And they're not stopping at 100, either.
Way back in October of 2009, Tilera Corp announced that it had managed to squeeze 100 cores onto a single chip. The announcement focused on a new line of multi-core processors called the TILE-Gx series. Tilera said it planned to release 16, 36, 64, and 100 core models with the goal of simplifying system architecture. So far, the company has launched the 16, 36, and 64 core models. So where's that 100 core model?
According to Wired, we don't have to wait much longer, as Tilera plans to ship the 100-core TILE-Gx later this year. Not only that, but the company's next line, code-named Stratton and set for release in 2013, will mean an expansion of cores in both directions: Tilera is claiming as few as four and as many as 200 cores for the Stratton line and a move from a 40-nm to a 28-nm process means they can cram more circuits into the same area.
Quanta is one of the manufacturers that Tilera supplies chips to, and though the servers it supplies to big-name web companies aren't powered by Tilera yet, Wired reports that the MIT-run company is on the radar of several companies. In fact, just last summer, Facebook pitted the TILE-Gx against Intel and AMD's Xeon and Opteron server-processors. The test saw a tuned version of Memcached on the 64-core Tilera TILEPro64 yielded at least 67 percent higher throughput than low-power x86 servers.
Though the social network was obviously pleased with the TILERPro64's performance, Facebook highlighted the limited amount of memory the Tilera processors support as a problem for the server chips. Thirty-two-bit cores can only address about 4GB of memory and, as Anant Agarwal, director of MIT's CSAIL (the lab behind Tilera), puts it, "A 32-bit architecture is a nonstarter for the cloud space." Tilera’s 64-bit processors are capable of supporting as much as a terabyte of memory and should more than address this problem, but Agarwal hasn't said if the improvement has been enough to sway Facebook, revealing only that Tilera and the social networking site have a good relationship.
Read more on Tilera, its 100-core chips, and its plans for future server processors, on Wired.
Further Reading:
I know this because I was that 14 year old once.
This. You take a P4 3.80 GHz and I'll take a Core i7-2960XM [2.70 GHz]. The latter DESTROYS the former despite the difference in clock speed...
I'll take $$$ put into architecture changes that yield 10x performance improvements over the same money going towards doubling clock speeds...
I know this because I was that 14 year old once.
This. You take a P4 3.80 GHz and I'll take a Core i7-2960XM [2.70 GHz]. The latter DESTROYS the former despite the difference in clock speed...
I'll take $$$ put into architecture changes that yield 10x performance improvements over the same money going towards doubling clock speeds...
You need to bone up on your CPU history, as do all the other people clamoring for "more GHZ."
The clock speed wars ended because it was totally unsustainable. The increase in heat generated from running at higher clock speeds is greater than the amount of extra speed you get. So, for example, if you increase the clock speed 15%, you increase the heat by 30% (not real numbers). Most people don't have access to plentiful liquid nitrogen, so you can't just keep ratcheting up the speed or the chips will literally burn up.
Instead, Intel and AMD have adopted different strategies to make the chips faster. Intel has aggressively optimized the chips to get a higher IPC (Instructions Per Cycle). This means that more gets done on every cycle. AMD was initially pursuing this as well, and was beating Intel in this arena, but in recent years they stagnated and even regressed (I'm looking at you Bulldozer) on their IPC. To compensate, they just cram more cores onto the chip. Works well in some cases, really poorly in others.
So, the chips we see today are a product of these diverging strategies solving the problem of "How do we make chips faster when we can't increase the clock speed?" Intel chips have fewer cores than AMD chips, but each core is much more powerful than an AMD core. This is why Intel totally dominates AMD in single threaded workloads. At a given clock speed, Intel chips are simply faster clock for clock than AMD's. On the other hand, AMD chips can really stretch their legs when presented a multithreaded workload and usually pull ahead of Intel if they can keep all their cores active.