You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
AI systems consume far larger amounts of power for processing and cooling than current data centers and systems do now. From an engineer familiar with the grid, it looks like the grid will be subject to major challenges to meet the power consumed by AI data centers and heavy users of AI. Any thoughts on this challenge and how it can be met will be appreciated, Thanks.
Also, the largest data center campus in the world is located in Tahoe Reno, Nevada and has a total of 7.2 million square ft. Far less need for air conditioning up there. Owned by Citadel
Same as other very large industrial consumers deal with their power requirements - Cogeneration. AI data centers in the future will generate most of their own power on site and only be connected to the grid for backup/emergency power.
Same as other very large industrial consumers deal with their power requirements - Cogeneration. AI data centers in the future will generate most of their own power on site and only be connected to the grid for backup/emergency power.
Same as other very large industrial consumers deal with their power requirements - Cogeneration. AI data centers in the future will generate most of their own power on site and only be connected to the grid for backup/emergency power.
Is cogeneration the right term? "Cogeneration or combined heat and power is the use of a heat engine or power station to generate electricity and useful heat at the same time. Cogeneration is a more efficient use of fuel or heat, because otherwise-wasted heat from electricity generation is put to some productive use."
Maybe cogeneration is not the right term, but on site generation. Most likely natural gas because you need baseload power but I'm sure they will throw a few solar panels on the roof to greenwash. A lot of mining companies do things like this. In the case of a mining company it is cogeneration because they are also using the heat. For data center they don't need the heat, they just need the power, but maybe they can sell the heat to nearby consumers.
I think as data centers scale up they will probably rethink locations. Maybe they will build data centers in places with a lot of good solar like in west Texas. Texas also has a lot of natural gas. This is already what a lot of bitcoin farms do.
Nevada also makes sense for a lot of the same reasons. Lots of solar.
AI systems consume far larger amounts of power for processing and cooling than current data centers and systems do now. From an engineer familiar with the grid, it looks like the grid will be subject to major challenges to meet the power consumed by AI data centers and heavy users of AI. Any thoughts on this challenge and how it can be met will be appreciated, Thanks.
Also, the largest data center campus in the world is located in Tahoe Reno, Nevada and has a total of 7.2 million square ft. Far less need for air conditioning up there. Owned by Citadel
Power was a big topic at the Intel event. I imagine TSMC will focus on it as well since they have a clear low power advantage (mobile SoCs). The AI boom is definitely going to stress the power grids around the world. Kind of like the crypto boom did but probably 1,000x more. The difference being AI will literally be everywhere and it will be an arms race of sorts for both military and commercial uses. AI will not only consume a huge amount of silicon, it will consume a huge amount of power as well. Silicon is the easy part, power will be difficult, especially with the violent weather we have had here in the US. Another big storm will hit this weekend with 10ft+ of snow and 100MPH winds in the Sierras so buckle up Art.
Agreed. For decades I've thought the nuclear power industry could have learned a lot from Intel - Copy Exactly! Instead, every new nuclear plant is a hugely complicated one-off design. SMRs need to be the future, ASAP.
Agreed. For decades I've thought the nuclear power industry could have learned a lot from Intel - Copy Exactly! Instead, every new nuclear plant is a hugely complicated one-off design. SMRs need to be the future, ASAP.
Interestingly the oil and gas industry seemed to finally this lesson. This is how they were able to successfully scale up unconventional oil plays with highly standardized modular well pads. So it can be done.
There are several types of small nuclear reactors being created and that could be mass produced and I feel this is the best answer to the challenges ahead. I also wonder if there might be a way developed to harvest power from lower temperature differentials with data centers in cold areas generating significant heat. Any thoughts on this would be appreciated. Might there be a way to generate electricity by solid state panels to take advantage of heat differentials? This would be a new frontier for the semi industry.
Is there any way to increase the efficiency at lower temperatures of thermocouples to generate power from the waste or excess heat and recycle it back into the servers? What about a sealed very low pressure gas operated power system? The temperature differentials, especially in winter in the Tahoe area would be fairly large.
Our aircraft carriers and submarines have proved safe and reliable since the submarine Nautilus was launched in 1954 and operated without incident for years.
Is there any way to increase the efficiency at lower temperatures of thermocouples to generate power from the waste or excess heat and recycle it back into the servers? What about a sealed very low pressure gas operated power system? The temperature differentials, especially in winter in the Tahoe area would be fairly large.
