Can you say 100 core cpu

Interesting that they don't seem to be using an established CPU core.
 
100 core, not fully x86 cpu... hmmmm..

I wonder what ATi and nVidia should be calling their GPU now... oh what! :D
 
would be cool if the gpu could boot and run code without any intel cpu needed. but in the meantime I wouldn't mind 100 or more cores if it could run windows. ;-)
 
mobusta1 said:
But will it play crysis?
Click to expand...

Not when it's seriously lacking on the floating point hardware, no. This is a chip mostly for shuffling data around. The cores are all mostly integer based, and FP instructions being used on Tilera chips are strongly frowned upon.
 
xxGriff said:
you say that like its a bad thing :p
Click to expand...

Not necessarily, but it seems disingenuous to spend all that engineering effort to design your own core that seems similar to existing licensable IP (MIPS, ARM, PPC etc.). Not to mention that you've got to build new development tools and get developers & operating systems onboard with your product.

If they can get it running Linux though, and it shouldn't be hard, it could make a very compelling product for the data centre.
 
Interconnect seems to be the key to unlock another big door in our progression of a better computational model. Currently, interconnect are, to say the least, very rudimentary. In that sense, 2D matrices are an immense step forward, at least conceptually. The interconnect on this chip is beyond your current CPU, and that is purpose-driven. Apply a full x86 implementation to this concept and you have a monster.

Then again, I'm a fan of CELL and Power. Tera-scale compute on a single chip... now if only we could get that in a more friendly version that will work with our current Windows gaming setups to combine with the GPU...
 
Matthew1034 said:
Still waiting for the clockless CPU.
Click to expand...

So if it were clockless.. would it then theoretically be infinitely fast?

You are going to have to have something based on something other than transistors in order to even try to accomplish that.

It would be very nice to not have to deal with multipliers though.... a system that allows a lot smaller adjustments than what is available now. Full and even half multipliers kinda anger me sometimes.

Also being able to adjust the speed of different things withought affecting the speed of other things would be way nice as well...

Like... speed of L1, L2, L3 cache, integer part of CPU, FPU part of CPU, etc.
 
cyclone3d said:
So if it were clockless.. would it then theoretically be infinitely fast?
Click to expand...
Maybe he's just referring to a fully static CPU that can run down to 0Hz or at any other clock rate up until its rated maximum? Most embedded CPUs have this property. Basically totally flexible clock scaling capabilities, though modern desktop/mobile CPUs are pretty strong here already, and relaxing the static requirement offers performance gains.

Like... speed of L1, L2, L3 cache, integer part of CPU, FPU part of CPU, etc.
Click to expand...

I don't think this would be very useful, though perhaps the L2 and L3 cache clocks would be nice to control independently. The rest of the CPU operates in tight sync though, and you want the whole pipeline to operate at the same speed or you're just going to be stalling parts of it while they wait for the next step to catch up - speeding up one component doesn't help since the rate at which you can actually do work depends on the other, slower components - it's an integrated system. And making asynchronous interfaces like this is possible, but it increases complexity quite a bit both increasing power consumption & size and reducing performance. Probably not a good tradeoff in most situations, at least where the design is not very application specific and the CPU's structure and clocks are tailored specifically to the code that will be running on it.
 
I wonder what kina overclock you could get on this chip? The power consumption is low so the thermals are probably damn good.

Someone should do the LN2 thing!
 
100 is 4 in binary, right?

i already have one...

:D
 
mrmylanman said:
I'm not sure if you were being a smart ass or not, but a clockless logic circuit just simply won't work. There has to be a clock input somewhere for it to work.
Click to expand...

This is incorrect.

Combinatorial logic doesn't require a clock obviously, and you can have asynchronous state machines. They bring their own host of problems to the table, but a clock is certainly not required.
 
cyclone3d said:
So if it were clockless.. would it then theoretically be infinitely fast?

You are going to have to have something based on something other than transistors in order to even try to accomplish that.

It would be very nice to not have to deal with multipliers though.... a system that allows a lot smaller adjustments than what is available now. Full and even half multipliers kinda anger me sometimes.

Also being able to adjust the speed of different things withought affecting the speed of other things would be way nice as well...

Like... speed of L1, L2, L3 cache, integer part of CPU, FPU part of CPU, etc.
Click to expand...
you are practically and theoretically bound by the smallest quantum of time that can exist. something about plancks constant or something. and also, if u dont have some ratio between all of the given things you mentioned, there will be increased latency. if i throw a ball in a straight line five times a second at a hole that passes by two times a second, the ball wouldnt line up with the hole and id have it bounce back or be lost
 
marylandman said:
you are practically and theoretically bound by the smallest quantum of time that can exist. something about plancks constant or something. and also, if u dont have some ratio between all of the given things you mentioned, there will be increased latency. if i throw a ball in a straight line five times a second at a hole that passes by two times a second, the ball wouldnt line up with the hole and id have it bounce back or be lost
Click to expand...

There's no "quantum of time". There's simply a time past which quantum mechanics and general relativity split into separately definable systems, about 10^-43s after singularity. The reason that this time is known as the "Planck time" and an "upper limit" is because the Standard Model of physics cannot explain what goes on before that, for some esoteric reason having to do with the ratio between wavelength frequency and energy (denoting a minimum quanta of energy that electromagnetic radiation can carry). Before this time gravity, strong, weak, and electromagnetic forces were merged, and this unified force has no explanation under the Standard Model. But time is inherently indivisible, unless you want to measure it.

Planck postulated that the explanation to "Maxwell's Oven" lay in discrete energy steps, where there is a minimum energy step (10^-34J), and that higher frequencies take more energy (that energy and frequency are tied). This apparently gives him a bit too much credit, but I have to find a source for that (the above is what is commonly taught). Maxwell already knew that wavelengths radiated in whole steps, (cannot have less that 1 full wavelength for a given frequency of em), and Planck realized that if you take that discrete sum concept and apply it to the energy each wavelength of a given frequency has, you'll come to a finite solution to the blackbody radiation problem (because only whole wavelengths radiate, and it takes increasing energy to radiate higher frequncies, and in a closed system you have finite energy).

edit: Btw the assumption that there is no "free energy" ala what this guy is describing, is poorly thought out. I'll agree that free energy doesn't exist when you can explain to me why electrons' orbits don't decay. hint: it has to do with "free" access to energy in a vacuum

Btw, when you're talking about microscopic "bounces" that ball will on average go through the missing hole a significant portion of the time. If it didn't leakage would be no problem.
 
Last edited by a moderator:
Can you say waste of money because no current program/os utilizes all of those cores?
 
The current graphics processors have more than 100 processors so this isn't unheard of. Seems like this would really only be good for extremel parallel processing...
 
When you guys and gals talk about performance with this 100 core CPU, please realize that each core will be slow compared to today's cpu cores. This is the same with the intel 80 core chip we that was announced last year. Every core of the Intel chip was less than 1/2 the power of a low end atom core. This type of cpu would be good for some very specific problems but definitely not good for a general purpose machine.
 
drescherjm said:
When you guys and gals talk about performance with this 100 core CPU, please realize that each core will be slow compared to today's cpu cores. This is the same with the intel 80 core chip we that was announced last year. Every core of the Intel chip was less than 1/2 the power of a low end atom core. This type of cpu would be good for some very specific problems but definitely not good for a general purpose machine.
Click to expand...

The best process for this CPU would be one where the number of cores was directly proportional to the amount of data processed (not saying Hz doesn't matter). The CPU can run the OS and whatever tasks it is assigned - it would probably be a great deal cheaper to install and run several of these in 1U racks instead of pay out the posterior in larger racks and the labour involved in fitting large GPUs to do the same task (just a comparison, since I imagine GPGPU tasks scale similarly).

Potentially. I do admit that its niche processor, but would you take 100 cores with half of the power of an atom machine each? The answer depends on what you would use the machine for.
 
TIWizard said:
100 is 4 in binary :confused:
Click to expand...

100 in decimal translates to 1100100 in binary, or "3". Yes 4 decimal is 100 in binary. Just looking at it from the opposite end, since the regular joke was already played. Like a mixed metaphor.
 
mrmylanman said:
I'm not sure if you were being a smart ass or not, but a clockless logic circuit just simply won't work. There has to be a clock input somewhere for it to work.
Click to expand...

That's not true. Handshake solutions makes a clockless CPU core. Among the other benefits, there's a huge reduction in EMI for this chip.

"ARM offers first clockless processor core"

Asynchronous CPU

The chip itself likely has a clock for the I/O that requires it, but the CPU core doesn't.
 
Additionally, getting rid of the clock tree eliminates an enormous chunk of the dynamic power usage.
 
You must log in or register to reply here.
Back
Top