Tilera unveils 100-core processor

Author: Gareth Halfacree
Published: 29th October 2009
Comments (17) Stumble
Tilera unveils 100-core processor

The Tilera TILE-Gx processor allows up to one hundred individual cores to be addressed individually.

If triple-core, quad-core, and even six-core processors just aren't doing it for you any more perhaps it's time to look beyond the mainstream duopoly: high-performance computing specialist Tilera has announced the creation of a 100-core chip.

As reported over on V3.co.uk, the manufacturer has claimed a world first with the production - due for commercial release in 2011 - of a processor with one hundred individually addressable cores. While the performance of each core on its own is less than that of a traditionally designed processor at between 1GHz and 1.5GHz with 256K L2 cache, the company claims that the overall performance is four times greater than anything else available today while offering up to ten times the performance per watt.

Omid Tahernia, Tilera's chief executive, believes that high-performance computing users will be able to "replace an entire board presently using a dozen or more chips with just one of our TILE-Gx processors, greatly simplifying the system architecture and resulting in reduced cost, power consumption, and PC board area."

The design of the TILE-Gx chip means that the grid of one hundred addressable cores can be individually toggled into multiple power states in order to conserve energy while idle, and each core can access the resources available to other cores - meaning it's easy for a process to spread across multiple cores. Furthermore, the chip can be configured to run a single operating system across all cores or a hundred operating systems across a single core each - or anything in between.

With Tilera concentrating firmly on web, cloud, and search services as their target market, operating system support is going to be key. This, sadly, is where the stumbling block currently lies: the design of the chip is sufficiently novel that significant work will need to be done in order to port current software and operating systems - including Linux and Windows, the two most popular choices for web service hosting - across to the new architecture. With that work still to be carried out, it's important that Tilera gets sufficient quantities out to market cheaply enough that developers can get to work as soon as possible - otherwise they'll be faced with attempting to bring a chip to market with no efficient software support for such a large quantity of cores.

Does this seem like the logical progression of multi-core processors, or do you believe that a smaller number of more powerful processing cores is still the way to go? Share your thoughts over in the forums.
Quote trig 29th October 2009, 16:00
i dont get this...100 cores...25 times what we have now, and only 4 times as powerful? so, and i haven't looked at intel's roadmap recently, but i would imagine by 2011 they will have an eight-core cpu at least in the works, this 100 core chip will be about as powerful as that? theoretically?

i know i probably dont get where this has a significant advantage somewhere...just seems like a waste...
Quote yakyb 29th October 2009, 16:04
it all marketing BS

4 times as powerful

at what some specific highly multi threaded app how doe sthis compare to cuda which is the current equvalent
Quote Radical_Monkey 29th October 2009, 16:24
This would be very useful in the field of grid computing and cloud computing, saving a shed load of space and less cooling for hopefully similar or better performance. But as yakyb said, GPUs will be competing with these CPUs head to head and cuda has had a huge lead on it. Infact where I work we're looking into using nVidia GPUs for our grid solutions, so I think its got an uphill struggle.

Saying that, if they are supremely power efficient then that could be the ace up its sleeve that it needs....
Quote barndoor101 29th October 2009, 16:32
http://en.wikipedia.org/wiki/Amdahl%27s_law

This is why do dont get a 25 fold increase in speed. Also bear in mind the cpus are clocked alot slower than other cpu cores.

the biggest news is the performance per watt. Increasing this by an order of magnitude is going to make cloud and server providers sit up and take notice.
Quote HourBeforeDawn 29th October 2009, 16:33
this is actually fairly old news, I mean really old news, Im not sure if it was this company but a 100core CPU has already been made and as for performance it was more focused for the manycore concept then the multicore concept that we know now.
Quote l3v1ck 29th October 2009, 16:51
Sounds a lot like Larrabee. If OS's have to add support for masses of CPU's for that, why not for this?
Quote ChaosDefinesOrder 29th October 2009, 17:27
I wanna see what the Performance tab in Task Manager looks like with that! ;-)
Quote identikit 29th October 2009, 17:46
Forget about the 100cores, the GHz, the watts etc. It's all about the architecture. Intel made an 80core CPU running at 10GHz years ago IIRC. The thing stopping them was the non-conventional architecture. Your hardware is only as good as the software you can run on it.
Quote Chocobollz 29th October 2009, 19:16
Quote:
Originally Posted by Radical_Monkey
Saying that, if they are supremely power efficient then that could be the ace up its sleeve that it needs....

No, even if they have an ace up their sleeves and successfully make a three of a kind aces, I'd say any GPGPU (ATI or NVIDIA) will at least beat them with a fullhouse, or a higher hands :p
Quote Bazz 29th October 2009, 20:25
Quote:
Originally Posted by yakyb
it all marketing BS

4 times as powerful

at what some specific highly multi threaded app how doe sthis compare to cuda which is the current equvalent

You are assuming it is being compared to a Intel and AMD chip then yeah??
What about other CPU's commercially avaliable??
Quote perplekks45 30th October 2009, 08:08
Quote:
Originally Posted by ChaosDefinesOrder
I wanna see what the Performance tab in Task Manager looks like with that! ;-)
You mean something like this?

BQ4shSQJTd0
Quote ChaosDefinesOrder 30th October 2009, 10:32
Quote:
Originally Posted by perplekks45
Quote:
Originally Posted by ChaosDefinesOrder
I wanna see what the Performance tab in Task Manager looks like with that! ;-)
You mean something like this?

BQ4shSQJTd0

Can't use YouTube here at work so can't see that for now, but I think it's the AMD Opteron with 16 cores isn't it? Seen that, with the task manager spread over two windows full of "cores" - still wanna see what the tab with 100 little boxes looks like! Would be epic!
Quote identikit 30th October 2009, 14:06
Quote:
Originally Posted by ChaosDefinesOrder
Can't use YouTube here at work so can't see that for now, but I think it's the AMD Opteron with 16 cores isn't it? Seen that, with the task manager spread over two windows full of "cores" - still wanna see what the tab with 100 little boxes looks like! Would be epic!

Vid description: Intel's Kennedy Brown and IBM's Kevin Powell show you a sneak peek of an 8 socket, 64 core, 128 thread IBM server based on Intel's Nehalem-EX processor.

Quote chrisb2e9 30th October 2009, 18:05
Three cheers for innovation. Ok, maybe its not brand new as some people have pointed out. But going against the norm is one way that we can advance technology.
Good job Tilera.
Quote JCBeastie 31st October 2009, 07:26
I wonder how this compares to GPGPU tech. On former side you can use existing hardware and plug more GPUs in to scale (like nVidias Tegra) or have a customised operating system and and entirely new hardware to support these chips...

Hmmmm.

I'm working off the assumption that these chips are built to handle massively parrallel but simple tasks, and thus have drawn the comparison to CUDA/OpenCL.
Quote TSR2 31st October 2009, 16:55
@Identikit: Something like http://blogs.technet.com/markrussinovich/archive/2008/07/21/3092070.aspx then? (just look at the picture)
Quote Timmy_the_tortoise 31st October 2009, 21:57
So these are x86 cores?

I think sticking with fewer, more powerful cores for now will suffice.. With CUDA/OpenCL for any massively multithreaded applications.

Yeah, that's probably better than a 100-core CPU... Since with GPGPU you get way, way more "cores" to thread through.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.