News - Firm aims to have 80 per cent efficient x86 to ARM in 2 years

[HEXUS]

Emulation progresses and, with superior performance per watt, there's serious appeal.

Read more.

[SciFi]

Makes me want to dust of the Archimedes with pride :-) Go Acorn !

[shaithis]

Wow, I am surprised at the numbers they are quoting. Emulation with 60% of the performance of the same generation.....that is pretty much unheard of.

[Brewster0101]

Didn't a company called Cyrix try this back in the 90s ?

Wonder if Intel will be reading these reports with very nervous eyes.

[Bagpuss]

Without doubt, from a purely aesthetic perspective, ARM processors are THE best looking CPU's out there..)

[shaithis]

Didn't a company called Cyrix try this back in the 90s ?

Is that why the Cyrix chips were woeful compared to intel (and the AMD made Intel clones)

[Steve]

IIRC Cyrix did what AMD and Intel later did, which was have an x86 frontend for a processor that did RISC micro-ops. That's how all x86 processors work now and have for quite some time.

Unless Cyrix also provided some emulation software for some other processor of theirs...

[kingpotnoodle]

Didn't a company called Cyrix try this back in the 90s ?

Is that why the Cyrix chips were woeful compared to intel (and the AMD made Intel clones)

Initially they were OK, good in some areas (productivity) but pretty bad at FP intensive games in the pre-3D era before it all got offloaded to the graphics card.

Cyrix merged with Nat Semi and then got sold on to VIA/Centaur, their MediaGX x86 SoC line was retained at Nat Semi and then sold on to AMD to become Geode.

Don't remember them doing any emulation, their designs were 5x86 and 6x86 IIRC.

[ET3D]

There was one company which did emulation, was it Centaur?

Anyway, I wonder how far they can get with it. I'm sure they'll be stepping on some Intel IP if they want to implement a full modern x86 CPU.

[wasabi]

Transmeta Crusoe is what you're thinking of.

Never come across emulation software that performs remotely well in anything despite massively optimistic claims. Hope to be proven wrong.

[Scribe]

Wow, I am surprised at the numbers they are quoting. Emulation with 60% of the performance of the same generation.....that is pretty much unheard of.

I think the key here is that ARM is RISC vs x86 CISC. It's difficult to optimise emulation of a RISC system on CISC as the latter relies on its advanced instructions to achieve the performance we've come to expect and it's difficult to detect which simple instructions would be suitable to convert to more complex ones and often not possible. The other way around, however, complex x86 instructions can be decomposed into simpler instructions with greater ease.

[d032sh]

Comma explosion

[Noli]

Interesting. If it works and outperforms, I find it hard to believe that Intel could have missed such a big trick all these years. I had read that the although RISC is often faster than CISC that in many ways Intel had denied that myth as its CISC chips were faster for the tasks at hand.

Even if RISC is more usually efficient (per watt), you can not discount Intel's huge improvements in driving down power requirements over the last few years and anandtech and others often cite that there atom / clovertrail developments should be competitive with ARM. Just look at the medfield Lolo phone they brought out - it wasn't wanting in terms of power or battery life vs the incumbent ARM competition.

[anubis44]

Didn't a company called Cyrix try this back in the 90s ?

Wonder if Intel will be reading these reports with very nervous eyes.

I believe you may be thinking of NexGen, which had a processor Nx686 which used RISC instructions at its core which were translated into x86. AMD bought NexGen in 1996 and the Nx686 became the AMD K6, K6-2 and K6-3 processors. I had a 450MHz K6-2, and it was quite good compared to the Pentium II at the time.