Synthetic Diamonds may make the fastest and coolest CPU's yet

[GoodBoy]

Binding microchips (including GPU dies) with diamond and other unique methods allows for much lower temperatures and up to three times the performance.
Lab tests on an undisclosed high-end Nvidia GPU reportedly granted three times the performance as a normal chip based on standard manufacturing materials, which is insane but a little hard to believe without publicly available benchmarks.

Full read:
https://www.tomshardware.com/tech-i...e-glass-may-make-the-coolest-fastest-cpus-yet
original source is WSJ but it is behind a paywall.

 

[Dan_D]

Binding microchips (including GPU dies) with diamond and other unique methods allows for much lower temperatures and up to three times the performance.
Lab tests on an undisclosed high-end Nvidia GPU reportedly granted three times the performance as a normal chip based on standard manufacturing materials, which is insane but a little hard to believe without publicly available benchmarks.

Full read:
https://www.tomshardware.com/tech-i...e-glass-may-make-the-coolest-fastest-cpus-yet
original source is WSJ but it is behind a paywall.

At face value significantly lower temperatures would allow for a great deal of increased performance so its hard to say exactly how much of an impact the technology would have.

 

[cdabc123]

At face value significantly lower temperatures would allow for a great deal of increased performance so its hard to say exactly how much of an impact the technology would have.

Would it? Excluding sub zero temps where electrons get funky, a 4090 cant clock much higher at 30c vs 70c

 

[T_A]

Name of the game is lower power consumption and efficiency

 

[TheBuzzer]

interesting, so lower temp allow them to overclock it 3x compared to normal gpus. or did they also watercool it and stuff.

 

[Lakados]

Thermal transfer is a big deal, the run of the mill 120mm AiO is more than capable of handling a heck of a lot more heat than anything that Intel or AMD is selling retail, We have a long way to go before we thermally saturate those things let alone the 360's or any of the similar 140mm fan variants. Anything that lets that heat get transfered out faster the better, and if it provides for cleaner power delivery that means less hardware onboard to deal with the noise power delivery creates which means fewer dummy gates which means more actual ones so faster chips which generate more heat and the cycle continues.

 

[LukeTbk]

If any of this is true, I would imagine it is far to be just because of cooler temp, liquid nitrogen cooled Nvidia gpu cannot triple their frequency or performance. Like the not sure if well written text of tomhardware that leach WSJ say: Instead of cooling, purified glass is instead being used for improving power efficiency and communication between chips...

The fact the GPU is made from material that conduct electricity better and do not generate the heat to start with is probably much more the reason here, than just lower temp. Better material with higher conductivity providing giant gain, sure one would believe that, super-conductor could easily multiply by 10 the computer per watt one could imagine, almost all of the power being lost in heat production right now, with imperfect but much better in between doubling-tripling, why not, how many hundreds of thousands does those wafers cost right now too... ?.

 

[Lakados]

If any of this is true, I would imagine it is far to be just because of cooler temp, liquid nitrogen cooled Nvidia gpu cannot triple their frequency or performance. Like the not sure if well written text of tomhardware that leach WSJ say: Instead of cooling, purified glass is instead being used for improving power efficiency and communication between chips...

The fact the GPU is made from material that conduct electricity better and do not generate the heat to start with is probably much more the reason here, than just lower temp. Better material with higher conductivity providing giant gain, sure one would believe that, super-conductor could easily multiply by 10 the computer per watt one could imagine, almost all of the power being lost in heat production right now, with imperfect but much better in between doubling-tripling, why not, how many hundreds of thousands does those wafers cost right now too... ?.

From what I have read, Intel's current Glass substrate research it's about reducing power bleed and interference to as little as possible, as much as 40% of an Intel CPU's surface area is made up of either insulating components or additional error correction to filter out the bleed from power delivery and gate noise. The Diamond substrate work is an extension of that research, but cost prohibitive to a degree where it isn't feasible because growing Diamond is not cheap, this could allow Intel to do some absurd things with data center density so it's something they need to work on as there is only so much more they figure they can do with Silicon in the current state of things.

 

[GoodBoy]

I don't know how they are applying the Diamond layer. If they have figured out a way to grow diamond crystal from carbon vapor it sounds like something that could be mass produced. It's not the same thing as a decorative diamond, it's sole value would be in the performance improvements that it enables. Creating the decorative diamonds is what costs more to achieve than the output diamond is worth. This might be cheaper to do.

Give it 5 years and see where it has led.

 

[Twisted Kidney]

I don't know how they are applying the Diamond layer. If they have figured out a way to grow diamond crystal from carbon vapor it sounds like something that could be mass produced. It's not the same thing as a decorative diamond, it's sole value would be in the performance improvements that it enables. Creating the decorative diamonds is what costs more to achieve than the output diamond is worth. This might be cheaper to do.

Give it 5 years and see where it has led.

I wonder how you would vaporize a diamond. Their whole schtick is that they absorb very little energy. I wonder how they'd excite the diamond enough to get it to sublimate.

 

[Tsumi]

I'm going to wager that 3x performance is in thermally limited situations, i.e. laptops. Expelling heat faster from laptops, especially high performance ones, will allow them to maintain boost states longer.

 

[StoleMyOwnCar]

https://www.eetimes.com/diamond-shines-in-high-power-devices/

Looks like it's been a thing in other places. Is the thermal performance primarily due to the thermal conductivity, or are there other advantages? I've been reading about the "critical electric field" thing mentioned in this article, but although some of these terms sound familiar, it has simply been far too long since college.

 

[OutOfPhase]

I wonder how you would vaporize a diamond. Their whole schtick is that they absorb very little energy. I wonder how they'd excite the diamond enough to get it to sublimate.

Show it a RTX4900 for 250 dollars.

 

[Tsumi]

https://www.eetimes.com/diamond-shines-in-high-power-devices/

Looks like it's been a thing in other places. Is the thermal performance primarily due to the thermal conductivity, or are there other advantages? I've been reading about the "critical electric field" thing mentioned in this article, but although some of these terms sound familiar, it has simply been far too long since college.

According to Google and Wikipedia, natural diamond is 5 times more thermally conductive than silver. Synthetic is about 7.5 times silver and only gets more thermally conductive as temperatures drop. I imagine the most difficult part is shaping it into a good mating surface.

 

[Lakados]

I wonder how you would vaporize a diamond. Their whole schtick is that they absorb very little energy. I wonder how they'd excite the diamond enough to get it to sublimate.

I'm going to wager that 3x performance is in thermally limited situations, i.e. laptops. Expelling heat faster from laptops, especially high performance ones, will allow them to maintain boost states longer.

Here is one of the industry leaders in the field and their high level breakdown on it.

https://compoundsemiconductor.net/article/116102/Creating_commercially_viable_diamond_substrates

It talks a bit about how they make the substrate and its properties.

 

[LodeRunner]

I don't know how they are applying the Diamond layer. If they have figured out a way to grow diamond crystal from carbon vapor it sounds like something that could be mass produced. It's not the same thing as a decorative diamond, it's sole value would be in the performance improvements that it enables. Creating the decorative diamonds is what costs more to achieve than the output diamond is worth. This might be cheaper to do.

Give it 5 years and see where it has led.

I wonder how you would vaporize a diamond. Their whole schtick is that they absorb very little energy. I wonder how they'd excite the diamond enough to get it to sublimate.

Chemical Vapor Deposition has been a way to make diamonds in various configurations and diamond compounds for a while now (several years at least to my admittedly faulty memory).

A light description of the process: https://www.alicat.com/synthetic-diamond-production-using-chemical-vapor-deposition/

 

[Twisted Kidney]

Chemical Vapor Deposition has been a way to make diamonds in various configurations and diamond compounds for a while now (several years at least to my admittedly faulty memory).

A light description of the process: https://www.alicat.com/synthetic-diamond-production-using-chemical-vapor-deposition/

Yeah, I was reading bits from here and there on the internet.

Materials science never fails to make me nerd out.

 

[Conman]

Hmm I can add this to my big list o' things that sound great but never seem to arrive. I'll put it right between the everlasting sodium ion battery and the miracle cancer drug that has a 99% cure rate in lab mice. Honorable mention? Room temperature super conductor.

 

[BrainEater]

While this is a novel implementation of the technology, diamond substrates are old news.

The repeaters that amplify fiberoptic signals under the ocean use diamond substrate thermoelectric cooled lasers.
---
Not only can CVD diamond be "sprayed" onto any refractory surface, the diamond can be " lattice-oriented" for thermal flow. Lattice oriented substrates can have thermal values of 1400w/cm +
I've actually priced these out ( you can order them).
a 1" diameter 1mm thick lattice oriented diamond disk was ~ $8000 US ( a few years ago).

 

[Lakados]

While this is a novel implementation of the technology, diamond substrates are old news.

The repeaters that amplify fiberoptic signals under the ocean use diamond substrate thermoelectric cooled lasers.
---
Not only can CVD diamond be "sprayed" onto any refractory surface, the diamond can be " lattice-oriented" for thermal flow. Lattice oriented substrates can have thermal values of 1400w/cm +
I've actually priced these out ( you can order them).
a 1" diameter 1mm thick lattice oriented diamond disk was ~ $8000 US ( a few years ago).

I think that’s why Intel is working with glass and not diamond. Last time I checked it was significantly cheaper.
Gonna be a long while before growing diamond in a lab is commercially viable.

 

[Gideon]

I think that’s why Intel is working with glass and not diamond. Last time I checked it was significantly cheaper.
Gonna be a long while before growing diamond in a lab is commercially viable.

Lab grown diamonds are cheap, use them all the time in industrial cutting equipment and such. Just gets massively more expensive when you want a certain size or lattice structure.

 

[GoodBoy]

https://www.eetimes.com/diamond-shines-in-high-power-devices/

Looks like it's been a thing in other places. Is the thermal performance primarily due to the thermal conductivity, or are there other advantages? I've been reading about the "critical electric field" thing mentioned in this article, but although some of these terms sound familiar, it has simply been far too long since college.

From the eetimes article, the relevant points:

“We synthesize and dope diamond epitaxial layers with a unique control; therefore, stacks of diamond doped layers are grown to form a high-value-added wafer ready for device fabrication,” Driche said.
Among all the industrial processes required to manufacture a diamond device, the growth of the epitaxial layer is one of the most critical because most of the electrical performance depends on the quality of these active layers.
Unlike most semiconductors, diamond has a decreasing resistivity with increasing temperature. Therefore, devices made from the material perform better at 150 degrees Celsius—the typical operating temperature for power devices—than at room temperature.
Diamond is also a good heat dissipator."

For power control components, like in an electric vehicle where it is very high current, the diamond made component is actually cheaper than the silicon.

For GPU or CPU, it's not the same use-case, but the semiconductor properties are still very useful.

Time will tell and all, where this goes for chips. It's already in use as a technology in Tesla's EV's.

If there is a big enough benefit and need, manufacturing will improve and costs will lower.