Intel demonstrated a 80-core, Teraflop CPU!!!

http://www.monitor.si/images/novice/...l_teraflop.jpg
http://i.n.com.com/i/ne/p/2007/Teraf...om_400X600.jpg
http://i.n.com.com/i/ne/p/2007/CUTer...ip_550x367.jpg
http://i.n.com.com/i/ne/p/2007/demo_...ps_550x367.jpg
http://i.n.com.com/i/ne/p/2007/Teraf...or_400x600.jpg

Quote:

---

Intel last week demonstrated a working processor with 80 individual processing cores. Each core, or "tile" as Intel calls them, consists of a compute element and a router that allows each tile to connect to its neighbor. The chip can deliver more than 1 trillion floating point operations per second (teraflops), depending on how fast it's running. This is only a research project right now, as there are a lot of challenges involved in making an 80-core chip that's a practical option for PCs and servers.

Intel demonstrated the teraflop research chip in a briefing for reporters last week at a San Francisco hotel. It built a special cooling system for the chip and ran a few applications to demonstrate its performance. Intel had the chip running at around 3GHz during the demonstration, but has gotten it to run faster in its Oregon labs with a water-cooled system, said CTO Justin Rattner.

This is how Intel demonstrated the chip for reporters. Several power supplies were required, and Intel also needed to build a special motherboard. Engineers showed how Intel can obtain different levels of performance by tweaking the chips' clock speed and voltage supply while running an application that solved complex mathematical equations.

---

C.net news

Quote:

---

Vara says the 80-core chip uses less than 100 watts of energy; a dual-core chip uses 60 to 70 watts and a quad-core uses 105 to 130 watts. Of course the numbers for the 80-core chip could be affected by the fact that it's lacking some functionality, but Rob Enderle, president and principal analyst of the Enderle Group, says it's still a significant accomplishment.

"We're quite literally creating a network mesh to let each little core communicate with the other cores and the rest of the system," says Vara. "The cores will want to know what the other cores will doing so they don't fight."

While it may take five to eight years to come out with a working 80-core chip, Vara says IT managers might start watching for what he calls "different flavors" of quad-core chips. "Maybe you'll have interim chips where they have more complex cores along with simpler cores, too."

---

EE Times

OMG! :slobber:

:slobber: :slobber: :slobber: :slobber: :slobber:
OMG thats really awesome.
Well, not really awesome, but pretty awesome.

just imagine a multi threaded superpi on that rig...

I tried to, but the calculation was finished before I really got a thought together :D

Thats pretty sweet!

cool but I thought they were also going to include stacked Cache with these

i love the theoretical language the intel engineer is using, they might have mixed chips, like he doesnt know, lol...

Quote:

---

Originally Posted by mr_knowitall15

:slobber: :slobber: :slobber: :slobber: :slobber:
OMG thats really awesome.
Well, not really awesome, but pretty awesome.

---

Stop stealing quotes from Jackass 2 :slapass:

wow what a rig :slobber:
read it some time ago that they are going "multicore" but didn't expected them to show it that soon.

even it still takes its time, impressive where technology moves...

looks like a lot of wires there , nice research :)

http://i.n.com.com/i/ne/p/2007/Teraf...or_400x600.jpg

I wonder what kind of heatsink that is. Looks a little like the ThermalRight SI-128 but with 6 heatpipes.

Folding at home
We need it.

Quote:

---

Originally Posted by CPLB

I wonder what kind of heatsink that is. Looks a little like the ThermalRight SI-128 but with 6 heatpipes.

---

looks like water, or chilled water maybe...
http://i.n.com.com/i/ne/p/2007/demo_...ps_550x367.jpg

looks like Scythe Andy Samurai Master

I really like that intel are diversifying more recently - from graphics, to chips with integrated nb's to massively multicore chips (rival to suns niagra in the server space?), as well as persuing their desktop architectures and (to a seemingly limited extent) Itanium.

Flexibility will be the key in the coming years, and by having the modular components ready they will be in a great position to mix and match elements according to market demand.

AMD bought ATi so that they would be ready for the changing landscape, and it appears that intel is doing their darndest to ensure they don't get caught off guard and stuck down an architectural path again.

Quote:

---

Originally Posted by gOJDO

http://i.n.com.com/i/ne/p/2007/CUTer...ip_550x367.jpg
http://i.n.com.com/i/ne/p/2007/demo_...ps_550x367.jpg

---

watta woman, that mobo.. :slobber:

ever gonna get some.. ?

Quote:

---

Originally Posted by NapalmV5

watta woman, that mobo.. :slobber:

ever gonna get some.. ?

---

yeah but 8 power connectors doesn't exactly screen energy efficient

Quote:

---

Originally Posted by nn_step

yeah but 8 power connectors doesn't exactly screen energy efficient

---

now thats some sexy curves.. :slobber:

And people complain about a PC's current power consumption - looking at all those thick wires.... ;)

Well that mobo will be a pain in the ass to insulate :(

Quote:

---

Originally Posted by nn_step

yeah but 8 power connectors doesn't exactly screen energy efficient

---

Its only 100W :p:

I bet Intel keeps their fingers crossed until quantum computer's test day :D

I think the ammount of wires has more to do with providing very precise voltages, rather than absolute power demands. As they said it outputs less than 100w of heat, and is on the 65nm process, theres no way it needs all that much power.

Wow 80 cores:slobber:

Quote:

---

Originally Posted by Supertim0r

looks like Scythe Andy Samurai Master

---

Yea I think you're right. It's crazy that they can use this on an 80 core CPU.

http://us.st11.yimg.com/us.st.yimg.c...s_1936_4873304

Probably low clocked cores, maybe round 400mhz, otherwise i wouldn't know how intel could cool that even if it would be 45nm.
This core structure has to be a lot differend then there current retail technologie.

Quote:

---

Originally Posted by ownage

Probably low clocked cores, maybe round 400mhz, otherwise i wouldn't know how intel could cool that even if it would be 45nm.

---

"Intel had the chip running at around 3GHz during the demonstration, but has gotten it to run faster in its Oregon labs with a water-cooled system, said CTO Justin Rattner."

I'm assuming that's 3GHz on air with the Scythe heatsink?

Quote:

---

Originally Posted by lawrywild

Stop stealing quotes from Jackass 2 :slapass:

---

This quote is really intense. Well not really intense but kinda intense :D

"3GHz"

3x80=240GHz/240,000MHz
oced - 3.75 x 80 = 300GHz/300,000MHz

someone get me a diaper..

Quote:

---

Originally Posted by gOJDO

http://i.n.com.com/i/ne/p/2007/CUTer...ip_550x367.jpg
:

---

WOW, does that socket ever look nice!!

Quote:

---

Originally Posted by ownage

Probably low clocked cores, maybe round 400mhz, otherwise i wouldn't know how intel could cool that even if it would be 45nm.
This core structure has to be a lot differend then there current retail technologie.

---

They're 80 simple cores - like many people have said in a rather similar way to how the cell is a 8 core cpu.

String a 1000 together for a Petaflop.

The thick wires are there because that mobo has no built in voltage regulators, and depending on the core voltage, there could be more then 100 amps running through those wires.

Quote:

---

Originally Posted by Lead Head

The thick wires are there because that mobo has no built in voltage regulators, and depending on the core voltage, there could be more then 100 amps running through those wires.

---

which can only mean that Power regulation for it is gonna be pricy

That looks really awesome. But why can't they just make single core processors much better? Instead of just adding more and more cores lol :p:?

Quote:

---

Originally Posted by nn_step

which can only mean that Power regulation for it is gonna be pricy

---

NO! It can only mean that it still is in development!

Quote:

---

Originally Posted by nn_step

which can only mean that Power regulation for it is gonna be pricy

---

Maybe not. Maybe they was working so hard on the CPU that completely forgot about the mobo :p:

Quote:

---

Originally Posted by gOJDO

NO! It can only mean that it still is in development!

---

Exactly! That huge box with the LED number readouts is a very precise power supply. It was obvious that this was thrown together very quickly and still in very early development.

Quote:

---

Originally Posted by Brettbeck

That looks really awesome. But why can't they just make single core processors much better? Instead of just adding more and more cores lol :p:?

---

And how do you imagine "much better single core CPU" ?

Quote:

---

Originally Posted by nn_step

which can only mean that Power regulation for it is gonna be pricy

---

Not at all, a well-made VRM is enough to provide power to this chip, but there's no VRM on the development PCB, so external power supplies have to provide lots of amps to the chip and it's not ideal because of ~100 A needed here (0.95 Vcore and ~100 W peak according to Intel). That's why there's a lot of capacitors near the chip to compensate losses induced by wires and difficulty to provide very stable voltage, regulation is a pain with such design implemented far from the chip.

Quote:

---

While it may take five to eight years to come out with a working 80-core chip, Vara says IT managers might start watching for what he calls "different flavors" of quad-core chips. "Maybe you'll have interim chips where they have more complex cores along with simpler cores, too."

---

.

;)

Quote:

---

Originally Posted by BlackX

And how do you imagine "much better single core CPU" ?

---

well that is exactly the point there are three walls that will forever limit performance.
1) the IPC wall (which is limited by the hardware design team)
2) the Thread wall (which is limited by the software programmers)
3) the Clock speed wall (Which is limited by Process and Ram)
Going for mega IPC will only result in low clock speeds and unused features (best example CISC)
The thread wall will seriously eat up Transistors that could be used to improve IPC (PS3 anyone?)
The clock speed wall demonstrated that the clock latency only gets worse as clock speeds increase (see Prescott)

Holy !@#$ I don't care if those are the simplest of cores, the fact that they have 80 of them on a single package and functioning at 3GHz is a feat in itself. In regards to the power management, have you looked at how bare the motherboard is? Just looks like me like they threw something together for the press release, current quad-cores can easily suck up to 100 amps of power but it's all done through onboard regulation and then distributed through dozens of traces/pins. Give Intel time and you'll see a better developed socket with onboard power regulation. This early in the game you can't expect miracles ;)

Probably every year we will see new CPU with double number of cores.. 4..8..16..32..

Quote:

---

Originally Posted by BlackX

Probably every year we will see new CPU with double number of cores.. 4..8..16..32..

---

2,048... 4,096... 8,192... :p:

:eek: no comment

I dont like the doubling of cores.
If you overclocked a northwood you could keep up with a later prescott, even the higher clocked ones. But now you can overclock your conroe but you cant keep up with the kentsfields.
From now on you cant keep up with the newer CPU's because of the doubling of cores.

Q - how will the chip function?

80 threads?
10/thread x 8?
5/thread x 16?
cpu/gpu hybrid?

even years from now the way programmers are treating multithreaded cpus.. only windows + 2 other apps will take advantage of 80-threads..

Quote:

---

Originally Posted by ownage

I dont like the doubling of cores.
If you overclocked a northwood you could keep up with a later prescott, even the higher clocked ones. But now you can overclock your conroe but you cant keep up with the kentsfields.
From now on you cant keep up with the newer CPU's because of the doubling of cores.

---

double the stock clock speed :rolleyes:

Quote:

---

Originally Posted by nn_step

double the stock clock speed :rolleyes:

---

I'm still busy with that :rolleyes:

It is 80 cores, but these are REALLY REALLY simple cores. The 80 of them combined have the same number of transistors as a single core Pentium 4. In fact, the cores are pretty much just simple stand ins to test the interconnect fabric until they can develop an X86 version. Having said that, they do each have 2 32-bit pipelined multipliers with accumulate, so they would make an insane DSP type chip (IE: graphics card) if they were beefed up a little more.

EDIT: and about the SRAM layer on top, look at the wafer, for every 80 core chip there is also another die that looks like it would be the SRAM cache which would be pasted on top of the 80 cores.

Isnt that old news? I think I read about that stuff a month ago.
As has been said, its a test to see if a 80 core cpu would work at all, because there are all kind of things that can go wrong.
From what I recall its even possible to disable parts of specific cores to prevent them from overheating and let other cores take the role of the disabled ones.

sweet i n eed that for wcg NOW

Quote:

---

Originally Posted by nn_step

cool but I thought they were also going to include stacked Cache with these

---

"Intel's prototype uses 80 floating-point cores, each running at 3.16GHz, Justin Rattner, Intel's chief technology officer, said in a speech following Otellini's address. In order to move data in between individual cores and into memory, the company plans to use an on-chip interconnect fabric and stacked SRAM (static RAM) chips attached directly to the bottom of the chip, he said."

http://news.zdnet.com/2100-9584_22-6119618.html

http://pic.xfastest.com/sxs112/09.jpg
64Core

Source

Ok considering you need 512Mb of ram per core for distributed computing, you would only need 32768Mb for that 64core ^ ^

But it looks damn sexy. Too bad that a quadcore is already overkill for most apps, because of bad multicore support :/
For distributed computing, such a beast would be definatly a nice addition.

Edit: According to that screenshot, that thing has 134gb of ram, so it must be a cluster of some sort, right? Dont think that much Ram would fit into 16 dimms.

Quote:

---

Originally Posted by sxs112

http://pic.xfastest.com/sxs112/09.jpg
64Core

Source

---

and 64 CPUs networked together has to do with 1 80core CPU how?

LOL at that pic

Quote:

---

Originally Posted by Fr3ak

Ok considering you need 512Mb of ram per core for distributed computing, you would only need 32768Mb for that 64core ^ ^

---

So even if these things somehow work out to sell for say under $3000, were still gonna be looking at around 5k for the ram *eye falls out*


Unless.....by the time these things come around to the consumer market, well all probably be using solid state hard drives as ram :p: :rolleyes:

interesting CPU although its not an X86 arch.
might prove useful for research though, as Itanium :rolleyes:

I think the real issue here, as NapalmV5 pointed out, is: How many threads are we talking about here?

I wonder how much ram you would need to run an 80 core setup ? 40GB ? :eek:

Quote:

---

Originally Posted by sxs112

http://pic.xfastest.com/sxs112/09.jpg
64Core

Source

---

omg 128 gb of ram :D

80 cores for 1 teraflop...

i thought g80 or r600 can do 500gflop already?

Quote:

---

Originally Posted by vitaminc

80 cores for 1 teraflop...

i thought g80 or r600 can do 500gflop already?

---

The R580 does 360 http://www.peakstreaminc.com/referen...m_technote.pdf
well the Xeon 5160 does 24 and Cell does 256.
and I couldn't find any numbers on the 8800GTX

But these 80 cores occuppies a smaller die size than the g80, so GFLOP/die size is better on the intel chip. Of course when it comes to practical use the G80 is far better, given theres probably very little code that the intel chip could run, let alone run optimally.

Quote:

---

Originally Posted by onewingedangel

But these 80 cores occuppies a smaller die size than the g80, so GFLOP/die size is better on the intel chip. Of course when it comes to practical use the G80 is far better, given theres probably very little code that the intel chip could run, let alone run optimally.

---

very true, it is far easier to optimize for a GPU than it is for many many CPUs

Quote:

---

Originally Posted by nn_step

very true, it is far easier to optimize for a GPU than it is for many many CPUs

---

I don't see why that would nescecarrily be true, modern graphics card pretty much are like a massively multicore processor, and the code they run is very easily parallelizable. I see absolutely no reason why an 80 core CPU couldn't have every core working jsut fine when running graphics code like pixel shaders, or raytracing so long as they all have acess to that top level cache and huge memmory bandwidth in order to keep up with all the cores. This particular processor wouldn't work since it doesn't have high enough precision ALUs, but its just a proof of concept anyways, so thats to be expected.

It is similar to Itanium (Also uses VLIW (Very Low Instruction Word)) in concept. They seem to be simple, small and efficient Cores, but they kick a whole bunch of complexity of it to the compiler and programmer that must do all the code optimization job if they expect to get any amount of decent performance.
Just think on most lazy programmers that are developing commercial Software (Usually games) that becomes bloatware pretty fast demanding an awful amount of resources due to the lack of optimizations for rushing it to market. Definately not for general purpose.

Quote:

---

Originally Posted by nn_step

The R580 does 360 http://www.peakstreaminc.com/referen...m_technote.pdf
well the Xeon 5160 does 24 and Cell does 256.
and I couldn't find any numbers on the 8800GTX

---

8800GTX should have more, but ya, not sure how much more.

Quote:

---

Originally Posted by onewingedangel

But these 80 cores occuppies a smaller die size than the g80, so GFLOP/die size is better on the intel chip. Of course when it comes to practical use the G80 is far better, given theres probably very little code that the intel chip could run, let alone run optimally.

---

Not sure if die size are that much different?

80 "cores" or 128 "pixel shaders". whats the difference? they are just naming simple processing units differently, except NVIDIA/ATI has devoted a lot of embedded software support while Intel is "lagging" behind somewhat.

Quote:

---

Originally Posted by zir_blazer

It is similar to Itanium (Also uses VLIW (Very Low Instruction Word)) in concept. They seem to be simple, small and efficient Cores, but they kick a whole bunch of complexity of it to the compiler and programmer that must do all the code optimization job if they expect to get any amount of decent performance.
Just think on most lazy programmers that are developing commercial Software (Usually games) that becomes bloatware pretty fast demanding an awful amount of resources due to the lack of optimizations for rushing it to market. Definately not for general purpose.

---

VLIW mean Very LONG Instruction Word. And "simple, small and efficient" describes everything Itanium is NOT.

I duno what to say. Great there really pushing. God know what they have done exactly, let alone how. seam fitting quad onto the LGA775 is pretty hard currently and die srink is needed so unless this ''cpu'' is 80 cores only and the catch is the ram or soemthing whom knows but scray the fact each core wants to know what the other is doing...

To make better to anyone to understand how they achieved that look in this video below.:)

Intel 80 cores Video Demonstration :slobber:

Quote:

---

Originally Posted by zir_blazer

It is similar to Itanium (Also uses VLIW (Very Low Instruction Word)) in concept. They seem to be simple, small and efficient Cores, but they kick a whole bunch of complexity of it to the compiler and programmer that must do all the code optimization job if they expect to get any amount of decent performance.
Just think on most lazy programmers that are developing commercial Software (Usually games) that becomes bloatware pretty fast demanding an awful amount of resources due to the lack of optimizations for rushing it to market. Definately not for general purpose.

---

:rofl: :lmao: :lol: it isn't even close to VLIW
The Reasons why that is funny (all about Itanium)
1) it uses 592,000,000 transistors which is more than kentsfield
2) it is in the 130 watts evelope, once again worse than any x86 proc
3) it originally only produced 3.2 GFLOP of work and now it is only at 6.4 GFLOP, FX-53 cranked out 4.8 @ 2.4Ghz single core
4) VLIW is the hardest thing possible to program for, Even if you spend thousands of hours making 100% usage of the hardware; the next version of it wouldn't be able to run that same code & if it could, there is no way in hell that it would use the potential of the hardware.
5) VLIW is like a boxer with a Glass jaw, even though it looks good on paper; it goes down in the first round every time

Quote:

---

Originally Posted by BrownTown

VLIW mean Very LONG Instruction Word. And "simple, small and efficient" describes everything Itanium is NOT.

---

That's my typo. Itanium logical Core (Ignore Cache) is smaller than that of an x86 Core. As it kicks instruction schedule responsability to the compiler, the logical Core itself is smaller and less complex than a similar x86 Core. It is efficient if you have a good team of programmers optimizing by hand an application code, that doesn't means that it is practical for everything though.

Quote:

---

Originally Posted by nn_step

:rofl: :lmao: :lol: it isn't even close to VLIW
The Reasons why that is funny (all about Itanium)
1) it uses 592,000,000 transistors which is more than kentsfield
2) it is in the 130 watts evelope, once again worse than any x86 proc
3) it originally only produced 3.2 GFLOP of work and now it is only at 6.4 GFLOP, FX-53 cranked out 4.8 @ 2.4Ghz single core
4) VLIW is the hardest thing possible to program for, Even if you spend thousands of hours making 100% usage of the hardware; the next version of it wouldn't be able to run that same code & if it could, there is no way in hell that it would use the potential of the hardware.
5) VLIW is like a boxer with a Glass jaw, even though it looks good on paper; it goes down in the first round every time

---

If it is not VLIW, what Dailytech is saying that it is?

Quote:

---

A major limitation of the current 80-core chip is that it is not based on the X86 architecture. Instead, it uses a 96-bit Very Long Instruction Word (VLIW) architecture, another architecture currently used in the Itanium server processors. A major hurdle that Intel hinted at will be moving from VLIW to X86 on its 80-core chip.

---

Again, when doing Transistors-based comparision, remove the Cache, as that 600M Transistors (Montecito I suppose?) monster got obviouly tons of Cache L3 (It will cripple its performance but that is not what I am comparing). Anandtech did an article with a in-depth architecture comparision of IA64 vs x86, but I wasn't able to find it after 30 minutes searching.
Intel did the same with this prototype, it is similar to Itanium as its trades lot of Core logic for simplicity of each them, but goes for massive thread parallelization instead of massive instruction parallelization like Itanium did.

Quote:

---

Originally Posted by Metroid

To make better to anyone to understand how they achieved that look in this video below.:)

Intel 80 cores Video Demonstration :slobber:

---

is there a way to download it?

Quote:

---

Originally Posted by -Sweeper_

is there a way to download it?

---

right click->save as

Quote:

---

Originally Posted by BrownTown

right click->save as

---

tks :p:

Now where's that demo of AMD quad core for all to see? Intel just made me happy that I brought a new Conroe rig. AMD is going to start losing money again if they don't show something bad ass.

Quote:

---

Originally Posted by safan80

Now where's that demo of AMD quad core for all to see? Intel just made me happy that I brought a new Conroe rig. AMD is going to start losing money again if they don't show something bad ass.

---

hm k10 will be soon on the market, i suppose an early q3 launch for the 'stars'.
unless intel bump considerably the speed of its 45nm, the green side will have its thrown back. :fact:

Quote:

---

Originally Posted by zir_blazer

That's my typo. Itanium logical Core (Ignore Cache) is smaller than that of an x86 Core. As it kicks instruction schedule responsability to the compiler, the logical Core itself is smaller and less complex than a similar x86 Core. It is efficient if you have a good team of programmers optimizing by hand an application code, that doesn't means that it is practical for everything though.



If it is not VLIW, what Dailytech is saying that it is?



Again, when doing Transistors-based comparision, remove the Cache, as that 600M Transistors (Montecito I suppose?) monster got obviouly tons of Cache L3 (It will cripple its performance but that is not what I am comparing). Anandtech did an article with a in-depth architecture comparision of IA64 vs x86, but I wasn't able to find it after 30 minutes searching.
Intel did the same with this prototype, it is similar to Itanium as its trades lot of Core logic for simplicity of each them, but goes for massive thread parallelization instead of massive instruction parallelization like Itanium did.

---

Technically it is a MPTP (Massively Parallel Threaded Processor) since there isn't an explicit requirement of Instruction level instructions. The parallelism that is exhibits is thread level not instruction level (which is what VLIW is)
I know it is picky but it is a very distinct difference when programming for it.
Itanic was instruction level parallel and this is a Thread machine gun, for lack of a better term.

Link from big anands article on the subject http://www.anandtech.com/cpuchipsets...oc.aspx?i=2925
________
Mexico Hotels

Quote:

---

Originally Posted by nn_step

yeah but 8 power connectors doesn't exactly screen energy efficient

---

Just seen this on the news...

Quote:

---

Originally Posted by News

There was a prolonged blackout this afternoon of up to 6 hours in some areas said PSEG, the cause according to authorititys was NN_steps new 80 core pc.

---

keke :rofl: