Quote:
Originally Posted by DoubleK
Great posts by SeeGeeWhy.
One of the projects I'm working on is in a 'data center' park near Reno. The size of these facilities is unlike anything I've seen in my life. I was as shocked as I was the first time I drove through Sherwood Park and saw the size of the refineries.
One of the emerging issues is the impact that these data centers are having on electricity costs. Arizona is looking at setting a separate tariff for data center customers and they are driving up power prices/causing supply shortages in Washington state.
The massive loads of these data centers are suspected to cause power system harmonics issues that will also need addressing. I do think that this is very similar to the issues Alberta is experiencing with inverter based resources causing frequency issues. Both of these are engineering challenges that are fixable, but it's something that I am keeping an eye on. |
The cluster that xAI setup was so big that they needed to install Megapacks to actually power the cluster, and the grid / diesel generators charged the Megapacks.
Megapacks can deliver consistent baseload power needs. There just needs to be enough of them to power the facility for x amount of time, and then renewable sources (solar, wind, etc) can be used to charge the Megapacks, with other non renewable sources also available for the brunt of it, i.e. natural gas.
Quote:
The electrical grid can easily crash with electrical volatility. Think about how power surges can fry your computer or your appliances.
Non-AI datacenters, with steady year-round power usage, were grid-friendly.
By leveraging Megapacks, $xAI effectively reduces grid pressure: “xAI found millisecond power fluctuations when GPUs start training, causing issues with power infrastructure.
The solution was to use generators to charge the Tesla Megapacks, which then discharge to power the training jobs.
|
https://nextbigfuture.substack.com/p...ai-data-center