AMD Releases 2 Low-Power 64-bit Processors

rwiggers writes "AMD has released two new low power processors for embedded apps. With a power of 18W and a chipset with 3W of average consumption [PDF] it seems we may have some interesting competition with Intel's Atom."

Comments

servers

[Lord Ender]

Low-power chips are great for low-load servers. I bought a cheap-o Atom nettop, no bigger than a DVD player, slapped a 2TB disk in, and installed Linux. Bam--instant offsite rsync server for my backups. The whole system uses less power than a lightbulb, makes almost no noise, and has a fanless CPU!

It may not be right for a high-load AJAX web app platform or for an HTPC, but the low power chips are more than enough for sufficiently responsive linux+ssh server.

Re:servers

[TheRaven64]

Yup, I'm looking for a low-power 64-bit chip for a home NAS. ZFS performance is much better with a 64-bit CPU (partially due to the fact it likes having a lot of kernel address space, partly because it's very heavy on 64-bit arithmetic), which eliminates the Via chips and the low-end Atoms from consideration. The power usage is much too high for a portable (21W total? What is this, 2002?) but for a low power non-portable it's quite reasonable. It will be interesting to see how it scales down when the machine is idle.

Re:servers

[wirelessbuzzers]

VIA Nano is 64-bit. Dunno how its price/performance/power compares to an undervolted desktop CPU or cheap laptop CPU. It's definitely faster, more power hungry and more expensive than an Atom. But like the Atom, it's also definitely available in Mini-ITX.

Re:servers

[linhares]

Here's a tip for you, son. Whenever you need to find out something, you may want to take a look at these interesting websites, for starters: Wikipedia or Google. Sometimes these websites have information for you. Now go on and give it a try. What's a goatse?

point of sale systems?

[jschen]

I notice on AMD's PDF (linked in summary) that they list some of their envisioned uses. Why would someone need a modern 64 bit system for a point of sale system? Wouldn't a Motorola (err... Freescale) 68000 be more than powerful enough for the task, and way cheaper? I don't understand why some seemingly rather simple applications would require a large amount of processing power.

Re:point of sale systems?

[grub]

I don't understand why some seemingly rather simple applications would require a large amount of processing power.
When your only tool is a hammer, every problem looks like a nail.

.

Re:point of sale systems?

[Albanach]

Wouldn't a Motorola (err... Freescale) 68000 be more than powerful enough for the task, and way cheaper?

.

Maybe, but not by much - reports suggest the Atom costs less than $10 to manufacture. At that price any savings between processor types is pretty tiny unless you're deploying a vast number of them.

There's so much x86 development though, I'd imagine x86, and especially windows programmers, are much easier to find and cost less to hire. The processor cost in a POS system is going to be a tiny fraction of the total when you add in touch screens, bar code scanners, cash drawers, scales etc etc.

From the manufacturer's point of view it can probably develop software faster and cheaper using .net and it's that saving that probably drives lots of x86 uptake in these sorts of devices.

Re:point of sale systems?

[fuzzyfuzzyfungus]

Blame software. If you look around, a surprising number of POS systems are running some sort of XP embedded + POS graphical bloatware combination. (Dell's offerings in the area are more or less representative, if you are morbidly curious). Obviously, POS functions could be(and were) done on seriously weedly embedded hardware. Trouble is, if your business is already running quickbooks or something, and they want their cash register to integrate, the path of least resistance is to buy quickbooks' cash register product, which is a giant pile of bloat that only runs on full systems. On a global scale, you'd probably save money by rebuilding it to run on embedded ARM or something; but any individual economic actor is better off just buying a (still shockingly cheap) general purpose X86.

You still don't need 64 bits for that; but all of AMD's designs(aside from some of their old Geode gear, and maybe embedded products based on Athlon XPs, if you can still buy those) are based on Athlon 64 cores, and they would save essentially nothing by disabling 64 bit capability, and might lose in certain applications that do require 64 bit support, so they might as well ship with it.

Re:point of sale systems?

[Cassini2]

The embedded market is known for its fondness of cheap hardware, and sticking to the status quo. For many years, DOS was a dominant O/S for Point of Sale applications. In recent years, Windows is getting more popular. Linux is a big portion of this market, because it is free, and has real-time extensions. You can control entire machine tools in real-time with Linux, implementing the servo-loops on a PC in software. You can even prototype embedded applications, like motor controllers, in real-time Linux on a PC, and then port to an appropriately sized embedded platform when you know your processor requirements. With embedded PC chips, like these, it may not even be cost effective to switch off the PC platform for some applications.

The 68000 / ColdFire line is getting rather old at this point. The problem is that only two platforms long-term can keep up the pace with the cost of modern processor development. The IBM Power series (including the cell for the PS3), and the Intel/AMD x86 platform. Everyone else, cannot keep up with the cost of modern processor development and the cost of fabrication facilities. Even AMD struggles to finance fabrication, with the resulting Global Foundaries spin-off. Intel struggles with the same problem too. Intel is unable to make the Itanium (Itanic) line competitive with the x86 line. The top 10 supercomputers are all either Power series processors, x86 AMD or x86 Intel. On the total Top 500 list, only a very few systems (6) use Intel Itanium. It simply costs too much money to develop a processor architecture, and only the largest architectures can remain competitive on a cost/performance basis.

There are many simple embedded applications that do use inexpensive processors. There is an entire industry developing 8-bit, 16-bit, and even some 32-bit embedded processors (including the ColdFire.) However, generally the PC is not competing in the same space. Interfacing between the 8-bit processors and the PC is becoming a challenge.

Does anyone know good ways to connect the embedded processor to a standard PC motherboard? RS-232 is becoming rare. Ethernet overwhelms the small processors with data. Any good embedded communication solutions for networked motor drive and control applications?

Re:point of sale systems?

[0100010001010011]

Cables. Duh... Search for MultiSeat.

http://netpatia.blogspot.com/2009/06/multiseat-in-ubuntu-904.html
http://linuxgazette.net/124/smith.html

Great if you have kids or a larger family. One decently powered machine can power multiple "computers".

The technology they used was sending VGA & Audio over USB.

Re:point of sale systems?

[Hal_Porter]

But one of the users would install all sorts of malware and viruses. My highly trained (MSCE A+) technician tells me it's impossible to build an operating system which is immune to infection. He also told me Ubuntu had a higher total cost of ownership than Microsoft Vista XP 7 and doesn't support industry standards properly.

Re:point of sale systems?

[TheRaven64]

Most ARM chips these days have VFP and NEON extensions, so FPU performance isn't much of a problem. The only time you get serious performance issues is when you have a chip with no FPU and try to run code that requires one. In this case, every floating point operation (including loads to floating point registers and stores to floating point registers, which happened a lot with the old ARM Linux ABI which passed floating point values in registers) traps to the kernel, which emulates the instruction. If you compile for softfloat and use the old ABI, then you need everything to compile for softfloat. With EABI, floating point registers are not used for parameter passing, so you can have some libraries using softfloat and some using hardfloat, which improves compatibility.

These days, however, it's quite rare to find an ARM core anywhere other than the very bottom end of the market that doesn't have an FPU (and a SIMD unit). Anything based on the Cortex A8 core, for example, has reasonable floating point support.

You're almost right about Java. Most ARM chips have a set of extra instructions for things like bounds-checked array accesses that are used to make it easier to implement a JVM (basically, an ARM chip has a few different instruction decoders, one of which can read Java bytecode, directly execute just under half of the JVM instructions, and quickly jump to other routines for accessing the others). These instruction sets are not not published, but ARM does provide a Java implementation for Linux if you pay for it. That doesn't mean you can't use Java with a completely open-source stack, it just means that you don't get the accelerated JVM. The real benefit from Jazelle is that it makes interpreted bytecode almost as fast as JIT-compiled bytecode without the extra CPU resources needed for the JIT compile and the extra RAM needed for the caching the JIT-compiled code. In terms of overall performance, a JIT compiler is as fast as or faster than a Jazelle-based interpreter. Presumably you could combine these two techniques and JIT-compile hotspots in the code and use Jazelle for the rest, but ARM's JVM doesn't do this (it would take more memory, which isn't really an option for mobile phones, which are their biggest market). If you've got enough memory for a full JVM, it's not a problem.

Re:point of sale systems?

[petermgreen]

The 8 bit processors on the other hand tend to be pretty good at the low level stuff, PICs for example can toggle a pin on one clock cycle and read back how the hardware responded to that pin on the next, they are also prety cheap. So in embedded systems you often see a 16 or 32 bit main processor to do the real work and then one or more little 8 bit microcontrollers to do all the fiddly hardware stuff that the main processor doesn't want to be bothered with. This design also may make the software easier as you don't have to learn about the driver development interface for the OS that runs on the big processor.

For $30 you can get a nice embedded processor that supports Ethernet, serial, PCI-E, generic I/O and more without requiring a separate bridge chip.
yeah but expect to spend thousands of dollars worth of design time and prototyping costs to develop a board but on it.

For low volume stuff I can certainly see the advantage of using a PC based board as the main processer and then using an 8 bit micro (which can be accomodated on a relatively cheap 2 layer PCB or even stripboard) connected by serial or USB as an IO processor.

Depends on the atom chipset...

[fuzzyfuzzyfungus]

Compared to the common Atom + 945, AMD's new offering should(assuming it is reasonably priced) absolutely murder the Atom. The atom itself is a pretty low power chip(albeit slower than any A64); but the 945 is a nasty power hog, and has lousy 3D performance. An A64 and Radeon IGP in the same power envelope is hardly even fair, no contest, game over.

On the other hand, intel also has a low power atom chipset, with the "GMA500" they licenced from PowerVR. That particular combination will be weaker than this AMD offering; but it'll come in at something like 25% of the power draw.

This should, assuming it can score enough design wins to actually be buyable in a form other than trays of 1,000, be excellent competition for the Atom+945(being substantially more powerful, in the same thermal envelope), should be quite competitive with Atom+Ion(GPU performance will likely be a wash, CPU performance will be better, power envelope similar); but it won't have much effect on Atom+GMA500(substantially faster; but markedly higher power draw will keep it out of the smaller devices).

I'd love to see these show up in mini desktop systems, or the new thin and light slightly-larger-than-netbook laptops that are showing up.

Re:Depends on the atom chipset...

[avandesande]

Intel's next iteration for atom called 'PINEVIEW' is going to have memory controller and GPU on the chip- it is easy to see why they didn't design a completely new chipset for a configuration that was falling off the roadmap. Don't forget too that Intel goes through a pretty strenuous validation cycle for their customers which the 945 has been through.

I am sure the next generation will address most of these power concerns and then AMD will be the one 'murdered'

Weak competition for netbooks

[PhrostyMcByte]

Intel's netbook Atoms run at 2.5W/11.8W right now -- already beating them out for power usage. Because of how important battery life is to netbook users, I don't think this has much hope of competing there. Intel does have other higher-power Atom CPUs that aren't meant for netbooks, so maybe that's the market AMD is going for. I'd be curious to see how large that market is, though.

Re:Weak competition for netbooks

[fmachado]

But do not forget the 945 chipset eats energy like there is no tomorrow, so combine Atom (~4W)+ 945 (~24W) and then compare to AMD + AMD Chipset and they end like almost same (even favoring AMD a bit) power envelope but AMD will be much more powerful. 945GC eats a little less but only because better idle control.

Even Intel acknowledges it and is using a new chipset will far less consumption, but still with very weak video.

ION plataform is powerful with video but eats almost same power than 945 chipset.

Re:Cool

[ExE122]

I didn't look at individual pricing, but the AMD Turion Neo X2 L625 is alread being offered in a laptop from HP - listed at a base of $569.99 but the processor is a $75 upgrade... or so you think, as soon as you select it you are told you need to upgrade the video card as well!

Either way, they wasted no time getting this on the market. The price seems competetive with the Intel Atom model.

I'm sure it's just a matter of time before Intel one-ups them though.

18W "Thermal Design Power"

[Anonymous Coward]

Okay, so the 18W number is "thermal design power"... sigh, another bloody spec.

Is this a typical spec that is used for comparison? I ask because I've been an electrical engineer for 15 years and, up until now, have done fine with "typical power consumption" (which is supposedly 3 W for this chip, compared to 7 W for the Intel Atom Z530) and "maximum power consumption", which is what you have to design the power supply around, lest the supply rails brown out.

Sigh... like they say: "A datasheet writer can get twice the performance out of a chip that an engineer can."

Re:18W "Thermal Design Power"

[Rob the Bold]

Okay, so the 18W number is "thermal design power"... sigh, another bloody spec.

Is this a typical spec that is used for comparison? I ask because I've been an electrical engineer for 15 years and, up until now, have done fine with "typical power consumption" (which is supposedly 3 W for this chip, compared to 7 W for the Intel Atom Z530) and "maximum power consumption", which is what you have to design the power supply around, lest the supply rails brown out.

Sigh... like they say: "A datasheet writer can get twice the performance out of a chip that an engineer can."

The Thermal Design Power is the spec for the cooling system -- so relax, it's the Mechanical Engineer's problem, we don't do thermo.

Re:18W "Thermal Design Power"

[EmagGeek]

"Sigh... like they say: "A datasheet writer can get twice the performance out of a chip that an engineer can.""

And a marketing manager can get 4 times that...

Re:Cool

[linhares]

I love the fact that this will lower Intel's ability to interfere in the netbook market in a heavy-handed way.

Re:Huh?

[danpritts]

It's a guarantee of availability.

The typical lifetime of a CPU package is a year or 18 months.

Embedded designers want to be able to design around something that won't disappear next year right when they've got the bugs out and they're ready to ship.

Re:Cool

[MBGMorden]

I'm sure it's just a matter of time before Intel one-ups them though.

Always has been that way. Hopefully AMD will in turn one up Intel again, and the competition thrives. I remember back in the days of the K6-2 series of processors when an AMD chip never beat an Intel chip at anything other than price. You bought AMD not for any performance reason but because it was "good enough" and cost half of what an Intel chip did. It's great that AMD eventually reached a point when they DID beat Intel on price AND performance for a while. I know they've been slipping some, but I hope they keep it up.

Don't get me wrong - I'm no fanboy (I've got 5 machines right now - a Linux box, a Mac, and a Windows laptop, all running Intel chips, and a HTPC and my Windows desktop running AMD chips, so I actually have more Intel than AMD at the moment), but I really do hope that AMD survives, if only to keep Intel in check. Their prices are also still very competitive. I'm looking at replacing the aging Celeron 2.66Ghz chip in my Linux machine, and figured I'd like to go quad-core on it (it's my only remaining single core machine). Cheapest Intel Quad Core? $160. Cheapest AMD Quad Core? $80. It's a tad slower, but as a bonus the AMD chip burns about 30% less power as well. Looks like it's gonna be an AMD for that machine.

Are You High?

[JoeSixpack00]

Assuming you're half the Linux fanatic you're coming off as, I'm sure you already know that no one has more open documentation than AMD/ATI. Hell their open documentation and support is pretty much keeping them alive on Linux, seeing how their drivers are still inferior to those of nvidia. So because they didn't have an entire article promising to do what they're already doing, you're complaining?

Re:Cool

[wild_berry]

The thing that swings me in AMD's direction each time I put together a computer is that the MB and AMD CPU together are comparable for performance at a lower price point than the Intel chip and its MB.

Re:Virtualization is even better for low-load serv

[Just Some Guy]

Virtualization is even better for low-load servers

Yep! Virtualization is great for settop boxes and remote backup servers, except for the 99% of situations in which it's impossible.

All AMD Has To Do To Kill Atom...

[Nom du Keyboard]

All AMD has to do to kill the Atom is to not impose asinine restrictions (e.g. screen size <11.7") on its usage. It's as simple as that. Do that, and you will kill a good piece of the much more expensive Core 2 Duo market as well since that's what Intel is trying to foist off on the anything-larger-than-what-we-define-as-a-netbook market.

Re:All AMD Has To Do To Kill Atom...

[TheRaven64]

It doesn't matter. The important restrictions come from Microsoft, not Intel. If you want the $15 OEM licenses for XP, you need under 1GB of RAM, less than 160GB HDD or 32GB SSD, and a single-core processor under 1GHz (with some specific exemptions for certain models, including most Atom chips). For Windows 7, the CPU restriction is a single-core CPU under 2GHz. The screen size for XP can't exceed 12.1", but Windows 7 restricts this to 10.2". If your specs exceed this then you have to pay the full price for a Windows OEM license, which drives up the cost a lot more. Of course, you don't need to run Windows, but if you're going with Linux then you may as well go with an ARM chip instead of AMD and get lower power and better performance-per-Watt for less money.

Re:True 64 bit processor

[level_headed_midwest]

What do you mean by "true 64-bit processor" or "32-bit processor with 64-bit extensions?" A CPU is either a 32-bit CPU (can only use at most 32-bit instruction words) or it is a 64-bit CPU (can use 64-bit instruction words). The CPU in question is based on the AMD Athlon 64, which was the original x86_64 CPU. These CPUs can execute 16, 32, or 64-bit code, depending on the OS that is installed. If it's running a 64-bit OS, the CPU runs in 64-bit mode, where is uses 64-bit instruction words. I would say it as well as all x86_64 CPUs are "true" 64-bit CPUs.