1,000 Core CPU Achieved: Your Future Desktop Will Be a Supercomputer

Wel as they said, specialized FPGA hardware (including hardware to do MPEG encoding/decoding) has been around forever.

I can't really see an advantage to any desktop to have to program it's cpu cores for every single task you might want to run on it at any one time. And really only specilized and wel contained tasks LIKE video encoding can really take advantage of this. Though you could maybe have photoshop program every single one of it's installed filters into a seperate core on launch. Just in case you might want to use one?

Some of this is actually being done right now in CS5 and much more so in CS6 using your GPU which already has dozens or hundreds of graphics oriented cores on it right now.

I would think that you would still need a healthy set of general purpose cores with a satellite of speciality cores that could be programmed on launch of a process for those very few well contained in/out transform processes at best.

seekermeister said:

I don't pretend to understand the implications and requirements of this type of processing, but in my imagination, I would think that those 1000 cores could be divided into blocks of cores...for instance, 100 cores each, in which one block would be for one purpose, and other blocks for other uses. Even a block for general purpose use, so that it would be compatible with software not designed with this in mind.

Well said......... Something like multi block CenCores =100 core blocks into 10 SuperCores...... That would run Massive Apps for multi processes at the blink of a eye

The problem is, the cores need to be programmed to perform a function, and that function has got to be well contained and static for it to be much use or efficient, such as a codec. The test they did did just that, but then try running word at the same time and performance may drop to 1/100th that of a I7. They are not any good for the other 99% of computing which is on the fly utterly unpredictably random requirements.

That is why the current trend for hybrid operation of generic processors combined with on the fly specialized CPU/GPU usage for "filter" functions is really the way to go. And the way things are actually already going as a decent GPU on sale today already has 512 cores, and more and more software is starting to actually use them (other than games of course).

seekermeister said:

I know nothing about using a GPU as you have described, but doesn't it also have some design limitations on how the cores can be used?

Yes and no, they are programmed in C or other standard languages but have parallel computing extensions which, if you want to really make use of the architecture one would have to master. But that's the same with using any massive set of processors. Even today's 4, 6 or 8 core systems are RARELY used properly due to the increased complexity in writing code to do so and the fact that opportunities to really take advantage of multiple cores are not as widespread as you might think.

In fact past 4 cores, the very idea of just slapping more cores on instead of getting the clock speed or efficiency up is for the most part a disappointing cop-out since 90% of all computing can't take advantage of it. (Atleast in a desktop setting, servers are another story)

Graphics are the one thing that really shine in that area though, which is why GPUs have been doing this for years now. It'll be interesting to see how much Photoshop CS6 takes advantage of it.

These CPUs might be great for very specialized purposes like video servers or multiple stream encoders, but I doubt they'll have much use in a regular desktop. It was the swipe at intel and the comparison to the desktop CPU that sort of got me a tad riled about it in the first place (Even if it was just in jest)

seekermeister said:

I said nothing in jest, but then I said nothing intended as a swipe at Intel either. ...

Sorry I didn't mean you. I meant the tone of the original article :)

"Scientists at the University of Massachusetts Lowell laugh in the face of Intel's weedy handful of cores in its new CPU lineup:"

and

" the team was able to process 5 gigabytes per sec of movie files, which is about 20 times the rate that existing high-end computers can manage."

I mean obviously it was in jest, but the comparison wasn't really warranted. The two aren't comparable. The 1000 core CPU they are talking about could not do 95% of what an intel cpu could do at 1/100th the speed. Though it can do a very few things that are very constrained somewhat faster. They should have compared it to a modern GPU instead. Since a modern GPU is in fact a large single/dual core CPU coupled with 500 or so smaller task oriented cores.

They should have billed this as a possible runtime programmable coprocessor to existing CPU tech. That would have been a practical and awesome suggestion for it's use :)

Incredible technology, but considering the average laptop still only comes with a quad-core and desktops come with 6-cores as standard, doesn't it seem like this would be a little bit far-out in terms of hitting public consumer markets? I could always be drastically wrong, though.

The main question raised in my mind by this article, however, isn't a matter of processing capability, but rather one of data carrying. What type of cable would be needed to carry all the data that can be processed by this? What does this do to our already super-fast SSDs?

Oh, and then there's this: how does one cool this?