AMD's CES presentation was heavily APU-focused, although the company also announced OEM availability of desktop Radeon HD 8000 Series GPUs.
UPDATE: The Richland APUs are likely to be 32nm parts, not 28nm as indicated in an earlier draft of this article.
AMD's Consumer Electronics Show presentations this year have been dominated by one initialism, with the company showing off a raft of new Accelerated Processing Units (APUs) as it looks to steal a little market share from Intel in the desktop and laptop and ARM in the tablet spaces.
AMD's announcements covered four product codenames within the APU family: Kabini, Richland, Kaveri and Temash. First, Kabini: designed for ultra-thin notebooks - Intel's Ultrabook segment, in other words - the chips are claimed to offer twice the compute performance of current-generation Brazos 2.0 APUs without sacrificing battery life through the use of new Jaguar processing cores and a true system-on-chip (SoC) design. Firm figures are not yet available, but AMD has confirmed that Kabini will launch in dual- and quad-core flavours.
Richland, meanwhile, is a family of APUs that have started shipping this year to original equipment manufacturers (OEMs) looking to build more traditional laptops and desktops. With a claimed performance boost of 20 to 40 per cent over previous-generation APUs, there's a fair bit of spare power available to the system - and AMD is making full use of that by planning to bundle gesture- and facial-recognition software with Richland-based systems. A follow-up chip family, Kaveri, has been confirmed as launching in the second half of the year, boosting performance still further and adding new heterogeneous systems architecture (HSA) features at a 28nm process node.
Temash, finally, is designed to sit at the very top of AMD's mobile-centric APU line-up. Claimed to be the highest-performance system-on-chip design for tablets, Temash replaces Hondo with double the graphics performance of its predecessor. As with Kabini, itself another true system-on-chip (SoC) design, Temash is expected to launch in dual- and quad-core flavours.
The first of these chips to launch will be the AMD A10-6800K, and while firm figures are not yet available
preliminary figures point to the chip boasting a quad-core design based on a 32nm process with updated Piledriver cores, heterogeneous systems architecture (HSA) improvements and the rumoured possibility of second-generation Graphics Core Next (GCN)-based Radeon HD 8000 integrated graphics in the same 100W thermal design profile (TDP) as its Trinity-based A10-5800K predecessor. A June retail launch is expected, but not yet confirmed by AMD.
AMD also announced the Radeon HD 8000 family of graphics processors,
previously confirmed for laptops and now officially available to OEMs for use in desktop systems as the Radeon HD 8000 Series. Official specifications for the
Radeon HD 8700M & 8800M and
Radeon HD 8500M & 8600M mobile GPUS and
Radeon HD 8000 Series desktop GPUs have now been published, pointing towards a series that tops out with the AMD Radeon HD 8970: a 1GHz (1.05GHz boost) GPU teamed with 3GB of GDDR5 memory on a 1.5GHz clock offering 288GB/s bandwidth, 2,048 stream processors across 32 compute units for 4.3 teraflops of single-precision compute and 1.07 teraflops of double-precision compute, 128 texture units, 128 Z/stencil ROP units, and 32 colour ROP units on a PCI Express 3.0 x16 interface.
Sadly, AMD has been quiet on firm retail availability and pricing for any of its announced products - but to keep you amused while we wait for more information, here's AMD demonstrating the benefits of the gesture-based computing software it plans to bundle with the Richland-generation APUs.
21 Comments
Discuss in the forums ReplyI can see why they sometimes rebrand old parts that fit well (usually at a lower price) into a new series, but why launch an entire line of rebranded parts?
Doesn't make any sense.
Wow... that's... lame... I didn't even like rebrands in the mainstream, but now they are doing it at the top of their line?
They should just keep the HD7970 name, I don't mind seeing the same name for 2 years...
Soooo sluggish.
In case it hasn't been mentioned elsewhere (can't find it). Here are a couple of links relating to the valve announcement...
http://www.bbc.co.uk/news/technology-20949071
http://www.prnewswire.com/news-releases/xi3-and-valve-to-unveil-new-product-at-ces-2013---an-xi3-development-stage-computer-game-system-optimized-for-steam-gameplay-in-big-picture-mode-185945182.html
We've not mentioned it because Valve and Xi3 haven't confirmed anything.
I couldn't care less about the GPUs aswell, so I'm sticking with the mid-level intel + nVidia for the next two years to come I guess, but I don't think I need to upgrade within the next two years anyways.
To be fair it's only an OEM rebrand, consumer products will still be 7xxx series. Basically as product refreshes get less common (longer development cycles from everyone), system builders have nothing new to sell, so request new-sounding parts to justify refreshes or re-releases.
This has been the case in all camps for a while now, though Intel's refresh cycle is currently fast enough for this not to be an issue.
Any news on new GPUs?
Someone will believe it...
"Richland" is 32nm. The follow-on "Kaveri" will be 28nm.
Proof 1 - See in the following image from the presentation, "Richland" is shown as being 32nm, the same as "Trinity".
Proof 2 - AMD says that "Richland" is "currently shipping to OEMs", which means that it has been in production for quite a few months for the wafer to be started and baked, the die packaged, and the finished APU sent to OEMs. It isn't being made at TSMC and GloFo 28nm isn't ready for production. Therefore, logically, it is 32nm.
If "Richland" is A10-6800K, then it is 32nm. If it isn't, then A10-6800K isn't "Richland".
Everyone saying 28nm has been referencing Fudzilla. He seems to be wrong.
If "Richland" is 32nm (see above) and shipping now, it means Piledriver cores and VLIW4 graphics. It is just Piledriver with enhancements.
No Steamroller and no GCN. More Fudzilla-branded FUD.
. . .which got their information from (I think you can guess!) Fudzilla.
Who is also claiming GCN for the core. I'd be really careful with claiming that, because "Richland" is looking like a refined "Trinity" rather than a new core, with the claimed performance increase possibly due to a larger/higher clocked GPU rather than a new GPU.
I would be very surprised at GCN showing up so soon.
EDIT: A bit of hunting and I found the 2013 Client Roadmap (PDF) on the AMD website.
"Richland" is 32nm with 2-4 Piledriver cores and "2nd Generation DX11 GPU", which is as far as I can count is VLIW4 (VLIW5 being the 1st and GCN being the 3rd). Three versions: desktop, notebook (35w) and Low Voltage (17-25w).
This close to release, if it was GCN it would say GCN, as it does for the 28nm Kaveri, Kabini and Temash APUs.
Personally, about the only thing interesting to me is Jaguar update of Brazos - I'm running an E350 as a home server and will be putting a car pc in next summer based on another. Not bad for what it is (and what they cost), but a little more grunt would be appreciated.
http://wccftech.com/amd-radeon-hd-8000-gpus-oems-radeon-hd-7000-rebrands/
At least that's my understanding of the situation, AMD should have really done something regarding the naming however. I'm guessing AMD will just skip to calling Sea Islands the 9000 series, or confuse the matter by running two parallel 8000 series at once.