More Details on the Intel Atom Emerge
by Anand Lal Shimpi on March 6, 2008 8:00 PM EST- Posted in
- CPUs
In early 2004 Intel assembled a team of engineers to design a core to be used in a many-core CPU. You may remember the following slide from Spring IDF 2005:
By 2015 Intel expects to have CPUs with many smaller IA cores, each with very low power characteristics but with the entire chip being very high performance. Intel commissioned its Austin design team to work on one of the cores for this type of many-core CPU.
Also in 2004, Intel commissioned a smaller group from within this team to look at the feasibility of turning one of these cores into a standalone CPU for use in low power mobile applications.
The team tasked with the pathfinding effort quickly concluded that based on the performance, power and cost requirements, it could not rely on any of Intel’s existing microprocessors to base such a CPU on.
The CPU that resulted from this pathfinding effort was Silverthorne, more recently given the name Atom. And today Intel is releasing some more details on the processor, as well as its first successor due out in 2009/2010.
More Menlow Details
The Austin team’s pathfinding effort resulted in the design of the Bonnell core, named after the tallest mountain in Austin, TX measuring only 750 feet. Given that Bonnell was a very small core, naming it after a very small mountain seemed fitting.
The Bonnell core was just that, a core, at first. Once Intel told its Austin team to begin turning it into a standalone mobile CPU, the team had to design a cache and bus I/O for the chip. The resulting chip was Silverthorne, which is a Bonnell core + L2 cache + Bus I/O.
Silverthorne, which is the ultra mobile version of the Intel Atom CPU, is built on Intel’s 45nm process as we’ve mentioned in our architecture piece.
Menlow is the name of Intel’s MID (Mobile Internet Device) platform, it consists of the Silverthorne (Atom) processor and the Poulsbo chipset.
Poulsbo is an entirely new chipset design, it’s a single chip solution that features integrated GPU and I/O controller - it obviously plays all of the standard chipset roles, just in a single chip. Poulsbo isn’t 45nm but Intel isn’t disclosing its manufacturing process just yet.
The integrated graphics core in Poulsbo is also a new design, but Intel isn’t disclosing too much about it yet. It will support full hardware accelerated HD video decode.
The Menlow platform will officially launch in Q2 of this year with products expected within 6 months of its launch.
Facebook
Twitter
RSS
12 Comments
View All Comments
psychobriggsy - Friday, March 7, 2008 - link
Through use of Moore’s law allowing for better integration of silicon and smaller, less power hungry transistors, Intel hopes to be at power-parity with ARM (both idle and active) within the next 3 or so years - all while being significantly faster, at least for the next 5.Yeah, yeah, and ARM doesn't benefit from Moore's Law?
Intel hopes to be where ARM SoCs (far more integrated that Mooretown) have been for several years ... in another three. Maybe. Intel will have the process on their side. That's about it.
In the meantime, given the excellent performance of the iPhone on a 90nm 412MHz ARM11 core (with 1GHz+ available, and multi-core as well with Cortex) I don't see any imminent desire for Apple to switch to these offerings for the iPhone and iPod Touch. I think the release of their SDK suggests this as well - sure, they could make everyone port their software in a mere three years ... or not.
I just don't think that even Moorestown++ will be anywhere near what Apple need - not integrated enough, not frugal enough. They're not going to make an iBrick (in comparison).
nubie - Friday, March 7, 2008 - link
Are ARM SOC's just now moving to 65nm? Intel can push because it has the fab capability and IP and talent and resources.The iPhone being x86 means no more seperate Dev on two processor platforms, the iPhone will use the same progs as the desktop OS, no porting will be done, it will open up the handheld to all the companies already making OSX software, a "port" of any desktop app reasonably suited will be very very easy.
I wonder if Intel will allow a separate memory bus for the RAM, if so this would make it truly competitive, even a second DDR3 populated with SoDimm low latency high speed would do wonders for it. I am thinking of the HTPC and desktop markets of course (and the non-existent high performance hand held gaming market)
bebesito21 - Friday, March 7, 2008 - link
Yeah, so by 2015 Intel plans on having this ready? Aren't they a little behind? Seems like IBM already has this plan rolling. The Cell processor already has multiple "children" cpus waiting for a controller CPU to tell them what to do. This is, in my opinion, the best way to design chips because no developer wants to waste time writing programs that take advantage of multiple thread capable CPUS - they rather the hardware figure it out. Its hard enough to use 2 out of 4 cores on a quad core now..imagine 32 cores!I wish they would give us a socket on the motherboards for a GPU chip. Then they can do multi-core cpus and multi-core dedicated gpus with new ultra fast buses to link. Plus I don't want Intel completly taking over all chip making...that would at the least cut out NVDIA who doesn't make CPUs. They backed the PS3 though so they might help push the Cell.
LSnK - Friday, March 7, 2008 - link
The PPE doesn't auto-magically tell the SPEs what to do. It's vastly more difficult to program for the Cell than a standard multicore CPU. It isn't going to save developers any time.Thatguy97 - Tuesday, May 26, 2015 - link
Intel was right lolbebesito21 - Friday, March 7, 2008 - link
Yeah, so by 2015 Intel plans on having this ready? Aren't they a little behind? Seems like IBM already has this plan rolling. The Cell processor already has multiple "children" cpus waiting for a controller CPU to tell them what to do. This is, in my opinion, the best way to design chips because no developer wants to waste time writing programs that take advantage of multiple thread capable CPUS - they rather the hardware figure it out. Its hard enough to use 2 out of 4 cores on a quad core now..imagine 32 cores!I wish they would give us a socket on the motherboards for a GPU chip. Then they can do multi-core cpus and multi-core dedicated gpus with new ultra fast buses to link. Plus I don't want Intel completly taking over all chip making...that would at the least cut out NVDIA who doesn't make CPUs. They backed the PS3 though so they might help push the Cell.
bebesito21 - Friday, March 7, 2008 - link
Yeah, so by 2015 Intel plans on having this ready? Aren't they a little behind? Seems like IBM already has this plan rolling. The Cell processor already has multiple "children" cpus waiting for a controller CPU to tell them what to do. This is, in my opinion, the best way to design chips because no developer wants to waste time writing programs that take advantage of multiple thread capable CPUS - they rather the hardware figure it out. Its hard enough to use 2 out of 4 cores on a quad core now..imagine 32 cores!I wish they would give us a socket on the motherboards for a GPU chip. Then they can do multi-core cpus and multi-core dedicated gpus with new ultra fast buses to link. Plus I don't want Intel completly taking over all chip making...that would at the least cut out NVDIA who doesn't make CPUs. They backed the PS3 though so they might help push the Cell.
sprockkets - Friday, March 7, 2008 - link
Maybe Windows Mobile will now have a browser capable of doing IE5! No more excuses!Soccerman06 - Friday, March 7, 2008 - link
Can they throw on one socket? Would be interesting to see if you could put 32 on a socket and perform similar to core2 duo if the program was coded for that many cores.Janooo - Thursday, March 6, 2008 - link
Well, I'm just curious who's going to be the first with CPU and GPU on one silicon.