http://www.pcworld.com/article/3085502/components-processors/meet-kilocore-a-1000-core-processor-so-efficient-it-could-run-on-a-aa-battery.html
>Kilocore
>1000-core processor
>Can be powered by a single AA battery
>Can execute 115 billions instruction per second
Yes or No?
sounds like weak-ass useless shit outside of a datacenter
>>55218533
Most processors run at or around 1.5v though.
>>55218566
This processor has a dissipation rate of 0.7W
WE NEED MOAR COARS
>>55218566
At very high amperage, unlike this chip.
it can run on dc power without an inverter?
>>55218533
well its basicly an ibm creation so for them i kinda expect such a thing...
but im really mad imagine the tech that company has in the form of patents and we are here trying bashing intel or amd while ibm just release such a massive cpu that destroys both of them..
>>55218533
This was done 10 years ago, they just gave it a speed bump
but the real question is
is it risc or cisc?
what architecture does this little shit run?
>>55218627
From the article:
>despite being built on old 32nm CMOS processor tech from IBM.
>>55218620
Pretty sure every modern architecture is either RISC or VLIW
>>55218653
I mean, is it x86, ARM, MIPS, custom, ...?
>>55218533
>So efficient, it can't provide power to USB
amazing, just amazing.
>>55218620
their press release has no real substance so there's no way to tell
not like it makes much of a difference anymore anyway, it's not the '90s anymore
>>55218533
Show this to some fucking buttcoin miners and they'll go wild.
>>55218721
Or show this to FBI and NSA. They will be delighted because they can now break all the encryption in existence.
>>55218620
they have "MUH CORES" i suspect RISC with 50MIPS
>>55218751
If it's building on the same project I'm thinking of (how many kilocore projects could there be?) it's 1024 8-bit cores with 1 32-bit core or something along those lines.
>>55218589
That just means it will run for longer on a single AA.
AMD University
>>55218940
how much is it? don't care about power usage but $/performance
>>55219012
You can't buy it
can it run firefox with three tabs on
>>55218533
Cool tech.
But not being x86 means it's of limited use to us.
1000+ 8-bit processors is better suited for something like software-defined networking, live encryption/decryption built into a RAID controller, managing a massive multi-blade server, or on-the-fly encoding of large throughput video broadcasts.
Because of this processor, we might see faster internet, more secure and transparent cloud operations, more cloud-processing for apps, programs, and services, and better internet and TV streams. But we'll likely never see it in our PCs for quite some time, if at all.
>>55219026
well it would be probably expensive then
I doubt it would compute faster than GPUs as well
World's first? This same product shipped over 10 years ago in 256 core form...
https://en.wikipedia.org/wiki/Kilocore
>>55219069
>But not being x86 means it's of limited use to us.
wut. if it's ARM, MIPS or some other well known arch, it's not really limited.
>>55219069
>But not being x86 means it's of limited use to us.
Who exactly falls under "us?" Gaymurs?
>>55218533
>software is still 90% single threaded with a handful of asynchronous tasks dumped in thread two
>>55219218
At least I don't have to feel bad about not owning an i7 extreme.
>>55219218
Implying servers can't make good use of this.
Off the top of my head, VPS providers would love this.
It'll take 10 years before this tech hits the market, I bet they're like 10k a pop right now.
>>55219297
That's what everyone says but:
>work has a handful of esxi, oracle apps/msql, and mail clusters
>boxes that cost in the realm of tens of thousands of dollars
>they're almost universally blocked by memory
>the odd exception is secondary storage iops
>not a single one is hurting for cpu
Don't really care about more cores. Don't really need it, even at the high end.
>>55219218
>there is only one program running on your computer
>>55219183
damn near everyone who uses a pc really.
>>55218533
really wish we could get co processors in pcs.
cuda is nvidia restricted, and opencl barely functions. would be nice to have a massive cpu cluster to offload rendering parallel tasks to and my faster but less numbered cpu can go on to do other tasks that demand high speed performance over parallel tasks.
hell if we went with a co processor, we could have one legacy x86 cpu to run shit in case, and use a different architecture to run newer applications.
>>55218533
Define core
>>55219737
we've already got one, it's expensive though
xeon phi yo
>>55219183
Anyone that uses software built to compile on x86.
So, anything that isn't incredibly niche.
If you could run x264/5 on it that'd already be pretty good, might as well buy the Xeon Phi though.
>>55218912
Nope. Higher amperage means more wattage. Higher wattage means more energy consumed per second.
Power=Current*Voltage
If you ran a 95w 6700k on an AA battery,it would be empty in about 1.5 minutes. I'm guessing this lasts at least a couple of hours.
>>55219899
>>55219714
This isn't as big of a problem as you would think.
Microsoft are the only company that really makes x86 exclusive system software.
Apple have had their shit running on two non-x86 platforms, GNU/Linux and many BSDs are compiled for fucking everything.
Worst case scenario is that these new platforms mean the death of x86 (which is a great thing --if any other platform had the investment that intel throws into x86, they would easily kick the arse of it).
>>55219934
>Microsoft are the only company that really makes x86 exclusive system software.
this is so wrong it's not even funny, the software ecosystem isn't all about big names, not everyone is a teenager who only uses their computer for running a web browser and microsoft office for school work (which, remember how much of a fucking utter failure Windows RT was despite having both of those?)
>Apple have had their shit running on two non-x86 platforms
because they started out that way, and controlled the software ecosystem to the point that hardware swaps were almost indiscernible for the end user
>Worst case scenario is that these new platforms mean the death of x86
as history has shown, this will not happen
>>55218614
This.
>>55219096
>MIPS
can't wait for a 1000 core router from tp-link.
>>55219318
We will always be constrained by bandwidth and latency. Not even photonics will help this, it's a fundamental issue of distance. Maybe if humans never die out and thousands of years from now someone finds it possible to crack the physics of the universe, but realistically we will never have parity between moving data and processing it.
http://vcl.ece.ucdavis.edu/pubs/2016.06.vlsi.symp.kiloCore/2016.vlsi.symp.kiloCore.pdf
>At 0.84 V, 1000 cores execute
1 trillion instructions/sec while dissipating 13.1 W.
Not bad for 32nm. GP104 is 9 trillion instructions/sec while dissipating about 180W. Though how much of that power draw is rendering components (texture units and rops)?
>>55219373
And these programs are totally not just sleeping 99% of the time waiting for an event.
>>55218589
Electric current, you filthy fuck
>>55219851
give me something not as retardedly expensive, fuck i would run a xeon though a pcie if i could, ANYTHING that isnt that retardedly overpriced and requires you to program specifically for it apposed to just using it.
i have an amd cpu, opencl never fucking works on anything i want,
cuda does, but i despise nvidia as a company for fucking me time and time again.
I would have a use for this.
Search.
Certain kinds of searches, like ones where you are looking for fuzzy matches on very long strings, are more easily done on multiple cores.
At my work I have a roughly 10 TB data set that I have to sift through, and it's a pain in the ass to run a match on a couple of Xeons, find out that you need to tweak your parameters, and then run it again.
Could this compete with an FPGA?
>>55220503
GPUs are made for price/performance anyway
>>55218704
I got you fâm.
They normally dont put shit like this in the press release because most are too stupid to read it anyway.
After giving the abstract a read it seems like a lot of hype tbqh. Its pretty interesting, it looks like they've actually used CPU cores as a form of L1 Cache in some scenarios. Neat idea.
http://vcl.ece.ucdavis.edu/pubs/2016.06.vlsi.symp.kiloCore/2016.vlsi.symp.kiloCore.pdf
>>55220503
Most of that power is actually the memory controller/chips believe it or not. DDR is fucking expensive power wise. HBM is better bandwidth/watt, but its still power hungry. This gets around that with some memory tricks.
>>55221310
Never, they're totally different devices.
