Thanks for the tip fehu, I hadn't even heard of Eagle until now, not that surprising given how little ARM appears to have said about it.
I don't to dismiss everything said in the SA article entirely but it's a little hard to take Charlie seriously when he doesn't really cite sources, makes a few incredibly dubious claims, and seems to have something against nVidia (and to a lesser but non-imaginary extent ARM).
This statement is particularly culpable, bordering on offensive:
"On the technical side, the problem is simple, speed. ARM A9 CPUs are great for phone level applications, and can reach into the current tablet space, but hit a glass ceiling there. If Eagle doubles the performance per MHz and doubles performance per watt, it will basically be on par with the low end of the Atom-class CPUs, and woefully behind the Nano/Bobcat level of performance."
The claim that per-MHz Cortex-A9 offers half the performance per MHz and half the perf/watt (ie, given first claim, same watt/MHz, which is also false) is incredible and suggests a severe lack of understanding as to how at least one of the architectures works.
Then the very following sentence:
"On top of that, you strap a translation engine to change the x86 code to ARM instructions, and you lose more speed, add latency, and drop performance."
Which is an attempt to inundate a point by saying the same thing three times.
Comparing this to Transmeta's CPUs seems to be missing the point to me. Crusoe/Efficeon failed because they were for all intents and purposes inferior x86 CPUs, and the power advantage wasn't enough to win in the markets that mattered, notebooks, which at the time had other components which overshadowed the CPU power saving.
ARM is already a proven and highly successful architecture, and with its foray into tablets it's not unforeseeable that ARM can take share in netbooks too. It's highly unlikely that nVidia would push forward an ARM product with x86 compatibility boosting features in hardware but lock it down to x86 execution only via protected code like Transmeta's CPUs were - that'd be insane.
If nVidia is incorporating x86 support it's not in an attempt to beat Intel at their own game, because ARM does a good enough job for the markets it (and nVidia) target. It'd be to give it a competitive advantage against other ARM offerings.
I also think Charlie is overestimating the amount of silicon and especially power consumption needed to aid in some kind of useful translation acceleration. A few very simple techniques can go a tremendous way in enhancing recompilation. Transmeta's CPUs provided a lot of hardware infrastructure because they were doing support for system level emulators. nVidia wouldn't benefit from this, and would actually be foolish to push it, because they don't want people to run Windows - they want people to run environments which can fully leverage the great wealth of say, Android code already out there, while being able to run Windows programs too (ie, qemu on WINE). With some intelligently picked acceleration features exposed in the ISA performance can be brought to a decent level, probably to least 50% native (see Godson 3 for example, which claims 70% with quite modest enhancements - much more could be done). Things can be taken to levels where say, 2GHz Eagles could beat 1.6GHz Atoms. Of course, I say this while knowing nothing about the micro-architectural improvements Eagle makes. I'm just assuming it's not nothing, or something largely throw-away for this like a move to 64-bit.
The market has proven that 1.6GHz Atoms are very successful, and Intel apparently thinks there's also market demand at half the speed. x86 has added value, even at under the speed of current Atoms. You could argue that even pure software emulation is desirable, but I think part of the problem is there has been too little attention given to x86 on high end ARM (especially from corporations). If nVidia pushes x86 features in hardware they'll spark interest in this, especially if they develop their own software to this end, which I fully expect. There'll also be a marketing advantage just by claiming anything to do with x86.
Anyway, this is all assuming nVidia is really going for this. But if they are there's no point wondering how they'll pull the chip off, when IMO the translation would only be one aspect of it and it not being stellar wouldn't necessarily tank the product.
EDIT: However, nVidia can't extend the ARM ISA to include x86 emulation without getting an architectural license and developing their own CPU (ie, no Eagle). It might be possible that ARM is collaborating with them to allow them to do an x86 acceleration coprocessor, but I think that for it to really be effective it has to share the same register file. If Eagle is really an architecture revision then it's possible that ARM themselves are offering variants with enhancements that lend themselves well to emulation. Before anyone says it, these would hardly necessarily be infringing upon x86 - some of the most useful accelerations for emulation are not even remotely x86 specific.
