely to be on this list somewhere;-)

The final remark to make is to reiterate that I don't think Apple is
going to be any farms on anything -- at least not without hedging them
signficantly.  If they port to a "custom" Intel-based platform, I really
thing they are a firmware rewrite away from being able to run on vanilla
64 bit x86 hardware from Intel OR AMD, although to run "optimally" might
require more energy, sure.  They've been driven here by necessity, there
are diehards in the organization that are going to try to preserve their
existing modus operandi in the market in the new model and that are
currently dictating policy.

They will fail (no good pundit should be afraid to make bold
predictions:-).  They will even fail >>quickly<< as customers realize
that Apple is now a Dell competitor, not a Microsoft/Intel/AMD
competitor -- Mark already pointed out that they are running head to
head against Dell in a lot of markets selling peripherals, handheld
electronics, etc.  In our nearby supermall, Apple's store is literally
60 yards away from Dell's kiosk, both selling pretty much the same
thing; Apple "high rent" with lots of glitz and neon and prices to
match; Dell in a booth (literally) but cheaper and with plenty nifty
hardware on site.  When they're both running the same CPU at the same
speeds, when one can get a Dell with linux or windows or an Apple with
MacOS (both basically unix plus a windowing system) and the hardware --
especially that all-important "CPU clock" is basically the same -- how
can Apple charge that premium that pays the rent on a warehouse sized
store in the hottest shopping center around?

I don't think they can, but I >>do<< think that they can do Dell's thing
as well as Dell can, and Dell makes plenty of money.  For Apple the real
challenge is to transition to some sort of sustainable model, hardware
and/or software, before the remnants of their "branding" is spent and
nobody cares anymore.

Really interesting, actually.

As I understood it, Apple was still not being clear about whether or not
they were going to do a CELL computer.  Does the latest announcement
clarify that?  Are they going to ONLY Intel, or is CELL still a
possibility?

    rgb

> ----- Forwarded message from Gregory Brittelle <gregory.brittelle at kirtland.af.mil> -----
> 
> From: Gregory Brittelle <gregory.brittelle at kirtland.af.mil>
> Date: Wed, 8 Jun 2005 11:21:04 -0600
> To: Apple Scitech Mailing List <scitech at lists.apple.com>
> Subject: Re: Intel?
> X-Mailer: Apple Mail (2.730)
> 
> Hello,
> 
> I cannot believe the things I am reading. I understand that Apple  
> Engineers are first Apple employees, but trying to spin this as  
> "everything will be OK" is crap. The Intel architecture is simply  
> inferior. I read a posting that pointed out Steve Jobs never said  
> that Intel was superior to the PowerPC/G5 but rather Intel can can  
> deliver the power and speed Apple demands. In what universe has Intel  
> EVER delivered on such promises? Perhaps an orthogonal universe where  
> dual-core doesn't necessarily mean on the same die and low-power  
> means 100+ Watts. Intel bet the farm on hyper-threading (which is a  
> complete joke) while IBM and AMD took the hyper-transport path. I  
> recall (with a certain fondness, I might add) Dell crying over the  
> Spec2000 benchmarks because "they looked at the makefile and hyper- 
> threading was disabled." Anyone remember the response? Hyper- 
> threading was disabled because the Dell machine ran SLOWER with it  
> enabled!
> 
> Steve made it clear that it's not all about speed. Perhaps this isn't  
> about speed to Steve, but to those of us interested in high  
> performance computing, it is about speed. The G5 cluster with the  
> velocity engine was an excellent solution - you cannot say the same  
> about Intel clusters. The Intel L1 cache sizes are pathetic as are  
> the L2 sizes unless one can bring to bear the resources of the seven  
> richest kings of Europe...and I for one don't want a Frinkiac-7.
> 
> If x86 is the path Apple must go, why not AMD? Kazushige Goto and the  
> FLAME project at UT-Austin have produced libraries that SMOKE the  
> Intel Math Kernel. By the way, wasn't Intel the company that didn't  
> seem to understand IEEE-754 standards from the very first floating- 
> point challenged Pentium?
> 
> Good thing I spent all that time mastering AltiVec...I can now swap  
> stories with all those x86 Assembly programmers at the Orphaned Code  
> Expo.
> 
> Cheers,
> Greg
> 
> ------------------------------------------------------------
>    Gregory E. Brittelle
>    Boeing Integrated Defense Systems
> ------------------------------------------------------------
> "Don't worry about people stealing your ideas. If your ideas are any
>  good, you'll have to ram them down people's throats."  -Howard Aiken
> ------------------------------------------------------------
> 
> 
>  _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Scitech mailing list      (Scitech at lists.apple.com)
> Help/Unsubscribe/Update your Subscription:
> http://lists.apple.com/mailman/options/scitech/eugen%40leitl.org
> 
> This email sent to eugen at leitl.org
> 
> ----- End forwarded message -----
> 

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list