Power-Light Power Chips

DD writes to tell us ZDNet is running a story about a new Santa Clara, CA based startup that is boasting a new line of low-power, Power chips, the same architecture found in current day Macs and IBM servers. From the article: "The company's first so-called PWRficient chip will feature two processing cores, run at 2GHz and consume on average about 5 watts, thanks to an emphasis on integration and circuit design. At a maximum, it will consume 25 watts, far less than the single-core Power chips that can hit 90 watts found on the market today."

Comments

Embedded market

[Thanatopsis]

According to the article they are going to focus on the embedded market. I guess they mean the embedded market that need 2 GHZ embedded chips.

Re:Embedded market

[supabeast!]

From the article:

"The PWRficient actually won't come out for two years, so it's hard to predict exactly how it will stack up against the competition."

In two years a 2 GHZ dual core will probably be a good option for a high-end embedded CPU.

Re:Embedded market

[afidel]

Uses
Print rasterizers: I have printers with imaging engines capable of 30+ppm but I rarely achieve it in the real world because the printers are hobled by a measly ~500Mhz rasterizer.
Networking equipment:If you want to do any kind of complex routing or switching in a truely flexible manner without ASICS you are going to need as fast of a processor as possible.
Complex analyisis of data in an appliance:Antispam appliances are often limited in the algorithms they use because the cost in processing time for some of the better ones are too expensive to apply to the volume of messages they are supposed to handle.
etc.

While I am aware that there are large swaths of the embedded market where nothing more complex than a microcontroller is needed I am also cognizant of the fact that there are many areas where a more powerfull embedded processor which is still energy efficient is still very usefull.

Re:Embedded market

[wowbagger]

Actually, YES, the embedded market that needs 2GHz chips - folks like me doing signal processing for communications, among other things. Do you have any idea how many operations per second it takes to do an echo canceler for a phone, or to do GSM or CDMA decoding in software (if you want a system that can adapt to new protocols - a software defined radio or SDR - you need to use a more general purpose part than the dedicated ICs for this), or to do the latest 802.11 protocols, or to do video decompression, or ....

Yes, Virginia, there is a market for 2GHz processors in the embedded space.

Re:Embedded market

[TheRaven64]

It's not like Tivo's video encoding/decoding takes up any processing time.

No, it's not if they've got any sense. At least, not general purpose CPU time. Dedicated video compressor / decompressor chips get much more performance per watt, and usually more performance per $ as well when compared to general purpose hardware. The iPod video can play H.264 clips that a moderately fast G4 struggles with - and not because the iPod has a faster CPU.

Re:Embedded market

[wowbagger]

Because the difference between a DSP and a normal CPU is very small now-a-days.

It used to be that only DSPs had multiply and accumulate instructions - now many CPUs do (the Power being among them).

It used to be that only DSPs had the register count to do an FFT without having to spill to memory during the butterflys - the Power also has enough registers to avoid having to spill to memory in the innermost butterflys.

It used to be that only DSPs had the fast barrel shifters for single-cycle shifts of more than one bit position - now most CPUs have them.

I can go on and on - but simply put, the only real difference between a DSP and a modern CPU is that very few DSPs are clocked at 2GHz, while many CPUs are.

The really fast DSPs are the ones like the TI C6X family - which get their "speed" from being very long instruction word processors, much like the Itanium. They don't have a very high clock speed - the fastest C6x is running about 1GHz. They are benchmark queens - the will do a 4096 point FFT blindingly fast. Oh, you wanted to do something ELSE with the data after you did the FFT? Sorry, but now you are going to lose most of that speed as the code falls out of cache, and as you run out of vectorizable code and stall most of the cores. Besides, you can get just as much speed-up using the vector instructions of a modern CPU (Altivec/SSE etc.) as you do from the C6X processor.

They also suck when you are doing protocol as opposed to signal processing - DSPs *hate* jump instructions, and don't EVEN think of asking them to do a context switch - they are like a drag racer, they go fast until you ask them to TURN.

In short, the days of the DSP as the king of signal processing are past - you can do more with a general purpose processor and an FPGA than you can with DSPs for the same amount of board real-estate, bill of materials cost, and power consumption.

Sorry, but since this is actually what I do for a living, I know from first-hand experience that DSPs really aren't all they are cracked up to be with respect to regular processors now-a-days.

Re:Embedded market

[Superfarstucker]

This reminds me about a story of an engineer, a toaster, a king and a computer 'scientist'.

Once upon a time, in a kingdom not far from here, a king summoned two of his advisors for a test. He showed them both a shiny metal box with two slots in the top, a control knob, and a lever. "What do you think this is?"

One advisor, an engineer , answered first. "It is a toaster," he said. The king asked, "How would you design an embedded computer for it?" The engineer replied, "Using a four-bit microcontroller, I would write a simple program that reads the darkness knob and quantizes its position to one of 16 shades of darkness, from snow white to coal black. The program would use that darkness level as the index to a 16-element table of initial timer values. Then it would turn on the heating elements and start the timer with the initial value selected from the table. At the end of the time delay, it would turn off the heat and pop up the toast. Come back next week, and I'll show you a working prototype."

The second advisor, a computer scientist , immediately recognized the danger of such short-sighted thinking. He said, "Toasters don't just turn bread into toast, they are also used to warm frozen waffles. What you see before you is really a breakfast food cooker. As the subjects of your kingdom become more sophisticated, they will demand more capabilities. They will need a breakfast food cooker that can also cook sausage, fry bacon, and make scrambled eggs. A toaster that only makes toast will soon be obsolete. If we don't look to the future, we will have to completely redesign the toaster in just a few years."

"With this in mind, we can formulate a more intelligent solution to the problem. First, create a class of breakfast foods. Specialize this class into subclasses: grains, pork, and poultry. The specialization process should be repeated with grains divided into toast, muffins, pancakes, and waffles; pork divided into sausage, links, and bacon; and poultry divided into scrambled eggs, hard-boiled eggs, poached eggs, fried eggs, and various omelet classes." "The ham and cheese omelet class is worth special attention because it must inherit characteristics from the pork, dairy, and poultry classes. Thus, we see that the problem cannot be properly solved without multiple inheritance. At run time, the program must create the proper object and send a message to the object that says, 'Cook yourself.' The semantics of this message depend, of course, on the kind of object, so they have a different meaning to a piece of toast than to scrambled eggs."

"Reviewing the process so far, we see that the analysis phase has revealed that the primary requirement is to cook any kind of breakfast food. In the design phase, we have discovered some derived requirements. Specifically, we need an object-oriented language with multiple inheritance. Of course, users don't want the eggs to get cold while the bacon is frying, so concurrent processing is required, too."

"We must not forget the user interface. The lever that lowers the food lacks versatility, and the darkness knob is confusing. Users won't buy the product unless it has a user-friendly, graphical interface. When the breakfast cooker is plugged in, users should see a cowboy boot on the screen. Users click on it, and the message 'Booting UNIX v. 8.3' appears on the screen. (UNIX 8.3 should be out by the time the product gets to the market.) Users can pull down a menu and click on the foods they want to cook."

"Having made the wise decision of specifying the software first in the design phase, all that remains is to pick an adequate hardware platform for the implementation phase. An Intel 80386 with 8MB of memory, a 30MB hard disk, and a VGA monitor should be sufficient. If you select a multitasking, object oriented language that supports multiple inheritance and has a built-in GUI, writing the program will be a snap. (Imagine the difficulty we would have had if we had foolishly allowed a hardware-first design strategy to lock us into a four-bit microcontroller!)."

The king wisely had the computer scientist beheaded, and they all lived happily ever after.

BUT... BUT..!!!

Re:Amazing

[Thanatopsis]

It is amazing expecially when you consider that many current Intel chips suck down 150 watts at 2.8 GHZ. This isn't like Transmeta either. The team at PA Semi are some pretty heavy hitters in the chip design world.

Apple

[ZachPruckowski]

So are they going to be regretting moving away from that? I mean, that would have an appeal in a low to middle end laptop that can run for 12 hours or something. I'd certainly pay for it. I'm impressed with my iBook battery as it is, but it is just shy of being able to cover all my needs in a day. Or at least, usually have to think about charging it. An 8 hour laptop would be great for people on the move, like students, or amateur filmmakers.

Re:Apple

[Egregius]

They won't regret moving away. This startup is aimed at embedded chips, not desktops. Furthermore, Apple hasn't completely *abandonned* Power just yet. And it's not like Apple can base it's productline on a small start up with no real guarantees it can meet Apple's demands.

Re:Apple

[yamla]

Hardly. These chips aren't due out for two or three YEARS. Let's assume for the sake of argument that they ship IN BULK in two and a half years, an obviously optimistic estimate. Should Apple be satisfied with dual-core 2 Ghz laptops in the spring of 2008? I certainly hope not. While the power usage is sweet, we are looking at less than a 20% increase in speed (assuming you can safely compare clock speeds which, as we know from Intel and AMD, is not a good assumption) for a single core over that time. Even with dual core, that's pretty pathetic.

Now, if these chips were shipping in bulk TODAY and were able to be ramped up to 3 or 4 Ghz over the next six to twelve months, then maybe Apple might start regretting moving away from the G4 and G5 CPUs. That is, it'd be a toss-up at that point. As it is, this is far too little too late for Apple's laptops.

Of course, this rests on the assumption that Apple cares about processing power.

I feel a great disturbance in the Force

[katana]

As if millions of Apple customers suddenly cried out, and were silenced.

finally

[kevin.fowler]

What a relief. Implement this en masse and a dormitory full of idling computers running aim won't use as much energy as a small country anymore.

I'm just wondering:

[mctk]

How much power do processors use relative to the rest of the computer? It seems that hard drives and fans would use the majority of power (not to mention monitors and speakers if present).

Re:I'm just wondering:

[masklinn]

It seems that hard drives [...] would use the majority of power

The average 3"5 (desktop) hard drive (aka 7200RPM SATA/ATA133) runs around 7W idle and about 10W in seeking, high-perfs being a bit higher (12W seeking for 72Gb 10000RPM Raptor drive)

Notebook 2"5 5400RPM drives run around 1W idle (0.8W for a Samsung M40 MP0402H) and around 3W seeking.

It seems that [...] fans would use the majority of power

The fans I can check right now all fall between 0.15 and 0.30A, 12V.

This means that running them at max tension (12V) you're looking at 1.8W to 3.6W. Undervolt them at 7V and you fall between 1 and 2W.

And these are specs for 80mm to 120mm fans

So no, hard drives and fan often ain't the worst offenders as far as power consumption goes.

This vs ARM Cortex A8?

[Sam Haine '95]

I wonder how this will compare to the ARM Cortex A8 in 2007?

Vaporously Delicious

[brogdon]

"The company's first so-called PWRficient chip will feature two processing cores, run at 2GHz and consume on average about 5 watts, thanks to an emphasis on integration and circuit design. At a maximum, it will consume 25 watts, far less than the single-core Power chips that can hit 90 watts found on the market today."

Also, thanks to our patented Vapor-based architecture, we've been able to build our level-2 RAM cache out of a giant cloud of gaseous water! And we've licensed our chips to be in the Phantom Game Console! And they'll even run Duke Nukem Forever! As we speak the SCO group is printing out some infringing Linux code with them to use as evidence in an actual trial!

Re:Vaporously Delicious

[Thanatopsis]

Well except the guys at PA Semi have actually designed and shipped chips.
Here's some Bios
Dan Dobberpuhl, President and CEO

Dan Dobberpuhl, President and CEODan Dobberpuhl, who cofounded P.A. Semi in July 2003, has been credited with developing fundamental breakthroughs in the evolution of high-speed and low-power microprocessors. Prior to founding P.A. Semi, Dobberpuhl was vice president and general manager of the Broadband Processor division of Broadcom Corporation. He came to Broadcom via an acquisition of his previous company, SiByte Inc., founded in 1998, which Broadcom acquired in 2000. Before that, Dobberpuhl worked for Digital Equipment Corporation for more than 20 years, where was credited with some of the most fundamental breakthroughs in microprocessor technology. In 1998, EE Times named Dobberpuhl as one of the "40 forces to shape the future of the Semiconductor Industry." In 2003, he was awarded the prestigious IEEE Solid State Circuits Award for "Pioneering design of high-speed and low-power microprocessors."

Dobberpuhl holds 15 patents and has many publications related to integrated circuits and CPUs, including coauthorship of the seminal textbook Design and Analysis of VLSI Circuits, published by Addison-Wesley in 1985. He holds a bachelor's degree in electrical engineering from the University of Illinois.

Nah - he knows nothing about processor design - but random dude at slashdot know more.

Rest of the team's bios

BSD

What a business plan!!!

[IGoChopYourDollars]

1) design a low-power-consumption high-performance PowerPC chip that would be ideal for Apple to use
2) keep the development so secret that spouses are kept in the dark
3) launch the product after Apple has already abandoned PowerPC
4) ???
5) PROFIT!

Yes

[Wesley Felter]

According to the Web site it has AltiVec. By 2007 I think Apple will have switched completely to Intel, never to look back.

POWER != PowerPC

[frankie]

These are (theoretically, since they don't exist yet) based on the POWER architecture used by IBM big iron servers, which is related but incompatible to the PowerPC chips in Macs. Different pinouts and almost certainly no Altivec.

Perhaps if this company had existed a couple years ago, Lord Steve might have given them an audition before jumping to Intel. But even if they somehow got their current chips to mass production in industry-record time, they would still be years away from being able to ship a PowerPC version.

Re:25w is way too high!

[Jeff DeMaagd]

The chip that you use should only consume a few nanowatts of power, so there's room for refinement.

I've used those 8 pin PICs, they are pretty nice. I ran straight from battery power, a linear regulator is too wasteful. It was for a hazardous material area where running power was undesirable and the battery cell had to last a year. I powered sensors directly from the output of one pin, so I can turn the sensors on only when taking a reading. Obviously, it's only useful for low current devices, but other than maybe the speaker, I see no high current devices preventing the technique from being used on several of the inputs.