I have been wanting to write about this for quite sometime now, right since Intel announced that it had a beast of a processor rolling out, with a whopping 80 cores on the same die. It was also good to learn later that most of the grey cell twitching regarding the design of the processor happened at the Intel center at Bangalore.
Intel’s dual cores, when they were announced, had charted out into unknown territory. For a lot of people involved with this technology of the smallest of the small, that was a very interesting moment, filled with lot of intriguing questions, and a promisingly uncertain future. But not many (except for some Intel honchos) would have imagined a day so close into that future, when a processor with 80 cores would even be conceived of. As they say fact is stranger than fiction.
Since the juggernaut of processor design began (gosh! that was like an age ago), the golden way forward has been to put more and more transistors on a chip and make them tick faster and faster. All this courtesy the divine FETs which continued to scale down amazingly. Ah! those were days. But then this free ride started showing signs of problems, not really threatening to stall completely, but requiring increasingly more fuel for its continuance. Fuel in terms of materials research and complicated nano-electronic innovation. The world has since stoutly addressed this requirement, microelectronics being the most funded science discipline (those queer defence robos dont count as science), and by far. But the results only trickled through like sand particles in an hourglass. The world was by then addicted to the scent of faster and faster chips coming out every other year, faster mobiles with all possible apps, faster macs, better digicams. It just had to be there. That is what made the electronics industry what it is, the desire of the consumers to replace their perfectly working device for a device that works twice better.
Enter power consumption. He is playing villian here. All through this modus operandi of faster clocks, tinier transistors and more moolah, the power of the devices was shooting off the roof. Now the world was becoming increasingly mobile and this wasnt taken in the right spirit. This was also why the elec companies were pouring billions on to the materials research front.
Somewhere in all this invisible mess, Intel came up with the idea of the dual core. Presentations at every possible tech meeting were all about how simple the trick was to solve the then dreaded problem of a “hardware slowdown”. Voila! Drink to the human spirit. Intel now is raking in billions by the day for what it did. AMD and others chipped in too.
Now enter the beast of a processor. 80 cores on the same die with 20 MB memory on each and hardware to do the routing very much like a macro sized communication network. So what are the challenges in making such a beast and more importantly putting it to use? First and obvious problem is that programs run on such a processor need to be massively parallel to even make sense of what is being done. Secondly, the interconnect between the processors is going to be the make or break step. Interconnect design as it is today is already taxed, being able to ensure that it does not become a bottleneck is going to take some doing. Thirdly, the memory subsystem problem. Now though there is a tiny bit of memory on each processor, there is going to be a mega bank of shared memory somewhere which every processor would want a share of. Building an optimized memory subsystem (to arbitrate accesses from all the processors to the memory banks) is going to be crucial. I personally find this particularly interesting for it can go a long way in affecting the overall performance, simply because most SoCs today run as fast as they can read and write data into the memory. Finally, with 80 cores capable of computing parallely, the amount of data that would fly in and out of this chip needs to be ideally enormous. So the transfer rates at the ports better stand up and deliver for the proccys inside. Problems are a plenty and that means interesting times are ahead. Whether this would be a case of too many processors spoil the broth or an excellent technological innovation only time will tell.

Man!
I am impressed by your writing!
Bookmarked your page, and now I will be a constant visitor.