Yes. 50 cores. Five zero. All on a single, tiny chip. It’s real. Intel beamed over their 50-core “Knights Ferry” processor yesterday at a supercomputing conference in Seattle, Brier Dudley of the Seattle Times reports. And they have good reason.
The tiny chip is capable of one teraflop of processing power — an esoteric way of calculating how fast a chip is at crunching numbers. By comparison, the fastest Core i7 can only crank out 109 gigaflops. Stuffing this many cores onto one processor has pushed Intel to the point of just labelling it a “many” core chip — at a certain point, the number of cores becomes an abstraction.
Knights Ferry isn’t meant to be a computer’s main brain, but rather a programmable co-processor that’ll do hugely heavy lifting handed to it by a CPU. But there’s no doubt that core count is the future, and the race to push the “many” in man cores is on. This tech will be in your tablet, eventually. Remember when it was about megahertz? Barely. [Seattle Times and MarketWatch]
Photo: Xbit
NOZ
Thursday, November 17, 2011 at 8:26 AMBarely. Mhz…Cores…and then…???…
rdiac
Thursday, November 17, 2011 at 4:26 PM…and then…???… AND THEN… Carefully remember this phrase:- “I for one, welcome our new multiphasic masters.” So it’s all good then.
S0ULphIRE
Friday, November 18, 2011 at 2:28 PMAN THEENNNNNN?
Lachlan Bromage
Thursday, November 17, 2011 at 8:36 AMI’m not sure on the difference between GPUs and CPUs but wouldn’t it be a far more efficient solution to use, say 20 of those cores, for graphics or something?
Like I said I have no idea how the two different processors work but i’m sure it could be tweaked as an experiment.
Linea A
Thursday, November 17, 2011 at 8:56 AMI think there’s certain calculations that GPUs can do that CPUs have a hard time with.
Matt L
Thursday, November 17, 2011 at 9:51 AMGPU is a processor, it’s just dedicated and built specifically for processing computer graphics.
Gabriel
Thursday, November 17, 2011 at 9:58 AMYeah CPUs can handle graphics (to an extent) but it puts huge loads on them. It’s better to off-load it to a dedicated processor for better performance. I suppose the advantage of having a GPU on the same chip though would be the high data transfer rate
Jono
Thursday, November 17, 2011 at 12:45 PMAnd the disadvantages? This is a (educated) guess but I’d say that the instruction pipe is way longer/more complex for a CPU than GPU – meaning that your control/instruction ratio will be way lower on CPU. Also; I’m fairly sure GPU’s and CPU’s have similar data transfer rates – modern PCIx -> north bridge -> ram surely wouldn’t be heaps slower than cpu -> north bridge -> ram, and GPU’s generally operate on easy memory addresses with sequential data rather than data from all over.
I guess it comes down to the fact that, as Matt L said, the GPU is dedicated to graphics – it’s what it knows. CPU’s have to be able to handle a much wider array of functions, hence added complexity.
HR
Thursday, November 17, 2011 at 9:47 AM1 teraflop? Like the one they showed in 2007?
http://news.cnet.com/2100-1006_3-6158181.html
I guess the fact this one can do it with 50 rather than 80 cores is an advancement though.
Ben H
Thursday, November 17, 2011 at 9:52 AM“Remember when we were excited about dual-core processors!?” will become the new “Remember when we were excited about 1gb hard drives!?”
Sam
Thursday, November 17, 2011 at 1:03 PMIt already is, Hexacore CPUs and i5s etc
jeremy
Thursday, November 17, 2011 at 11:18 AMhybrid GPU/CPU. Basically a GPU where the cores can run “slow x86″ – hypercore intel atom with maths goodness added basically. Nowhere near the performance of the fastest NVidia/ATI chip for graphics stuff though … very interesting – according to some “projected” benchmarks this can run current games nicely on about half its cores.
Pat
Thursday, November 17, 2011 at 1:18 PMHurry up and make a Neural Net processor. I want me a learning computer!
John Connor
Thursday, November 17, 2011 at 1:47 PMSkynet already use these processors in their cybernetics systems.
meh
Thursday, November 17, 2011 at 1:49 PMCore count is only the future because it’s cheaper and easier than dealhng with the constraints of higher frequency. You can’t go massively multilayer with carrier plates.
Tim
Thursday, November 17, 2011 at 2:35 PMA co-processor hey, Takes me back to installing an overdrive co processor for my 486 SX (my memory is rusty as i was only 8 or so at the time so i don’t recall the specifics except i was the popular kid at school cause i had 16meg of ram when everyone else only had 8)
Matt
Thursday, November 17, 2011 at 3:44 PMSweet, you could have played Doom full screen!
…I had a 486DX…sooooo much better than your puny SX ;o)
Michael
Thursday, November 17, 2011 at 10:55 PMI was playing doom on a 386DX/2 66 with 8 megs of ram. I did have to seriously alter the autoexe.bat and config.sys files. possibly a .ini somewhere there too. In the end I just created a custom boot disc because my mum got sick of me “stuffing” the computer up.
I didn’t see it as “stuffing” it up, I seen it as making the computer better. I work in a supermarket these days because of my mum. But I do step outside the firewalls set up to get onto the web during my lunch breaks on the stores computers. Storenet is my bitch.
Tim
Friday, November 18, 2011 at 12:23 PMI hated having to screw around with autoexec and config.sys to gain more usable memory in dos, nothing worse than being about to rain hell onto the drug lords in ATAC and it dumps you out because it was out of memory.
Tim
Friday, November 18, 2011 at 12:21 PMWe did eventually upgrade to a DX, dad brought one home from work.