Pssst...Subscribe to Our Free Newsletter  
 

THE LATEST NEWS
Wednesday, January 19, 2005

Priming the Intel Pump
Not that I'm up to here in Mactrogen after these last 10 days, but I am. I haven't read anything about Intel's dual/multi core processors in a while so I figured it was about time I wrote something definitive. Here's the deal...

The second half of 2005 will see the ramp-up of Intel's family of dual core (and multi-core, Intel suggests), processors. The buzz here is called "thread level parallelism" and you can run that through the crapometer if you want because it's much easier to say that Intel wants to get two (or more) things done at the same time. Intel hasn't laid down the nitty or the gritty yet so while HyperThreading will also be part of the multicore equation, they can't (so far) say whether or not it will be implemented on each core but it will be there. More news on that as it becomes available.

Why do you want that? A while ago (and by that I mean we were standing in the shade under the belly of a Brontosaurus), an old friend, Bob Hotto, and I were contemplating the properties one needed to make a computer most excellent. Although we had a variety of ancillary thoughts, multiple processors was the one quick and sure point of agreement. Whatever you wanted the computer to do, you had a CPU to handle that operation. In those days it was very simple: the vast majority of PC stuff was internally I/O derived with graphics the hold out. GPUs took the latter problem away.

These days, we're still I/O dependent to some degree, especially with WiFi, but there's a lot more overhead involved -managing the WiFi connection, running anti-spam/spyware software, indexing files on your hard drives, etc., etc. Even with HT, trying to get all of that done at once in a single processor environment is a drag, literally. (IBM's big claim for OS/2 as a multitasking operating system was that you could format a floppy disk while downloading your e-mail... I was impressed.) But imagine if you had a CPU core dedicated to each of those tasks? Assuming the memory management is handled correctly, you'd never know about those background tasks.

If you believe Intel's claims for HT operations (and you shouldn't, they're wildly optimistic even though there are noticeable gains with HT v. non-HT systems), adding the physicality of multicores would create an environment even Apple's Mini Mac could boast about --without all the disadvantages of on-die cross-talk and boodles of heat. If you apply it to Intel's Digital Home concept, you could be ripping CDs, burning DVDs, checking your mail, paying your bills electronically --all while doing whatever it is you'd be doing in a Digital Home. Parallelism in the Digital Home is a hoot. (As much as the Digital Home concept itself might honestly end up as a hoot....)

As I said, if you believe Intel's timeline, both mobile and desktop dual core processors should be out this year (90nm). Think about it... Finally buying a portable computer that performs like its (single core) desktop cousin! Volume production is expected to hit 70+% by 2006 (65nm).

   10 comments      Email this Link


Comments on this Item:
 
The way I understand it it takes roughly five years to go from design concept to sellable silicon for CPUs. Intel demonstrated this with the P4 by releasing it early and then spending the next two years implementing all of the features the original design was to incorporate.

Now we're told that they'll have a dual core product ready in less than two years. Cough%#2t!

What's interesting is where Bill is getting this information. When I check out the roadmap on Intel's website it only runs through June '05 and there's no mention of dual core products. I do a search and I get Paul Otelini talking it (dual core) up but no design info, no nothing.

I go to AMD's website and it's all there: roadmap, design info, history of development, I've even seen pictures floating around the web showing the core design of older Opterons with the outline for where the second core will go.

So why talk about Intel's dual core offering and not AMD's? Clearly AMD has been working on this longer than Intel given the level of current progress. If you want to shed some light it would be interesting to know when the arm went out the window for the "right hand turn" that Intel is taking. Was it when they realized that there would be no 4.0Ghz P4?



 
Does it need priming?


 
Since I have a true dual processor workstation (or two) I have to agree. To an extent.

Unfortunately Windows, and applications have a tendency to allocate stuff to the wrong CPU, and it's not cost effective when it comes to memory use to move stuff to another CPU (in my case, I have memory per CPU).

So, you get more performance than you would with one CPU, if the software you are using is written well.

So far virtually none is, but if you use task manager to set the affinity to applications you can happily be encoding an MP3/movie and zipping/raring a directory on one CPU, and watching a DVD at the same time and still be able to load outlook and check your mail, and run azureus downloading a torrent, with it playing in the background with no stuttering or other problem. Fail to set the afinity of the processes, and you can end up with stuttering on the DVD when you start loading large applications and stuff, although it's pretty hard to make a DVD jitter whatever you do using Opteron CPU's.



 
So many Todds, so little time...

As you might suspect of someone with my infamous past, I have contacts at Intel et al. I was given the time line (actually a PDF version of a very nice Power Point Presentation) yesterday. Don't hate me because I'm connected.

AMD will come in time. You're like a little kid. "Okay, I got the milk already. Where's the chocolate syrup!" Be patient.

Yes, both multiprocessor and multicore technology depend heavily on smart software. Ain't nothing you can do about it.



 
By the way, that little CPU in the piece is the Yoda, er, I mean, Yonah processor.


 
I would agree, most software isn't written properly:
1) Should be multithreaded, when it is not
2) Should not be multithreaded, when it is
3) Or, poorly multithreaded
4) Or, "almost" multithread, software works great until you actually run it on a dual-processor machine (not hyperthreaded).

I forsee an excellent future for those who actually know how to do proper multithreaded programming.

My other comment: Windows XP does a fine job with processor affinity as it is. Could it be that the apps themselves are choosing an affinity?

Another thing could be if your apps are using I/O that requires frequent help from the CPU?

I haven't had any problems with XP and dual-Xeon boxes (each hyperthreaded). XP see 4 logical CPUs and (supposedly) has the smarts to favor scheduling of threads on the "real" other processor over scheduling on a logical one: for example, if you CPU 1 is busy, then some threads should starting moving to CPU 3 (if they do a lot of work) instead of CPU 2, which is on the physical CPU 1.

Anyway, could be wrong, it could be that I simply haven't tested Windows XP "hard enough".

-Anonymous Todd 42



 
I should have mentioned...
One real reason apps would explicitly be choosing a process affinity is to avoid problems on SMP systems. This is the *real quick* fix.

And yet another question, does anyone have good data on the differences between multicore and multi-chip systems? I suppose this will depend on what Intel, AMD and others call "multi-core".

Another last point: Intel's docs on HyperThreading are much more realist than their marketing lit:
http://www.intel.com/business/bss/products/hyperthreading/server/index.htm
This mentions " benchmark tests show some server applications can experience a 30 percent gain in performance", much different than the marketing graphic!

-Anonymous Todd 42



 
I should have mentioned...
One real reason apps would explicitly be choosing a process affinity is to avoid problems on SMP systems. This is the *real quick* fix.

And yet another question, does anyone have good data on the differences between multicore and multi-chip systems? I suppose this will depend on what Intel, AMD and others call "multi-core".

Another last point: Intel's docs on HyperThreading are much more realist than their marketing lit:
http://www.intel.com/business/bss/products/hyperthreading/server/index.htm
This mentions " benchmark tests show some server applications can experience a 30 percent gain in performance", much different than the marketing graphic!

-Anonymous Todd 42



 
So is a NDA keeping you from showing the .PPS?


 
I've only shown my pps to a very select group of people.


Post a Comment

 

 
 

The RealTechNews Official Collection of Interesting Technical Websites
(In Alphabetical Order)

RealTechNews.com -Hooray!

All About Technology
Apolemia
As the Apple Turns
Adam Bosthworth
Channel 9
Cincom Smalltalk Blog
CodeStore
CompHobby.org
CreativeBits
Cult of Mac
Daily Dose of Excel
Dan Bricklin's Log
Dan Gilmore
eHomeUpgrade
Engadget
Enterprise System Spectator
Fozbaca.org
Fullasagoog
Future Now
Gadgeteer
Gadgetopia
Gadgetryblog

Gemal's Psyched Blog
Geomblog
Gizmo
Gizmodo
 

Good Morning Silicon Valley
Google Blog
Google Weblog
Hack a Day
Hack the Planet
Hackdiary
Impact Lab
Internet Alchemy
I4U
IT Facts
Java.blogs
Joel on Software
Jonathan's Blog
Live Digitally
Lynch, Kevin
Matt Heerema
Mavromatic
Mehack
MobileMag
MobileWhack
Mobitopia
MSNsearch's WebLog
Napsterization.org
Onlineblog.com
PatrickWeb
Paul's Time Sink
Picturephoning.com
Player Blog
Ployer Technology News
 
PVRblog
Release 4.0
RFID Privacy
Scripting News
Scriptygoddess
Search Engine Watch
Shiny Shiny
SiliconBeat
Six Apart
Slashdot
SpaceNews
Swaine's World
Tech Digest
Techdirt
Threadwatch.org
Tip of the Day
UberGizmo
Unofficial Google Weblog
Unofficial Yahoo Weblog
Useit.com
Web-Graphics
What Ralph Knows
Wi-Fi Networking News
Wingedpig.com
Wohl, Amy
Wrist Dreams
Yahoo! Search Blog 
 

 

Don't See Your Favorite Tech Site Here? Contact Us to Add it Today!

 

(Please!!)
 
Subscribe with Bloglines

Hey Gang! Please sign our guestbook and say hello to the whole RealTech community: independent tech lovers like you. Take a sec to say hello. -- Sign it!

Please help us stay independent. Donate whatever you can today. (Even $1 will make a HUGE difference.)


 


Email the Editors

RealTechNews.com

Google
Search Our Website:
Web RealTechNews.com


Hosted by: Dreamhost
Underground Networks, Inc. Copyright 2005
All Rights Reserved

Place a Text Ad on RealTechNews

 
 
 
Sign up for PayPal and start accepting credit card payments instantly.