Uploaded November 2024 | Updated September 2026, 1 week ago
So your team of semiconductor researchers have designed and built a brand new type of chip that has the potential to be far more efficient for AI computations. But how do they test that it actually works like they intended? That’s where IBM Research Hardware Engineer JohnDavid Lancaster comes in.
In this behind-the-scenes look at his lab in Yorktown Heights, he demonstrates his team tests and validates the “power budget” — a “wattage-to-inference” ratio — of IBM’s latest AIU (Artificial Intelligence Unit) chips, which were specifically designed and optimized for AI applications.
“One of the key reasons you would use our AIU card is the super low power for how many inferences you’re able to compute.”
Lancaster shows how mapping the voltage of the entire card can ultimately help inform sustainable (low-power) encoder and decoder model design.
Read more about our AIU here: research.ibm.com/blog/aiu-chip-family-ibm-research
Subscribe and follow for more lab tours!
http://ibm.biz/subscribe_IBM_Research
0:00 - Intro
0:15 - What is the AIU?
0:47 - Parts & components
1:09 - Performance per watt advantages
1:35 - AIU workloads
2:23 - Power grab demo
3:16 - Outro
#ai #chips #artificialintelligence
So your team of semiconductor researchers have designed and built a brand new type of chip that has the potential to be far more efficient for AI computations. But how do they test that it actually works like they intended? That’s where IBM Research Hardware Engineer JohnDavid Lancaster comes in.
In this behind-the-scenes look at his lab in Yorktown Heights, he demonstrates his team tests and validates the “power budget” — a “wattage-to-inference” ratio — of IBM’s latest AIU (Artificial Intelligence Unit) chips, which were specifically designed and optimized for AI applications.
“One of the key reasons you would use our AIU card is the super low power for how many inferences you’re able to compute.”
Lancaster shows how mapping the voltage of the entire card can ultimately help inform sustainable (low-power) encoder and decoder model design.
Read more about our AIU here: research.ibm.com/blog/aiu-chip-family-ibm-research
Subscribe and follow for more lab tours!
http://ibm.biz/subscribe_IBM_Research
0:00 - Intro
0:15 - What is the AIU?
0:47 - Parts & components
1:09 - Performance per watt advantages
1:35 - AIU workloads
2:23 - Power grab demo
3:16 - Outro
#ai #chips #artificialintelligence
