Clockless processors... why haven't we talked?

The forum for non-component-related silent pc discussions.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
PS
Friend of SPCR
Posts: 81
Joined: Tue Jul 29, 2003 7:31 am
Location: Arcata, CA USA

Clockless processors... why haven't we talked?

Post by PS » Wed Aug 17, 2005 12:09 am

I did a search for "clockless" in our forums and found nothing. I also searched for "ARM processor" and some other things and couldn't find anything. Now I'm mentioning this because as I understand it, there are some significant benefits to clockless processors that we, as silencing enthusaists would be terribly interested in:
(1) Dramatic power draw improvements
(2) Dramatic heat dissipation improvements
(3) Efficiency improvements

From the little I've been able to find about clockless processors, I guess I've come to the conclusion that they could be more efficient, because executions are self-timed. In a traditional processor, there's a little quartz(?) clock that ticks back and forth thousands of times per second, this is where we get our MHz rating from. Each time the clock ticks, an execution occurs, but sometimes the execution may take a tiny amount of time, and be long finished by the time the clock ticks back again. So for all processes like this, a clockless processor would be more efficient, resulting in an overall gain in efficiency. Plus, the power required to keep this stupid clock ticking back and forth is significant. Finally, the heat generated by this little thing moving furiously back and forth is also significant.

Remove this little clock, and solve the execution timing problem in some other way, and doesn't it seem like this would be an all-around great idea? ARM has been making these processors for things like cell phones and car navigation systems, but I can't understand why we haven't seen a version marketed for home PCs. I'm SURE that someone here knows way more about this than I do. Care to enlighten me?

Oh yeah and I also read this, which I assume is just a rumor. It said (dated 2001) that Intel had developed a clockless Pentium that performed three times faster than a standard Pentium... it MUST be a rumor, right? Why wouldn't they release something that great?

mellon
Posts: 105
Joined: Thu Apr 21, 2005 12:17 am
Location: Helsinki, Finland

Post by mellon » Wed Aug 17, 2005 3:31 am

You didn't find anything because the issue is not relevant in todays world. Clockless (or asynchronous) prosessing is far too difficult to implement in PC systems to be of use to the general community. Developing a totally clockless system easily ends up using more space on silicon than a clocked implementation of similar speed. Developing time and cost of a clockless system would also be several (or even tens) of times longer/higher than those of a normal clocked system as a clockless system is wildly more complicated than a clocked system.

In short, it just doesn't pay off in the end. Thinking clockess system would be the solution to heat/speed problems is as much a fallacy as thinking having three states of logic instead of binary would make things somehow faster.

Megaman
Posts: 23
Joined: Wed Feb 16, 2005 6:23 am

Post by Megaman » Wed Aug 17, 2005 7:45 am

I had to jump in on this one. First, my credentials: I design CPU's for a living and I've designed asynchronous (clockless) circuits of the type you describe.

The basic reason you don't see asynchronous circuits is that clock based circuits are easier, faster and cheaper to design and manufacture. The entire industry is geared to building clock based devices. All the design tools, process, manufacturing, testing is based on clocked methodology. There are very few CAD tools in the industry to design self timed circuits and most of those are academic projects. Another reason, that mellon alluded to is the extra silicon cost. Most asynchronous logic is three state logic. The implication of this is that you need roughly twice the number of transistors to implement the same function. More transistors = more silicon = more cost. The control circuits are much more complicated in self timed circuits adding more complexity and cost. Another reason is that very carefully crafted and timed clocked logic (processing pipelines in CPUs for example) waste very little time between clock ticks. This means they run at very close to the speeds of self timed circuits.

Bottom line is that it's easier, faster and cheaper to build a CPU with a clock than without.

Having said that some of the cool things about asynchronous circuits:
- they always run at max speed and are limited by heat and voltage. Want to increase processing power? Put a better heat sink. Want more? Liquid cool. Even more? Turn up the voltage.
- the entire circuit runs at a "natural" speed. No need to make all fast components wait for the slowest one.
- You don't have to redesign them for different technologies. Take a 130nm CPU, build it in a 90nm process and you're done.
- No clock = no crystal = smaller, cheaper motherboards (important for small devices and stuff)
- they do tend to run cooler (leakage current is a big problem in both cases though)

Cheers,
Megaman

BTW, some of my comments are oversimplifications. I tried to stay away from the really technical issues like, clock tree distributions, clock skews, timing issues, retiming, two phase/four phase protocols, transister level design of asynchronous circuits, etc.

PS
Friend of SPCR
Posts: 81
Joined: Tue Jul 29, 2003 7:31 am
Location: Arcata, CA USA

Post by PS » Wed Aug 17, 2005 8:50 am

Megaman, Mellon... thanks for your replies! I really appreciate that you took the time. It finally makes sense why we haven't (and probably won't) see clockless processors for us anytime in the future. As for the problem of heat, power draw and efficiency, do you have any idea how much of the heat in a processor is generated by the crystal? Or how much of the power is required to keep the crystal moving properly? Or maybe how close clocked processors can run to their self-timed equivalents? (In terms of percentages, that is.)

Like was that thing about the Pentium that was self-timed and three times as fast really just a rumor, or was there some truth behind it?

Because as manufacturing processes improve and related costs go down (processors are steadily getting more transistors, a la "dual core"), would it not be worth it to deal with the more complex logic and design, IF heat and power draw for the self-timed circuit would be significantly lower than a clocked equivalent? Another consideration, (according to anandtech.com) gains in computing power are severely limited in dual core CPU's (when compared to single-core counterparts), the gains were mostly 0-10%. When they revamped testing to stress multi-threaded applications, gains sometimes shot up to ~30% or even more, but it was far from a sure thing. Isn't it true that sometimes processors just need to be faster in order to yield significant gains in benchmarks (or at least certain benchmarks)?

Megaman -- HOW COOL! It must be great to design CPU's for a living!! And how convenient to us everyday grunt workers (like myself) to have an expert on the SPCR forums willing to enlighten us about these matters... :D :D

Megaman
Posts: 23
Joined: Wed Feb 16, 2005 6:23 am

Post by Megaman » Wed Aug 17, 2005 9:38 am

PS, power draw from crystals/oscillators and clock distribution is not very significant. The majority of power losses in digital circuits are a combination of leakage current and switching power. Leakage power is the power disipated while the circuit does nothing and switching power is the power need to change a binary value from 0 to 1 or 1 to 0.

I don't know about any self timed Pentiums. The only asynchronous CPU project I can think of that you can directly compare to a comercial CPU is the AMULET series of CPUs which were ARM instruction compatible. If I remember correctly, AMULET CPUs only ran at about 80% DMIPS performance of the equivalent ARM CPU and disipated about the same amount of heat. So in practice, no superstars.

Cheers,
Megaman

PS
Friend of SPCR
Posts: 81
Joined: Tue Jul 29, 2003 7:31 am
Location: Arcata, CA USA

Post by PS » Thu Aug 18, 2005 3:03 am

Well Megaman, thanks again for your enlightening comments. Oh and thanks for the oversimplifications... at the present moment, I have no need to learn about.... ummm... all that stuff.

I guess that if the prospect of clockless processors was as great as I had imagined, they'd already be on the market. :? :lol:

dedogs
Posts: 86
Joined: Wed Sep 01, 2004 3:31 pm

Post by dedogs » Sat Sep 17, 2005 3:52 pm

Thank you PS for making me learn something new and very interesting everyday! :)

Megaman u are just cool :P

now if I can only think of a way to make less power when the circuit changes state hmm... a free energy source like magnetism ?

Myth!
*Lifetime Patron*
Posts: 151
Joined: Sat Oct 01, 2005 3:30 am
Location: Beds, UK
Contact:

Post by Myth! » Sat Oct 01, 2005 4:07 pm

i think you apply a voltage to the quartz...and it vibrates physically. It might be same technology as cigarette lighters where u press a button to get the spark (it squeezes a quartz crystal) in reverse. Perhaps the quartz is held tight so that its vibrations get turned into an a/c ripple. this is then conditioned by a circuit to produce a square wave which becomes the clock signal.

If thats how it works I deserve a drink 8)

Magnetism......

moving a metal wire through magnetic flux gives voltage...but you use more energy to move the wire than what you get out. Using the attraction of two magnets works briefly...but then you have to move them apart again.

Perhaps we could have bicycle pedal driven generators under the pc desk... get fit and save electricity :D

sthayashi
*Lifetime Patron*
Posts: 3214
Joined: Wed Nov 12, 2003 10:06 am
Location: Pittsburgh, PA

Post by sthayashi » Mon Oct 03, 2005 7:49 pm

Myth! wrote:i think you apply a voltage to the quartz...and it vibrates physically. It might be same technology as cigarette lighters where u press a button to get the spark (it squeezes a quartz crystal) in reverse. Perhaps the quartz is held tight so that its vibrations get turned into an a/c ripple. this is then conditioned by a circuit to produce a square wave which becomes the clock signal.

If thats how it works I deserve a drink 8)
More or less. You've kinda described piezoelectricity

Myth!
*Lifetime Patron*
Posts: 151
Joined: Sat Oct 01, 2005 3:30 am
Location: Beds, UK
Contact:

Post by Myth! » Tue Oct 04, 2005 2:48 am

that makes good reading. I'm off for a drink and a piez now :wink:

Post Reply