He is drinking at the Harrow when he should be at the plough
|
|
|
Secretive startup's new chip might challenge Intel
MONTALVO SYSTEMS, a Silicon Valley upstart chip design firm, is reportedly working on an innovative microprocessor that will be capable of running the same x86 instruction set as chips from Intel and AMD.
Santa Clara, California based Montalvo's new chip will reportedly incorporate multiple cores, but they won't be symmetrical, that is, identical circuitry. Rather, Montalvo's as yet unnamed microprocessor will be an asymmetrical design that contains some full-function, high performance cores along with some other, less capable and lower performance cores, all on the same slab of silicon.
Montalvo expects that its asymmetrical chip design will be able to reduce power consumption by shunting streams of simple instructions to lower performance, less power consuming cores. Conceivably, some applications that enter an idling mode after an initial burst of startup processing might also be moved to lower performance cores, saving power with no discernable impact to user-observable responsiveness.
Asymmetrical cores can also offer higher performance if optimised for certain specialised applications such as video processing, say fans of the architecture.
The Cell processor was the first highly successful example of asymmetrical chip design. It contains one high performance primary processor core, along with a number of "synergistic processing elements" programmable for simpler tasks such as shuttling network traffic or multimedia processing.
The Cell processor is perhaps best known for powering the Sony Playstation 3 games console, but IBM has also put Cell chips into server blades and Toshiba plans to use the Cell processor in TVs and maybe PCs, as Mercury Computer Systems has done.
Montalvo hasn't had any chips fabbed based upon its design yet, but it's said to be looking for applications in portable computers, notebooks, handheld devices and smart phones. If it takes on Intel in the x86 chip market, Montalvo will be joining several firms that tried previously and failed, including Cyrix, Transmeta and Rise. But if it does, the competition will no doubt be good for the industry. µ
L'INQ
Cnet
Share this:
Share
There have been asymmetric DSPs for a long time... and they are in all sorts of devices. I imagine there are far more of them than there are Cell processors.
That kind of chip will lose every benchmark except power consumption. How would it challenge AMD or Intel.

It sounds like it is just fitting into a niche market like Transmeta was.
Via got their license from their acquisition of Cyrix.

As for the original article, I will believe such a thing when I can install it :)
I thought it was something new but if the "architecture" has "fans", I guess it isn't, so that means that Intel is already working on their own or has already scrapped the idea. If I were gonna start making x86 chips, I'd want something more revolutionary than that.
Per below, using GFLOPs as the only measurement of a "high performance" chip is a bit misleading. Since when did a computers ability to do solely scientific calculations on its FPU become the measure of all the CPUs performance? From the people I know at this startup - I suspect that this chip wont be sitting in a server cluster spanking away at Linpack to the joy of people watching its GFLOP score.
How will Montalvo acquire a license to build an X86 core? It seems like I remember it being written several times on the Inq that there were only 4 companies with X86 licenses out there. I believe they are Intel, AMD, Via, and one more that I am a little sketchy about. I think this 4th license was browed by Cyrix and that its use is actualy a little sketchy from a legal point of view. I could be wrong about this but isn't the licensing issue one of the big factors that keeps others (i.e. Sun, IBM, Motorola, ect.) from entering the X86 market? I would love or someone to educate me a little on this issue.
Ahti! This shit turns a CPU into something similar to the Cell with all SPE. Even better in fact.
Looks like a good target for a cretin green fables chip maker. That is if it can deliver!
But I bet by choosing an asymmetrical design their manufacturing costs are going through the roof. Doesn't anyone remember Transmeta?
>The Cell processor was the first highly >successful example of asymmetrical chip >design. It contains one high performance >primary processor core, along with a number >of "synergistic processing elements" >programmable for simpler tasks such as >shuttling network traffic or multimedia >processing.

There is nothing high-performance about the primary 3,2GHz PowerPC core, as benchmarks under Linux have demonstrated. It hovers around the performance of a 3GHz Celeron. 6,4 GFLOPS at 100% efficiency is the best it can do, while each of the SPEs, the real heavy lifters, can do 14 - 25 GFLOPS (double- or single precision correspondingly).