It is estimated that the IBM blue Gene project will use 80,000 chips and produce 80 to 256 teraflops of computing power. What we need is 80,000 processors on a single chip. If we can scale the postage stamp size intel 80 core chip by 1,000 percent, then we could have a serious workhorse that could give us the raw computational power needed for strong ai.

In the 1970's and early 1980s, large chips were common place. Even the first 486 chips in the early 1990's were larger in servers than in the later 1990s when they were in consumer machines. So what if the processor is the size of a dollar bill, or double the size of a dollar bill?

80,000 processors per chip is what we need to start the strong ai revolution. 80,000 processors per chip is an interesting number, a good number, because it is estimated to be similar to the same amount of neurons in a single cortical column. I think we will need 80,000 cores per chip to solve the strong ai and autonomous computing platform and make it work as a technology.

We need to rethink why we buy a computer. We buy a computer to process information, we buy new machines so we can process more information, and access more features that the additional computation allows us to do. Better speech recognition, video editing, money management, education, schedules, write papers and do more work than the previous machine.

Since computation is the reason we buy machines, shouldn't we put raw processing power on a chip? Let the machine optimize the code. Let the machine code the solution once we program the seed ai and nueural networks? Why stop at one processor per motheboard. For the problems of strong ai, we need 10 processors per motherboard. This would be more logical than single chip computer designs, and save many older computers from the landfills.

The intel 80 core chip, is a start. If it was not for the patent law and large corporations, if we could use the best of the intel 80 core chip design, and supersize the chip. We need to make the chip big by a factor of 1,000. We need an 80,000 core processor on a chip with 32 GB of memory on or under each core.

I my vision of the future, I call this computer the “Dave 80”.

The real reason we want 80,000 cores on a single chip is so we can build a 10 processor computer that has 800,000 cores from 10 processors. This would allow us the freedom to model complex datasets, and crack the problem of strong ai, nautural language processing. A 10 processor system could have different configurations given your goals, and keep the machine out of the landfills for a few additional years longer than a single processor system.

You might want to have all 800,000 cores in a seed ai configuration. If you were building a gaming machine, you might divide the 10 cores one for game input, several for game AI, several for object recognition and object understanding, and two processors for real time graphics output. Think pov-ray distributed on 1 or two 80,000 core processors. Imagine 160,000 processors generating photo-realistic 3-D. I call the this machine the “Dave 800”.

Network ten 800,000 processor systems into a simple network and you have a very powerful system for strong ai applications. I call this the “Dave 10M” or “Dave Pro” system. With a blazing 8 million cores, I think it would have the raw power to understand language, create self improving software based on scientific benchmarks and at some point exhibit strong ai.

I think that 8 million cores should be enough to model a strong self-learning ai. However, a “Dave Pro” system could be as large as 100 to 1,000 computers. I do not think we would want or need a 100,000 machine system, but the rate that progress makes machines obsolete, once the first “Dave 80,000 processor core machine is in mass production, there will be a need to increase the amount of cores by a factor of 10, 100, and 1000.

The end result of an 80,000 core machine, and 80,000 core machines in networked configurations is that we can hopefully create computers that can help us think faster and validate results faster than it would be possible by human thought and cognition alone.

Now if I could just get the venture capital. We could have 80,000 core machines in 15 years if we worked to make it happen. Why go for 80, when you can scale by 1,000 percent. We will need to scale by 1,000 percent to solve the computer problems that our current generation of machines are incapable of performing. The only way to get out of the problem box is to scale processing power by 1,000 percent from existing multi core designs.

10 processors per machine, at 80k cores each, times 10 machines give you 8 million processors total. I think that would be enough power to brute force create functional multipurpose ai.

http://www.bit-tech.net/hardware/2007/04/30/the_arrival_of_teraflop_computing/1

http://domino.watson.ibm.com/comm/wwwr_thinkresearch.nsf/pages/20010611_cellular.html