Graphene 1thz CPU + Oculus VR

[McTurbo]

http://wccftech.com/graphene-transistors-427-ghz/

while we are still years away. i cant help but be giddy at the thought of a 1 thz cpu and what graphene is going to do to electronics such as Oculus VR. 8 to 12k resolution, with graphics better than Avatar, and a world bigger than GTA V..

Think 1 thz cpu (20+ cores) solid state drives. (your cellphone will be far more powerful than the best personal computer we have now...) now hook that up to VR and your not locked into a seat 😛

the future is exciting and i cant wait to play in it!

[Alci]

that's a reason why we stopped at around 2GHz. And only very small part of processor (like ALU) can run faster. Also reason why we went the road of simplifying PUs and parallel processing instead. Because GPU can run hundreds of
"threads" at once. Same as internet and any network is an answer for gigantic TFlops needs. Not a single chip.

There is small space for improvement with density. And some opportunity if we could make 3D chip. But there are obvious problem with cooling.

Cheap hispeed transistors will find better usage at hi frequency transmitters/receivers and measurement devices.

Also the very first link is full of nonsense about how CPU rendering is superior to GPU rendering and THz processors. While they both are just computation units and both will compute pixels exactly in the way you tell them to do so.

[us11csalyer]

So which is more likely for future gaming:

A: multiple cpus (like some servers) that run in par like duo and quads do today

or

B: A single cpu that is 100Ghz "fast"

Personally I think we might see a gaming rig use a single cpu that handles a few other cpus that do all the work, and handling the i/o to video cards. I think this would be amazing. One cpu to handle everyday computing, and 2+ more processors each being say 20 cores for gaming/rendering/VR. The everyday cpu would be running the OS/AV/ect.

[McTurbo]

i think you will see option B in the long run. we already are doing option A for servers.

even though i linked talk about 1thz cpu's i dont think thats where we will start off.. even if we started off between 200-500 ghz we will still be in place that is amazing.

[RiftXdev]

A friend of mine is a research engineer in the nano materials industry and he informs me that graphene is a long way off from being a viable production material.

It's old news these days but the reality is carbon nano tubes is still very much the focus, certainly in his industry (developing manufactureable nano scale materials).

[geekmaster]

Considering that a human body emits about 28 THz thermal radiation (same frequency as a CO2 laser), and room temperature thermal radiation is significantly lower frequency, I suspect that electronic circuitry operating in the multi-THz range will have to take electrothermal coupling and feedback effects into account. Perhaps it could use such properties to advantage by design, rather than as a technical hazard to be avoided.

[alejux]

"RiftXdev" wrote:
A friend of mine is a research engineer in the nano materials industry and he informs me that graphene is a long way off from being a viable production material.

It's old news these days but the reality is carbon nano tubes is still very much the focus, certainly in his industry (developing manufactureable nano scale materials).

This happened just this month.

http://www.extremetech.com/extreme/179874-samsungs-graphene-breakthrough-could-finally-put-the-wonde...

On the micro-electronics industry, there is A LOT of effort being done with graphene. This kind of technology is advancing pretty fast. Five years from now, we could be talking about a totally different scenario concerning all this.

[lossofmercy]

Again, this isnt a cpu. This is 3 logic gates. Its not the same comparison.

[Calanar]

"lossofmercy" wrote:
Again, this isnt a cpu. This is 3 logic gates. Its not the same comparison.

Yes to show that it worked. The crystal method just pioneered by Samsung apparently works in thin sheets though and that should be able to achieve the densities we need for computation. There are several breakthroughs being worked on at once. When they come together someone is going to make a fortune and they know it. Just as SSDs are pricey but have more performance than harddrives so too will Graphene CPU/GPUs in 3-10 years be pricey but ridiculously better performing.

[lossofmercy]

Again, there are a ton of additional issues in a normal cpu. I know that IBM managed to get one of the old chips at 350 ghz/sec in 2002:

http://www-03.ibm.com/press/us/en/pressrelease/444.wss

[mstdesigns]

"lossofmercy" wrote:
Ok, after some more research... good attempt but try again:
http://www.reddit.com/r/technology/comments/1kr5fw/mit_got_graphene_to_behave_like_a_transistor/cbs5...

So, let's take that 845 GHz number. Loosely speaking, that's a single transistor changing states 845 billion times a second, meaning a single state change takes 1/845,000,000,000th of a second.
Let's suppose for a moment that information could be transmitted through the chip at the absolute fastest speed possible, the speed of light. In that 1/845,000,000,000th of a second, information could have traveled a maximum of 0.354 millimeters.
For comparison, Intel CPU dies for the current 22nm node range from 160 to 264 mm2. If the die is a perfect square, that comes out to 12.65 to 16.2mm on each side. That means that the longest dimension is going to be 17.9 to 23mm, from corner to corner.
That means that in this ideal scenario, where our information is traveling at the speed of light, the results of one calculation will take between 50 and 65 cycles to reach the other end of the chip, and twice as many cycles before the signal can get all the way back to the original end of the chip.
I hope that helps to show why the speed of a single transistor is pretty meaningless when we're talking about an entire CPU. It's kind of like quoting the fastest speed a tire has ever gone, probably by being shot out of a cannon or something similarly ridiculous. It tells you nothing about how fast an entire car can travel, because a single tire and an entire car simply aren't comparable.

The point is the information does not have to reach the ends of the chip at all times. I imagine it could work with 10,000 cores or more working on independent tasks and exchanging information only when needed. For sure the current architectures would need a lot of modifications but it certainly is doable.

The main problem will be that the memory will not get the same hundredfold boost. I can see cache taking up as much as 90% space inside the graphene chip!