Tilera Announces 100-Core Processor

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Who needs a quad core processor when you can get a CPU with 100 cores? According to this article, Tilera’s chips will start shipping next year, with the 100-core chip scheduled to ship in early 2011.

The two-year-old startup's Tile-GX series of chips are targeted at servers and appliances that execute Web-related functions such as indexing, Web search and video search, said Anant Agarwal, cofounder and chief technology officer of Tilera, which is based in San Jose, California. The chips have the attributes of a general-purpose CPU as they can run the Linux OS and other applications commonly used to serve Web data.
Click to expand...
 
PFFFT still wont be able to do physix.

Because nvidia says BUY BATMAN!!! DX11 is useless!
 
Steve said:
Who needs a quad core processor when you can get a CPU with 100 cores? According to this article, Tilera’s chips will start shipping next year, with the 100-core chip scheduled to ship in early 2011.
Click to expand...

yeah as early as 2011 BUT the *games* aren't going to *utilize* 100 cores until 2060 or so...
 
DASHlT said:
PFFFT still wont be able to do physix.

Because nvidia says BUY BATMAN!!! DX11 is useless!
Click to expand...

You got it wrong. :)
It should have been: " But can it play Crisis?" :D

On a serious note, how efficient will each core be? Will this be like 100 mosquitoes trying to lift a pound or more like 100 ants.
I guess anyone can make 100 cores, but if it can't do the work of say, one efficient core or 2, than what's the point? For bragging rights?
 
Is that 1.5ghz per core or 1.5ghz total for all cores. And if the first do i have to build a pc using teh core in a chest freezer, and how big is teh chip itself
 
They can run Linux? This would be great for my dev box for work but I wonder how much money they'll cost.
 
TeeJay88 said:
Is that 1.5ghz per core or 1.5ghz total for all cores. And if the first do i have to build a pc using teh core in a chest freezer, and how big is teh chip itself
Click to expand...

thats 1.5ghz per core, available in 100, 64, 36 and 16 cores.

the 100 core chip is 45 x 45mm BGA.

I would assume that a cpu using 55watts isnt going to be extremely hot, they claim their chips to be "low power" and are marketing them as such.
 
Valkhorn said:
They can run Linux? This would be great for my dev box for work but I wonder how much money they'll cost.
Click to expand...


the end of the article says bulk pricing will be from $400-$1000 per chip when they ship. Since they range from 16-100 cores im guessing the 100 core chip will be in the $1000 range. Which seems pretty cheap if it can indeed compete core for core with existing cpu's.
 
Hulk said:
yeah as early as 2011 BUT the *games* aren't going to *utilize* 100 cores until 2060 or so...
Click to expand...

50 years? Probably some kind of insane stuff out there by then. Look what kind of progress we made in the last 10 years.
 
Hulk said:
yeah as early as 2011 BUT the *games* aren't going to *utilize* 100 cores until 2060 or so...
Click to expand...

Jospeh said:
50 years? Probably some kind of insane stuff out there by then. Look what kind of progress we made in the last 10 years.
Click to expand...


I already have a prime number generator I made that can utilize thousands of CPU cores.

Now if other programmers would get off their rears and learn how to do multithreading properly we would be all set.
 
Imagine 100 Via cores all mushed into one die.

Sounds interesting. Surely there is a real world application for such a CPU?
 
A two year old startup leap frogs Intel and AMD by 25x? Something smells like vaporware.
 
Im sure each of those cores can't hold a candle to a single mainstream intel or amd core. Probably 1/50 - 1/100th of the raw power, but then, its designed for huge multithreading (but low demand threads) apps like web serving.
 
They try to compare themselves to Intel in the article, but this CPU can't be x86 architecture, no license... so they can't even be dropped into the same kind of systems.

Video cards often have this many or MANY more processing cores, that are used for general applications such as folding, encryption, vidoe encoding.... tons of GLOPS... and the mainstream videocards can be found in fairly low power form as well.

Not impressed...
 
ehh, ok so they are coming out with a processor? Where is the rest of the hardware platform that supports this processor? Kinda confused.... Their site gives little detail about implementation...
 
DASHlT said:
PFFFT still wont be able to do physix.

Because nvidia says BUY BATMAN!!! DX11 is useless!
Click to expand...

What has this thread got to do with batman or DX11?


That said the chip is in direct competition with fermi which has similar lofty aims I suspect.

Like most little cpu companies they will get a load of investor cash, release Mk1 which won't work half as well as it should but get enough cash to last a while longer, then hope some big company buys them up. Most likely they'll die and some big company will snap up the remains cheap for the patents.
 
Monkey God said:
Im sure each of those cores can't hold a candle to a single mainstream intel or amd core. Probably 1/50 - 1/100th of the raw power, but then, its designed for huge multithreading (but low demand threads) apps like web serving.
Click to expand...

Exactly the same way I looked at it.
 
Jutsu said:
ehh, ok so they are coming out with a processor? Where is the rest of the hardware platform that supports this processor? Kinda confused.... Their site gives little detail about implementation...
Click to expand...


the article is announcing the release of a line of processors of theirs. If you want to know more about it you could just google it...
 
If it runs Linux, it can fold right?
Though I bet a PS3 beats it at folding too. So a system with 100 cores and the CPU cost $1,000 gets beat by a console that costs $800 less?
 
Jutsu said:
ehh, ok so they are coming out with a processor? Where is the rest of the hardware platform that supports this processor? Kinda confused.... Their site gives little detail about implementation...
Click to expand...

IIRC, it's going to be an ARM architecture processor, so plenty of hardware and software will run with it.

I think there was supposed to be an ARM version of Windows 7, too; that might still be in the works.
 
Cool, a 100core, not x86 processor...

ATi, nVidia... you should do this!

Oh... wait....

I think they already have.
 
does anyone know the manufacturing cost of a multicore processor? What kind of margin is set for AMD/INTEL cpus'? How could 100 core become profitable in 2years when 12 core AMD cpus will sell for more in next year.?
 
Menelmarar said:
They try to compare themselves to Intel in the article, but this CPU can't be x86 architecture, no license... so they can't even be dropped into the same kind of systems.
...
Not impressed...
Click to expand...
I have a feeling you're not the one they're trying to impress...

jeremyshaw said:
Cool, a 100core, not x86 processor...

ATi, nVidia... you should do this!

Oh... wait....

I think they already have.
Click to expand...
Shader units are not cores, they're functional units. That's like calling a P4 "multi-core" because it has more than one ALU. GPUs are superscalar single-core processors.
 
Has anybody thought about the licensing costs for software on a 100 core processor. Buying vsphere for that machine would bankrupt companies. $40k in the host, god only knows how much for everything else.
 
Phelptwan said:
Has anybody thought about the licensing costs for software on a 100 core processor. Buying vsphere for that machine would bankrupt companies. $40k in the host, god only knows how much for everything else.
Click to expand...
Ah well, if it hastens the death of silly per-processor (or per-socket) licensing, all the better.

I really don't think this product is targeted at 'generic' computing though. It doesn't seem like it would be particularly useful for typical tasks, however there are plenty of applications where it would be great. Sounds like it would be great for flow classification and similar tasks in high-end routers, or for applications that need to run lots of realtime or tightly synchronized tasks simultaneously (probably some audio/video applications, control systems). Maybe some use for 'cloud computing', but my understanding is that most of these applications tend to by I/O bound, and gain their parallelization improvements mainly by partitioning data (or requests) across machines, not across cores, but I'm hardly an expert.
 
You must log in or register to reply here.
Back
Top