Intrinsity Claims 2.2 Ghz Chip 308
PowerMacDaddy writes "Over at SiliconValley.com there's an article about an Ausin, TX startup named Intrinsity that has unveiled a new chip that utilizes a new logic process with conventional fab processes to acheive a 2.2GHz clock rate. The company is headed by former Texas Instruments and Apple Computer microprocessor developer Paul Nixon. The real question is, is this all FUD, will the real-world performance be part of The Megahertz Myth, or is this thing for real?"
MHz (Score:5, Informative)
It doesn't matter if it is real or vapour, it will still fall prey to the "Megahertz Myth". Maybe someday, people will understand: non-similar architectures can't be compared by MHz alone. And even most similar arch's can't be compared via MHz, as the Intel v. AMD war will tell you.
It is even worse than that! no single metric will ever give you the whole story.
Re:Why is everything non-Apple a myth? (Score:3, Informative)
Good theory. And it is what Be said.
Do you know how long it took for the PPC Linux developers to get the Linux kernel running on the new G3 machine? About 2 weeks. How many people work on the PPC specific parts of the Linux kernel? About 2 or 3. I can only guess how many software engineers worked at Be at the time, but I imagine more than 2 or 3. So, how stupid do you think people are? Be didn't get BeOS running on the G3 because -THEY DIDN'T WANT TO- just as Elwood said in a parent post to this. The fact that they lied and whined that it was Apple's fault made me lose a great deal of respect for them.
I'd also like to point out that Apple is a HARDWARE VENDER. Do you think Apple makes money selling MacOS X for err, $89 or so? Of course not. It's a loss-leader to get people to buy their hardware which has a higher markup than most consumer PC hardware. People have been talking for years about how Apple should give up on hardware and moving to software. It won't happen. Apple losing control over their hardware platform would greatly reduce the added value that their products give over consumer PCs.
A more technical article is available at... (Score:4, Informative)
Dynamic logic is nothing new .... (Score:3, Informative)
Given that net delays are becoming the gating factor in big chip designs dynamic logic seems to me to just be a sideshow - unless the long wires are themselves the dynamic nodes (transmission lines with solitons moving on them?) now that would be interesting ...
Potentially much more interesting IMHO is clockless asynchronous logic - but CAD tools just aren't up to supporting this methodology (oh yeah and the synchronous clock based mindset is pretty entrenched too).
Re:Just a guess... (Score:3, Informative)
There really is some intelligence and talent working for this company, I'd like to see what they can produce. Maybe in a few months, if there's no decent benchmarks (by that time, someone somewhere should have written code to use their logic, right?), then I'll jump on the "it's a myth" bandwagon, but I'm willing to give them a chance first.
Re:Why is everything non-Apple a myth? (Score:4, Informative)
Saying that their OS was running apps slower is kindof silly when it's not preemptively multitasked. If you really wanted to, you could just steal the processor from the OS and never give it back.
And Apple stopped sharing specs because they didn't want harware competition.
That said, Be didn't stop porting because they needed the specs. They didn't need the specs. They stopped porting because they wanted to stop. Perhaps because they wanted to know that Apple would support them in the future, but whatever.
Depends on a lot of factors (Score:2, Informative)
It's not MHz that determines the speed. It's just one of them. The rest would be:
And many more. If you have learnt Computer Architecture, then you'd certainly able to list hundreds more.
Moreover, Apple wants to play catchup [theregister.co.uk] with x86... Hmmm... Do you smell something fishy?
Re:What is dynamic logic? (Score:5, Informative)
Both dynamic and static logic use logic gates or blocks that are wired together. The difference is in how the gates are implemented internally, and how they pass data back and forth.
CMOS is a good example of static logic. It uses pull-up and pull-down transistor networks to make sure that outputs are always strongly asserted. This makes CMOS gates big and makes input capacitance larger than it otherwise needs to be. But, it's well-understood, has a few attractive features, and has a whole slew of design tools built for it.
Precharge logic is a good example of dynamic logic. It uses the parasitic capacitance of the output line to store the output value. The output node is charged up on one half of the clock (precharge phase), and left floating on the other half (readout phase). During the readout phase, the inputs are asserted. Inputs are fed into a pull-down transistor network that drives the output low if it should be low, and leaves it alone if it should be high. This style of logic takes up half the space of CMOS logic, has half the input capacitance, and has stronger driving capability (NFETs pulling down typically drive 2x-3x more strongly than PFETs pulling up). This means that if you play your cards right, you can make precharge logic circuits that are faster *and* more compact than CMOS logic circuits. The downsides are that designing and verifying precharge logic is a royal pain, and that you have to have a clock input into the logic block.
The article describes a more complicated dynamic logic scheme with a four-phase clock. These kinds of schemes have been floating around in research literature for years, but are usually not used because of the greater complexity and fewer tools available.
I used to work there, when they were called EVSX (Score:5, Informative)
They were the Austin branch of a company called Exponential Tech. Doing a google on that should bring you up to speed on the Apple connection. I wouldn't really consider them a startup as they've been around for several years and have designed a number of very popular things (e.g. DSPs for other chip manufacturers).
They were a great bunch to work for, especially for being kind to a rather wet-behind-the-ears sysadmin like I was. The only downside to working there was the gawd-awful commute I had to do from far NE Austin to far SW Austin. (If you're an EE type who'd like to live in Austin, they'd IMHO be a great place to work [intrinsity.com]for)
For embedded systems (Score:2, Informative)
1) This is old news. You can find a much better story [eetimes.com] from yesterday over at the EETimes.
2) This is for embedded systems and is not really relevent for PC based systems.
3) This isn't even taped out yet... matter of fact they are not even planning to have the design done for another 18 months... it is vapour until you can actually buy it and that isn't slated until sometime in 2003.
4) This might give Transmeta a serious run for its money if it is ever produced, because they are both in the same space... Of course, TMTA being still around in 2003 is a bit on the presumptious side.
5) Oh never mind, why do I even bother...
Re:Why is everything non-Apple a myth? (Score:3, Informative)
There seems to be some confusion. SPARC, Athlon, Alpha, and Itanium are not faster performers than P4 (except the Itanium which beats P4 at FP).
Let's have a look:
P4/1.8GHz: SPECint - 574, SPECfp - 618
Athlon/1.4GHz: SPECint - 495, SPECfp - 426
Alpha/1001MHz: SPECint - 561, SPECfp - 585
SPARC/900MHz: SPECint - 439, SPECfp - 439
Itanium/800MH: SPECint - 314, SPECfp - 655