Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
AMD announces G-series embedded SoCs; hints at Kabini, PS4, Xbox 720 specs
AMD announced five new chips today.
http://www.amd.com/us/press-releases/Pages/amd-embedded-gseries-2013apr23.aspx
You're not going to buy an embedded chip for consumer use, so you're not going to buy one of these. However, the dies are exactly the same as in the upcoming Kabini and Temash chips that you might buy if you want a Windows-based ultraportable laptop or tablet. Clock speeds and TDPs about these chips give you information about what will be available soon.
Some embedded chips aren't that sensitive to power consumption (e.g., the difference between 10 W and 15 W doesn't matter much) and don't need that much performance, so AMD gets rid of a lot of defective dies here. That accounts for the bulk of their SKUs in the previous generation G-series APUs. But they did sell some good chips, too, and it's pretty obvious which are the good chips that AMD has announced today.
There are two that are interesting, really. Both have four Jaguar cores and (presumably, though AMD didn't explicitly state this) two GCN CUs. First is the GX-420CA, which clocks the CPU at 2 GHz, the GPU at 600 MHz, and has a TDP of 25 W. Second is the GX-415GA, which clocks the CPU at 1.5 GHz, the GPU at 500 MHz, and has a TDP of 15 W. The GFLOPS numbers on the former forces it to be at least 2 GCN CUs, and I'd assume that the latter is, too.
The former looks like they simply took Jaguar cores and clocked them as high as they'd go while being comfortable that it can be the stock speed for a major SKU. But that tells us pretty plainly that Jaguar cores can do 2 GHz reliably enough that AMD has no qualms about selling chips with 2 GHz as the stock speed. The PS4 and Xbox 720 may clock the cores lower or disable a core or two for reasons of yields. But they're not that power sensitive, as you can readily dissipate 100 W in a console form factor if you care to, so I wouldn't expect to see them clock the cores much lower than 2 GHz.
The former is also what you'd get for a nettop, as that's a form factor where you can readily dissipate 100 W if you care to, so 25 W isn't much.
The latter, meanwhile, is what you'd likely want in an ultraportable laptop. Hopefully the laptop version of it will have some dynamic turbo so that it can turn off a couple of cores and clock the others up to 2 GHz when you need better single-threaded performance.
Notably missing is any low wattage part that isn't completely castrated. I wonder if AMD is holding back all of the ULV-binned parts for Temash.
AMD also says that prices range from $49-$72. Expect prices on the two parts listed above to either both be $72 exactly or one to be $72 and the other very close to $72.
Comments
Kabini and Temash only have a single memory channel. It does allow up to 1600 MHz DDR3, but still, only one channel. So no, an OEM who wants to give it a ton of memory bandwidth can't. But on the bright side, at least it means that OEMs won't ship it with mismatched channels or a memory channel left vacant.
Still, look at how the GX-415 GA compares to the E-350 which was the analogous thing on the previous process node. The GX-415 GA offers:
1) Twice as many CPU cores,
2) Better performance per CPU core (Jaguar cores are substantially better than Bobcat cores),
3) GCN CUs instead of VLIW5 SIMD Engines, which means better graphics performance at a given clock speed,
4) Higher graphics clock speed,
5) 2 CUs rather than just 1 SIMD engine,
6) An SoC, so additional functionality integrated into the chip rather than also having another chipset drawing power,
7) Higher clocked memory, albeit still only one channel, and
8) A substantially lower TDP: 15 W rather than 18 W.
So basically, you get better than double the performance while using less power than before. It's a full node die shrink, but that's a lot bigger gains than you'd expect from a die shrink.
I suspect that much of the gains are because Kabini will be able to use up more of the TDP than Zacate by dynamically clocking things up and down as appropriate. That was the bulk of the gains that AMD got in going from Llano to Trinity on the same process node.