AMD Says It's 'Ambidextrous,' Hints It May Offer ARM Chips

J. Dzhugashvili writes "Today at its Financial Analyst Day, AMD made statements that strongly suggest it plans to offer ARM-based chips alongside its x86 CPUs and APUs. According to coverage of the event, top executives including CEO Rory Read talked up an 'ambidextrous' approach to instruction-set architectures. One executive went even further: 'She said AMD will not be "religious" about architectures and touted AMD's "flexibility" as one of its key strategic advantages for the future.' The roadmaps the execs showed focused on x86 offerings, but it seems AMD is overtly setting the stage for a collaboration with ARM."

Re:let's hope that...

[the linux geek]

It's a tough question. The Intel Atom has an edge on ARM, but it's not a big one, and while a high-performance ARM chip costs below $20, the Atom is significantly more. On the other hand, right now there are no ARM implementations that are really competitive on the PC front, and probably won't be until ARMv8 (64-bit) chips, or at least until Cortex-A15. A15 chips will probably come out in late 2012 and be a bit faster than the Atom, but a long way from Sandy Bridge and the other current Intel designs.

Re:PowerPC

[the linux geek]

The most powerful general-purpose processor in the world (Power7) is a huge seller for IBM, and is a PowerPC implementation. PPC is also big in telecom applications, and Freescale does a number of fairly high-performance designs for that market.

The PPC used in the AmigaOne X1000 is a PA Semi PA6T - not very fast, designed as a low-power chip, and long-dead. Apple bought the company a few years ago, and I'm pretty sure new PA6T's are not being made. I suppose that speaks volumes about how many X1000's they reasonably expect to sell...

Re:Could we have a hybrid?

[scheme]

That's tough enough to do when all the processors use the same instruction set, but if the system has processors with different instruction sets, it makes it much harder to have the OS/system switch from a lower powered mode where it's running on the ARM processors to a high performance mode where it's running on the x86 processors. It's not impossible, it's just very complicated and I don't see companies lining up to do the work to implement something like that.

Re:sub-45nm ARM?

[Btarlinian]

AMD lost its fabs a while ago. (Their fabs are part of GlobalFoundries now, and they're a bit ahead of TSMC, but not anywhere close to Intel in terms of process capabilities.)

Re:let's hope that...

[Guspaz]

The price tag is directly comparable, because ARM doesn't make processors, they sell licenses to designs. The only relevant metric is really performance at a given power point.

The closest competitor is Intel's Atom chips. At comparable power points, the current ARM chips seem to substantially outperform Atom chips, and the ARM chips scale far lower than Intel's do. It becomes a bit murkier at higher power levels, since until recently nobody was really making ARM chips that high, but we'll see a lot more competition in this field in the future with the ARM Cortex A15, which is intended to be a lot more scalable. The current design is planned to go from 1.0GHz single-core, up to 2.5GHz eight-core, depending on what the integrator wants. On top of that, they've got the new Cortex A7 that they've designed as an ultra-lower performance chip, which is intended to be a much simpler architecture that's still ISA-compatible with the A15. The intention is actually to put an A7 and A15 in the same SoC, so that the SoC can entirely turn off the A15 cores when only low performance is needed (like playing audio or video, since that's done almost entirely on a DSP). This is similar to what nVidia did with the Tegra 3, just taken even farther.

Re:let's hope that...

[Andy Dodd]

Much of this is a change of focus... Instead of beefy desktop CPUs running bloated OS, the focus is becoming more on portable devices.

Basically, this is "We're hanging in there in the desktop/laptop market, but rather than hang on to our piece of a shrinking pie, we want to get in on the pie that's getting bigger".

ARM is superior in low-power applications. It's highest-end CPUs maybe match Intel Atom, but often have far more peripherals (such as a fairly decent GPU and 1080p multi-format video decoding all on a tiny chip about the size of your thumbnail. Seriously - I can almost completely cover an OMAP4 with my thumb.)

Re:PowerPC

[the linux geek]

Power7 is fully compatible with the PPC 2.06 spec. How is it not a PPC ISA implementation?

Re:Where does AMD come into the picture?

[petermgreen]

AIUI ARM do HDL design of processor cores, then they pass that HDL on to other companies who make complete chip designs based on it. Those companies in turn pass the designs onto fabs (which may be in-house or external) for manufacture. IIRC some vendors also do their own HDL work and only license the basic architectural design from ARM.

Re:StrongARM

[lostmongoose]

You're confusing AMD and Intel. StrongARM was bought by Intel not AMD.

Competitor for Tegra?

[Tapewolf]

The Tegra is basically an ARM SoC with an nVidia video system. Maybe they're looking at doing an ARM SoC with the ATI video core...

Re:let's hope that...

[Anonymous Coward]

Its also worth noting that ARM has never been about performance until the semi-recent smart phone (mobile computing) surge. And even today, performance takes a backseat to power consumption.

It was a long time ago [wikipedia.org], but not "never", when ARM was about performance and running circles around the 80286 and 68000 CPUs.

Re:As a complete NOOB on the subject...

[Koen Lefever]

Or maybe all CPUs today are very generalized RISCy architectures with everything taken care of in microcode (or maybe nowadays it's nanocode)? That would make it (comparatively) really easy to do, right?

Sounds like you are reinventing the Crusoe processor [wikipedia.org].