World’s First 1,000-Processor Chip

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
A UC Davis team has created a microchip containing 1,000 independent programmable processors. One of its main strengths is power efficiency; 115 billion instructions can be executed per second while using only 0.7 watts.

“To the best of our knowledge, it is the world’s first 1,000-processor chip and it is the highest clock-rate processor ever designed in a university,” said Bevan Baas, professor of electrical and computer engineering, who led the team that designed the chip architecture. While other multiple-processor chips have been created, none exceed about 300 processors, according to an analysis by Baas’ team. Most were created for research purposes and few are sold commercially. The KiloCore chip was fabricated by IBM using their 32 nm CMOS technology.
 
still waiting for the quantum nanotube laser processor. Seriously, I'd like to see a major breakthrough in the processor arena.
 
I'm reading the dissertation where it's described. The summary of the power consumption above is kind of way off. Going by the numbers in the paper, the average power per instruction and other power usage numbers given in the paper, are that "KiloCore" uses around 15W just in the execution cores for that level of performance. (Pascal and Polaris are more than 2x more efficient per watt than that for a reality check, and mobile versions should increase that gap even more.) The KiloCore chip lacks caches and has very little scratchpad RAM. The simplicity of each core makes it fairly worthless for PC-like work, but it works decently for ALU-heavy highly parallel problems (just like a GPU!). It's a research design and it wouldn't be competitive with anything on the market since it's not optimized for sustained throughput. It does seem to be able to grab headlines though. :p
 
Considering the masks alone normally cost in the millions, who paid for this and why? A smaller version could have been tested with FPGAs. No one cares how many cores it has or how fast it is if it isn't commercially viable.
 
I thought someone collected the first 1000 microchips manufactured.
 
ecmaster76 said:
Considering the masks alone normally cost in the millions, who paid for this and why? A smaller version could have been tested with FPGAs. No one cares how many cores it has or how fast it is if it isn't commercially viable.
Click to expand...
Who says it used that kind of mask set? :p KiloCore was made in collaboration with IBM, and was basically a low volume prototype. There would be no reason to spend a fortune on a production quality mask set. Even if it used a mask set, the prices for non-production ones are in the tens of thousands of dollars (assuming there's not too many metal layers). IBM has demonstrated various electron beam and heat-based chip manufacturing prototypes in the past, so that's also a possibility. I don't think anyone cared if it took dozens of hours to make a single chip at a time, when that is weighed against the fortune it would cost to make many at once.
 
  • Like
Reactions: Xrave
like this
pxc said:
The simplicity of each core makes it fairly worthless for PC-like work, but it works decently for ALU-heavy highly parallel problems (just like a GPU!).
Click to expand...

i was going to say... 1000 cores on a die sounds something like a massively parallel processor, say, CUDA cores on an nVidia GPU.
 
ecmaster76 said:
Considering the masks alone normally cost in the millions, who paid for this and why? A smaller version could have been tested with FPGAs. No one cares how many cores it has or how fast it is if it isn't commercially viable.
Click to expand...

As someone who works for a semiconductor company, mask cost, even for this 32nm process, is only about $10k-50k.

I would be interested to know the yield, assuming they made more than the 16 die tray shows.
 
You must log in or register to reply here.
Back
Top