Pat Gelsinger says chips have to keep getting faster
ON THE EVE of Monday’s Penryn debut appearance, The INQUIRER caught up with Intel Digital Enterprise Group SVP Pat Gelsinger. A 28-year veteran of the company, Gelsinger dates back to 286 and 386 design teams and has held a series of key positions at the company.
We grabbed an hour of his time to catch up on pretty much everything we could squeeze in before Hexus TV did its thing. In that time, we talked about computing formats, GPUs, the relationships with Microsoft and AMD, phones and the true, shocking story behind why INQ editor Mike Magee calls him Kicking Pat.
In this first part, Gelsinger discusses why we should care about yet another new chip…
Having seen four decades come and go, beginning when punk ruled the music scene and Margaret Thatcher was moving into Number 10, and TV consisted of three channels, I asked whether he isn’t a bit jaded by yet another processor generation.
“Some of it gets to be [tiring when] you deal with the same questions for 30 years but it’s the magic of what we do - this business of doubling transistors and putting computing and communications into the hands of every human on the planet. If you take that definition [of the challenge] we have a long way to go.”
Sure, but back in the days of the 386 it was pretty obvious why you wanted the extra CPU cycles. I mean, Windows 3.0 made the 386 a slam dunk, right…
“You think it was a slam dunk?”
Well, you know…
“What a revisionist historian you are! We had people saying, ‘32 bits! That’s a minicomputer, that’s a VAX. Who ever could use that much physical memory?’ The question was the same at the time as we get now. Moore’s Law is reasonably linear. We’re doubling performance of our chips approximately every two years but software follows hardware. The job is to create a vacuum and the available performance jumps up to be the new normal.”
Fair enough, but everyone always asks 'what am I going to do with this new stuff?' Where’s the killer app? (Yes, I did say killer app. Yes, I am ashamed.)
“Well, there was a time when Mitch Kapor wasn’t Mitch Kapor,” Gelsinger says, referring to the Lotus founder who built the biggest PC killer app of the lot, 1-2-3. “We don’t always know what the killer app is but there are a handful of areas that are interesting.”
These include graphics rendering, where he contends “polygonal representations are soon going to disappear” and global illumination and ray tracing will become mainstream. Processor advances will help depict, "how the cloth acts when it’s moving or how the skin stretches across the frame … the behaviour of objects interacting with each other so that when the ball hits the glass, you can model the behaviour of breaking glass".
Another near-term possibility is a new user interface that can interpret not just speech but also gestures. Visualisation, for example in healthcare for real-time MRI imaging, is another area that needs hardware power but Gelsinger says that sometimes you just have to build the hardware and let the software folks get on with it.
"If I knew what the killer app I’d get rich and start a company," he says. “That’s a joke by the way. But if you look at some of the hard problems out there, they require levels of performance we can't yet build." µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
GEL was yaya X derivative in 19x8-, Certainly linear processors that could write check & display image or even send message to another computer WHERE PROMISE. Software was already to migrate from mainframe long long ago, yet public just Couldn't get hardware EVEN FOR ERAS SMALL TACKY PROGRAMS. in fact LOTUS 1,2,3 came rather late in whole bargin, 30+ YEARS AGO CONCEPT WAS DEVELOPING, when BIG$ where foreseen in software, business. Remeber those Spread sheets A6+m121="Print" hahaha.ahhh.

EVERY day lady would dutifly find out new enrollment number by entering box with label for it & machine finds files A6 ,m121, then adds & displays answer. Wow. Complex yet almost so simple operator could cry.HEY-I Wrote Radar Screen & its all over VISTA as smooth shift mechanism between subroutines, not just clock display that went ballistic.Pascuale.
Todays processors are more tuned to plenty of action & thruout entire system, maybe you maka d'barn movies? yes,yes?.

You can always make processor go faster than better designED PROCESSORS, just use older engineering.

QUALITY IS TEDIOUS FEAT. It only makes sense most complex design will run slightly slower than simple design. 

4044 or 4 core its square wave, yet its not same capability at all. 

Hail to those whom opened field, PRAY get this Media Machine done well, as HOW CAN YOU HAVE NT7, WITHOUT PERFECTING NT6? It does little good to squawk about Nt5, XP basic Vista Starter, its OVER. DONE DEAD, COMPLETE, ENDOFPENIS.

I bet that guy knows some people, yet I believe Msr. Ballmer in saying no true 4 core has been made to date & it will be late spring '8 before such items start to be manufacturered.Yet I DO NOT Nessessarily hope they are FASTER, Just BETTER.

Signed:PHYSICIAN THOMAS STEWART VON DRASHEK M.D.
"If I knew what the killer app I’d get rich and start a company," he says. “That’s a joke by the way. 

Thanks for the get rich quick tip, Pat.

I aint no joke when you have the killer app....and you're looking for a company to start a company. The SMARTer route right to the top with XXXXPerience Leading the Ride and Reaping the Profits and Interest paying the Source Handsomely/indecently.

PS,..... That Comment Title? 
More than just a little to do with AI Blighty NEUKlearer V Force. dDNotice Stuff?
Kicking Pat Gelsinger may be a geek of some sort, but he is also a living representative of the bad old paranoid Intel culture. And it shows. The Inq should be very critical when interviewing such a person, not the common shit chat that he is used too. Check the history behind his infamous "Kicking" nickname and editor Mike Magee, his relationship with former Intel lunatic Andy Groove, his attitude on competition ("I hate AMD"), and ofcourse the Itanic that he's currently pushing although nobody wants it. Inq, time to get real and ask the really though questions. Don't let this so-called geek shine, because he doesn't deserve it in my opinion. I don't think your editor do either.
Chips have to keep getting faster so that Microsoft's 'cosa nostra' can keep getting sloppier.
would be an operating system that is reasonably sized, has point and click functionality, is reasonably priced,...........and works!!!
We're fifty years away from having real-time ray-tracing renders to replace existing polygon solutions. Sure, one can imagine processing power catching up to real-time for renders performed fifteen years ago; e.g. Terminator 2 graphics. 

The problem lies within our desire to improve not just the core, but the peripherals--higher resolutions, higher detail, higher lighting requirements, etc. As the capacity to render more quickly pushes on, the requirement continually heaves itself above that mark.

Improving our technology generally in accordance with Moore's Law is absolutely necessary to innovation--that, and nobody can make a buck on technology if it doesn't constantly improve.
A lot of the software he describes might benefit from a CPU without math errors, I wonder if he ever thought about that then.

Not sure about fifty years, remember theres Daniel Pohl has quake3 running at a good rate
Kip, RE: 'long road ahead': you have any science or sources or citations to back up that claim?

David

P.S. Almost everybody else posting above Kip is a nutter. Completely insane. I thought your readership was up to more than this, Inq...
The i386 was introduced in 1985/early-86, while Windows 3.0 came out only in May 1990. Not even Windows 1.0 (introduced 11/85) was around when the i386 taped out in 10/85.

Inq - get your chronology straight!
@ aman from mars 

It´s not true that we are 50 years away from ray-tracing. Infact we are as close as never before. 2003 you needed 50!! Xeon cpus to render 480x480 at 3-4 fps with raytracing. With the next gen Intel Xeon you can render 768x768 at 90 fps WITH ray-tracing. I think that photorealizm will be approached 2012 and all the details of physics and material interaction will be done programmed by 2020. There will just be the question how this will change our world.
@David: Welcome to the internet!
Of course they do - otherwise how could the cpu industry keep the money rolling in? Since the chips dont wear out (at least not quickly enough for the bean counters) then they must be made obsolescent instead.
"Processor advances will help depict, 'how the cloth acts when it’s moving"

It seems obvious that there are many people at Intel who are desperate to see Superman with his cape on a computer screen and believe what they see.
All this talk about polygon rendering concerns games, so I understand that they are searching for the ultimate cape-renderer for City of Heroes.
I, on the other hand, believe that we have enough shiny on the screen already. Look at Crysis if you want proof. I do believe that we need more processor power, but I would really like to see it go to a more realistic environment, and not to calculate just how awesome that cape is supposed to be.
Crysis is on the right track. I hope that it shows the way for other games to follow.