The CEO of chip giant AMD has said that the high-performance computing (HPC) industry needs to start becoming significantly more efficient, or it will need to turn to nuclear energy.
AMD CPU and GPU performance typically double every 2.4 years which is good progress by any means, but power efficiency has not seen such considerable improvements.
Speaking at International Solid-State Circuits Conference, Dr Lisa Su noted that with zettaflop HPCs on the horizon, we should turn our attention to performance per unit of energy in order to make zettaflop computation a reality.
Going nuclear
According to current trends and assuming that progress doesn’t slow down, AMD reckons that zettaflop computation could be here in about a decade (although Intel reckons it could be here in about half that time).
The reality is that when they arrive isn’t as important as when they will become available on a usable, energy-efficient scale.
Currently, Su generously estimates that an HPC capable of performance measurable in zettaflops would use 500MW of power, or the equivalent of several thousand homes. To provide such power, consistently and reliably, it’s likely that a nuclear reactor would be required.
While there have been noticeable improvements to efficiency in previous decades, progress is beginning to slow as the room for further improvements diminishes.
“This flattening of efficiency becomes the largest challenge that we have to solve, both from a technology standpoint as well as from a sustainability standpoint,” said Su.
She explained that compute efficiency should become the number one priority in order to solve the problem. While our existing means have slowed down, Su explained that there could be other solutions including employing artificial intelligence and machine learning to high-performance computing.
While there could be a nominal sacrifice to accuracy, working on improving this could be the way forward. Whichever combination of methods AMD and other companies ultimately use to deliver performance improvements, it’s clear that the journey between the first zettaflop-capable HPC and real-world use is likely to be a long one.
Via The Register (opens in new tab)
