Slashdot Mirror
Intel's "Terascale" Vision
Vigile writes, "Intel is pushing the envelope with its latest vision — 80 cores on a single processor. Dubbed 'Terascale' computing, Intel aims to bring low-powered, massively interconnected cores and unleash a new era in data-mining, media creation, and entertainment." For balance, read Tom Yager over at InfoWorld imploring AMD to stop at 8 cores while everybody gets the architecture right.
Heh,
Not really.
This chip (as designed) would be one CISC CPU core and 80 Mini cores (kinda like Cell?)
Anyway, where this will be awesome is in rendering &&|| Cryptography, where the memory bandwith requirements is not as high as CPU compute requirements.
I personally hope these come out in a 4xPCIe expansion card:)
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
That is true. A lot of applications do not heavily use multithreading. But, in the scientific community a lot of applications require it. Where I work, we process several GB of MRI data a day. We are able to parallelize the overall processing, so the more processors, the better. However, I wish Matlab would become multithreaded! Our servers have 4 processors and if matlab used them all, we could process 1 dataset in 1/4 the time, instead of processing 4 datasets at once to utilize the CPUs. Processing one dataset at a time would reduce disk I/O.
http://github.com/gbook/nidb
A lab prototype like this can help them with something important: Given multi-core processors look to be the way future computers will be built, how do you feed them data? The current paradigm won't scale past 4 cores on a single chip's worth of FSB, and there are folks who don't think that even 4's going to be a useful increase over 2.
Even if Intel never sells a chip bigger than 16 or 32 ways, an 80 core lab mule will teach them many things about how to get information to a processor and keep those caches full of appropriate data.
-F
It's not the number of processors or cores that they're afraid of, it's the fact that with the exception of a very few cases, your performance does not scale linearly with the number of CPUS, it is less than 1:1. To make it worse, as the number of CPUs rises, the cost to intelligently, quickly deliver sufficient bits to and from all of the CPUs gets exponentially higher.
Recently, some of our managers wanted to see what it would cost to purchase a system that would significantly outperform our 8-way Opteron for RDBMS work. I got numbers on machines from various manufacturers, and when the managers saw them, the conversation was instantly over.
steve
Oh, you're not stuck, you're just unable to let go of the onion rings.
Your comments are a lot more true than many people realize. Specialized hardware always wins.
As an example, people talk about using using multi-GHz machines for TIVO-type appliances, and "getting away" with 600 Mhz or so if your card has hardware MPG encoding. Some of the original TIVOs, because of their reliance on specialized chips and ASICs, used measly 33 MHz CPUs - and worked just fine.
steve
Oh, you're not stuck, you're just unable to let go of the onion rings.
The first 80-core chip will actually look live a conventional kitchen hotplate. You add a pot of cold water on top of the chip, then with a dial on the unit you determie how much heat you want to produce. The CPU will automatically run the correct number of instances of Seti@Home to generate the desired level of heat.
The 4 X 80 "stove top" model will come out later that year. It will include an "oven" that has its own chip and convectional cooling.
That didn't work because AMD worked out that architecture can trump speed. They innovated, and then did it again with decent dual-core (as in NOT the two-dies-on-one-chip cack that you churned out at first).
So, you improved your architecture and implemented dual-core properly, to produce the fantastic Duo. You got back in the race.
And then there was talk of more cores. And you went "Fuck that, bitches, stay DOWN - we is gon' fuck you up good with 80 cores, bitch, an' dat hard!". Yes, you decided to try and dominate the pissing contest of multi-core instead of megahurtz.
Jesus guys, didn't you learn a fucking thing? STOP trying to turn out something that little bit "more" than the competition, just get on with innovating and coming up with damn good chips. That's how AMD threatened you and, if you go on with this "anything you can do" shit again, you'll be back to square one.
Meta will eat itself
Because the chickens actually work better now and most oxen are now really just three chickens yoked together--hell, most chickens are three chickens yoked together!
Please, for the good of Humanity, vote Obama.
What's a memory bus? Oh right, that thing you use to access the DDR4 swap device when the page you want to access is no longer in the on-CPU RAM. ;-)
Seriously, look at the growth of L2 caches, and tell me the day isn't coming when they just call it "RAM" instead of "cache." If Intel and AMD want to keep piling transistors onto their chips, this'll give 'em something to do.
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
if you read his article, you would know that he wants them to create cpus that actually perform in the real world, not just add marketing numbers that will have very little effect.
The complexity of the research is due to the fact that it is so very, very refined. There isn't much out there that's revolutionary, so instead of "Look at this great concept... a wheel" we are getting "The benefits of XYZ fractal tread on a vulcanized rubber tire under wide tempurature variations spanning water phase changes." The second paper is much more educated on the subject, but much drier and not at all revolutionary ;)
:) But there is some Blue Sky research going on.
In Genetic programming, we had heirarchial GP a couple years ago, breaking through the long-running GA problem we have suffered from for 20 years which renders GA useless for anything but trivial search spaces. This was a brain-dead, forehead slapping stupid solution that had just been missed by everyone. MIT press broke the paper, the widespread implications are not yet being felt. This is just a simple example, but hopefully this unlocks some bigger solutions.
Type-calculus is still a very hard unsolved problem, even though the concepts of data types have been around longer than computers. A refined calculus here will open up alot of doors. There are alot of people working on it.
But yeah, we haven't had a Turing come through in quite a while, and it's about time. I just don't think you can "buy" genius. Throwing a bunch of money at the problem may just be throwing it to the wind, or it may indeed foster an environment that this generation's Church comes from... there is no way to know.
Oh, and I can't comment on MS's research... they'd be mad
It's pretty clear you don't have the slightest clue what you're talking about. Nor has that stopped you from forming an opinion, and posting it too. :)
IO bandwidth is exactly the issue missing on PC class machines.
Do you have any idea of the setup time for DMA, and then interrupt overhead involved for each tranaction? For one disk, it's not a lot. Start driving "a lot" of disks so that you saturate the bus, it adds up. These are cycles that steal away from your CPUs main processing capability. Now add 80 CPUs, all starved for IO. It's a total waste of resources.
There's only so much that you can do in memory.
If Intel doesn't do something which can feed those CPU's constantly, you're going to end up with 79 space heaters, and 1 CPU. Given the crunch to conserve power, this clearly isn't feasible.
Let me put it simply for you. Without a major redesign, Intel is simply wasting its money. That's why IO bandwidth is a serious issue here. Without it, they are just talking out their butts.