bit-tech.net

AMD launches G-T16R embedded APU

AMD launches G-T16R embedded APU

Systems based around AMD's new G-T16R APU are currently aimed at embedded computing markets, but could an x86 Raspberry Pi-alike be on the cards?

While ARM and Intel play out a very public battle for supremacy in the low-power chip market, AMD has declared that it has no intention of being left out and has officially launched a new entry in its G-Series Accelerated Processing Unit (APU) family with a 4.5W Thermal Design Profile (TDP.)

Intel and ARM continue to chip away at each other's market share in areas where the opponent is strongest: ARM's low-power chips are seeing increasing interest from the cloud computing and microserver market - although Intel isn't taking that lying down - while Intel's system-on-chip (SoC) Atom variants offer an alternative to ARM chips in the smartphone, tablet and embedded computing markets.

AMD, meanwhile, is continuing to push its APU design as an alternative to both. The latest chip in the G-Series family, the G-T16R - aimed firmly at small-form factor and cost-sensitive embedded computing designs - is designed to replace the ageing 500MHz Geode LX 800 chip, which remains a popular choice in embedded computing circles for its low power draw and x86 instruction set.

As an APU, the G-T16R includes both general-purpose and graphics-processing hardware, although it's not a true system-on-chip design: manufacturers looking to build a system around the G-T16R will need to include an A55E chipset for connectivity to the rest of the system. Although this makes the overall design bulkier than one based around a true SoC from the likes of AMD rivals Nvidia or Intel, the new chip still comes in at a fraction of the size of the Geode: overall footprint for APU and chipset is 890 square millimetres, compared to 2,129 square millimetres for the Geode.

AMD's selling point for the G-T16R isn't just a reduced component footprint, however. The company claims that the chip draws seven per cent less power than the Geode - a 2.3W average power draw in a 4.5W TDP - while offering three times the performance as proved by SciMark2 benchmark suite in which the Geode scored 22.05 to the APU's 65.55. It has not, however, confirmed the chip's clock speed.

Traditionally, AMD's embedded product lines have been exceptionally niche products aimed at industrial control and monitoring, medical infrastructure, point-of-sale and digital signage markets. The role of low-power, passively-cooled embedded processors is beginning to change, however: the staggering success of the Raspberry Pi ARM-based single-board computer - which still requires pre-registration to purchase despite two manufacturing partners and round-the-clock production - has indicated a growing demand from users for cheap, compact low-cost computing systems.

Although AMD's hardware partners for the G-T16R - Advantech, Aewin, Arbor, aValue, Axiomtek and MEN Mikro Elektronik - are currently aiming firmly at traditional embedded markets with their designs, it wouldn't surprise us to see a manufacturer taking the G-T16R and producing a Raspberry Pi-alike - but one which boasts support for up to 4GB of DDR3 memory and a full x86 instruction set with full support for running Windows or any modern Linux variant.

10 Comments

Discuss in the forums Reply
Andy Mc 25th June 2012, 14:08 Quote
If someone did put this on a Rpi a like and gave it 2 gig nic's it would be an awesome base for a pfsense box.
Alas that will never happen.
b1candy 25th June 2012, 16:26 Quote
Chip has less power because it's not a SoC. So what's the chipset power draw, eh?
schmidtbag 25th June 2012, 17:18 Quote
Quote:
Originally Posted by b1candy
Chip has less power because it's not a SoC. So what's the chipset power draw, eh?

hmm good point. i'd also like to know how it performs in comparison.
Gareth Halfacree 25th June 2012, 17:36 Quote
Quote:
Originally Posted by b1candy
Chip has less power because it's not a SoC. So what's the chipset power draw, eh?
Eermmm... The Geode isn't an SoC either, which is what AMD is comparing it to...
fluxtatic 26th June 2012, 07:38 Quote
Boioioioing. Do want.
BLC 26th June 2012, 09:59 Quote
Will have to do an awful lot to beat or equal the Pi - like getting a complete SBC out for $35. I doubt the chip will even be that cheap.

To be honest, I think it will take an awful lot to beat the Pi's achievements. Even if a manufacturer somehow manages to get their price down to the same sort of level, they'll have to do a lot of work on the software side to equal the amount of developer attention on the Pi; the foundation themselves might not be able to put out vast amounts of code, but the developers for the Pi are legion.

Comparisons with the Pi aside, it still looks like an impressive achievement. I looked into Geode-based SBCs a while back, but the boards were really expensive. It's good to have the option of an x86-based embedded SBC out there.
Gareth Halfacree 26th June 2012, 10:06 Quote
Quote:
Originally Posted by BLC
Will have to do an awful lot to beat or equal the Pi - like getting a complete SBC out for $35. I doubt the chip will even be that cheap.
Yes and no. I mean, the Pi is unlikely to be beaten on price any time soon - unless someone chooses an even older SoC, or a company takes a major loss somewhere in the chain - but I think people would be willing to pay a premium for x86 support. The ARMv6 instruction set of the current-generation Pi limits it quite a lot: it cuts the number of operating systems available down a great deal, and the effect on the client software is even more pronounced. A Cortex-based version would help, but an x86 equivalent with support for any Linux/BSD/whatever release as well as Windows and a graphics unit with ready-built drivers for 2D and 3D acceleration? Price it anywhere below $100 and it'd sell like hotcakes.
mclean007 26th June 2012, 13:02 Quote
I'm going to be a pedant and object to your use of "mm squared" - either use the superscript (mm2) or use "square millimetres". 890mm squared describes a square with sides of 890mm, i.e. 890mm x 890mm or 792,100mm2, which would be a mighty big chunk of silicon.
Gareth Halfacree 26th June 2012, 13:41 Quote
Quote:
Originally Posted by mclean007
I'm going to be a pedant and object to your use of "mm squared" - either use the superscript (mm2) or use "square millimetres". 890mm squared describes a square with sides of 890mm, i.e. 890mm x 890mm or 792,100mm2, which would be a mighty big chunk of silicon.
And I'm going to be a pedant and point out that the measurements were copied and pasted directly from AMD's press release. If it's good enough for AMD... (But yes, you're right - it should really be square millimetres, not millimetres squared.)
mclean007 26th June 2012, 13:50 Quote
Quote:
Originally Posted by Gareth Halfacree
And I'm going to be a pedant and point out that the measurements were copied and pasted directly from AMD's press release. If it's good enough for AMD...
If it's good enough for AMD...it's still wrong :-D

Maybe AMD has had to cutback on the people who sense check their press releases, because that's a pretty basic, if minor, error. Yes you can divine the meaning with a bit of common sense, but that doesn't really make it ok. If a news report said that an area 100 miles square had been affected by, say, flooding or forest fires, that a massively bigger problem than an area of 100 square miles.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums