Intel Reinvents Transistors Using New 3-D Structure

lotsa stuff
People are just silly and obviously don't read or understand what graphene transistors would compete against. ;) ZOMG 100GHz! 300GHz cut off point!

FUN FACTS: EXACTLY HOW SMALL (AND COOL) IS 32 NANOMETERS?

A 32nm transistor can switch on and off over 300 billion times in one second. It would take you 4000 years to flick a light switch on and off that many times5.
...
5 Assumes a person can flick a light switch on and off 150 times per minute.
That's something that has been in production for over a year and 22nm tri-gate is even faster. B-b-b-but how can this be? It's not graphene!!!!1
 
Intel rocks, now lets see em make a vid card with lasting performance that performs and will run max settings for more than 3 years on top dog models.....DOOOOOO IT! ;)
 
Intel rocks, now lets see em make a vid card with lasting performance that performs and will run max settings for more than 3 years on top dog models.....DOOOOOO IT! ;)

It'll be a few years, but I have a feeling that we won't be seeing discrete video cards as we know them today. It'll probably be incorporated into the CPU and later on a complete SoC, and while many, including myself, find it hard to believe that an integrated graphics solution could give us great performance, I think it'll get there as technology progresses. I mean we're already seeing Intel and AMD make huge pushes in that direction with Sandy Bridge and Bulldoze, and while its not up to snuff for enthusiast usage, like I said I think it'll get there.

I know I'm not taking into consideration architecture issues and the like, or the GPU vs CPU issues, but if anything can be gleaned from technology innovations like this 3D transistor development, its that we shouldn't limit ourselves to the methodology that is being used today.

On a related note, look at who is failing miserably in terms of market share and overall outlook: Nvidia. Look who is gaining considerably and have set themselves up considerably well for the future: AMD and Intel. I know its weird to think of now, as I own an Nvidia SLI setup with 2xGTX460 and LOVE it, but I just don't see Nvidia staying around in the next 5-10 years with the current business model they're running. They lost out big on the chipset business and well, the SLI licensing to AMD should tell you a few things as well :eek:.
 
A CPU has no where near the memory bandwidth to compete with videocards. A videocard today is pushing 160GB/s that's gigabytes per second. CPU's have no where near that bandwidth.

If a CPU can have 200GB or faster then on die GPU with CPU can compete with videocards.

I just don't see it happening on a CPU even in a few years. Maybe 10+ years our possibly if ever.
 
Intel does something Apple can't do. They can push new tech down your throat and you like it.

x86 is in a prime spot to take over the smartphone market. Hopefully they start going into the market by the time my contract is up on my Evo Shift.
 
Intel does something Apple can't do. They can push new tech down your throat and you like it.

x86 is in a prime spot to take over the smartphone market. Hopefully they start going into the market by the time my contract is up on my Evo Shift.

this may help but ARM still rules here. for that matter AMD still is actually closer then Intel here. but ARM is so far ahead the idea of a single process advancement closing the gap just doesn't fly.
 
"According to Intel, the 3D Tri-Gate transistors yield up to 37 percent higher performance over today's 32nm planar transistors. Further, these transistors consume less than half the power as the planar variants."

http://www.infoworld.com/t/processors/intel-transitions-chips-third-dimension-019

This will NOT translate into a 37% increase in IPC all by itself. They will have to change other things as well.

I am also curious if it is 50% less power used at the same clock speed or if they are going for 50% less power at the same performance level (lower clocks).

Does this mean that we will start seeing CPUs that run at temperatures that are similiar to what AMD processors already run at?

I am at work now so I will have to watch the video later.
Sounds like apples to oranges... They are comparing 32nm chips to 22nm chips AND 2D to 3D gated transistors. Once you subtract the power differences from the die shrink, the gains are significantly less impressive.

All new tech takes time to stabilize, so whatever generation first receives these improvements may not realize the benefits for a few years until they perfect the technology. Will be interesting to see what kind of improvement this eventually yields.
 
How does this compare to SOI, which as they note also achieves full depletion on the channel?
 
How does this compare to SOI, which as they note also achieves full depletion on the channel?
AMD/GF uses partial depleted SOI (PDSOI). FDSOI is a completely different animal. GF is dropping PDSOI after 32nm and moving bulk silicon for 22nm (its 28nm is already a bulk silicon process).

Intel's HiK-MG and prior processes (strained Si) had surpassed AMD/GF PDSOI transistor performance years ago. And now tri-gate transistors at 22nm cut power usage by "50% at the same performance" over Intel's HiK-MG at 32nm.
 
Once you subtract the power differences from the die shrink, the gains are significantly less impressive.
Intel showed that in the presentation. I'm pretty sure Intel's PDF of it was linked earlier in the thread.

BTW, it's still very impressive vs planar transistor performance at 22nm. Yesterday I read a better description of what tri-gate accomplished: at 22nm it effectively gave Intel a 2 node jump over 32nm instead of one. i.e. 22nm tri-gate gives 14nm (planar) power and performance. ;)
 
AMD/GF uses partial depleted SOI (PDSOI). FDSOI is a completely different animal. GF is dropping PDSOI after 32nm and moving bulk silicon for 22nm (its 28nm is already a bulk silicon process).
So do we know how it compares to fully depleted SOI?
 
No one said it wouldn't work, dude. That is a research project, a science experiment. It is nowhere near ready for production or even competitive with silicon:
The transistor gates are only 550 nanometers long, and the entire "wafer" is the size of a grain of salt, IBM says
Quotes added as appropriate. 500nm gates are very early 1990s production, and research from the 1980s to put it in perspective.

Intel is also evaluating graphene, but obviously didn't feel the need to breathlessly release a paper.
http://download.intel.com/newsroom/...search_Enabling_ Breakthroughs_Technology.pdf (see slides 11/12, especially notice "now" on slide 12 and peek at slide 10 for some of the challenges of the possible next material [III-V] which is in far more advanced stages than graphene)

Graphene is way out into the future, if it even turns out to be feasible for use in mass production of MPUs.
 
yeah sorry I'm trolling a little bit. but to be completely honest I just think these companies are being lazy and super greedy.
 
Someone else will make something equivalent and cheaper. No one is going to pay a premium on licensing. That's why light peak, ia64, and all these other proprietary Intel technologies will never take off.
 
Back
Top