Dan Ernst
@ernstdj.bsky.social
HPC Architecture Ronin, currently at Nvidia. ex-MSFT, but forever a Crayon.
I am proud of my home state (MN) and you will hear about it.
All posts (especially the dumb ones) reflect my own positions and not that of my employer.
I am proud of my home state (MN) and you will hear about it.
All posts (especially the dumb ones) reflect my own positions and not that of my employer.
www.theregister.com/2025/11/18/f...
You're right, it's not dead! 😁
Ozaki does not replace FP64. It just makes DGEMM faster. Why is that bad?
You're right, it's not dead! 😁
Ozaki does not replace FP64. It just makes DGEMM faster. Why is that bad?
The future of supercomputing will be AI accelerated
Interview: Nvidia's Ian Buck on the importance of FP64 to power research, in a world that's hot for inferencing
www.theregister.com
November 21, 2025 at 6:01 AM
www.theregister.com/2025/11/18/f...
You're right, it's not dead! 😁
Ozaki does not replace FP64. It just makes DGEMM faster. Why is that bad?
You're right, it's not dead! 😁
Ozaki does not replace FP64. It just makes DGEMM faster. Why is that bad?
That was the best analogy, though!
November 21, 2025 at 5:47 AM
That was the best analogy, though!
Reminds me of Seattle in 2011. Didn't want to be a vendor in the "kid's table" hall.
November 17, 2025 at 1:41 AM
Reminds me of Seattle in 2011. Didn't want to be a vendor in the "kid's table" hall.
Same. Not even close.
November 5, 2025 at 10:16 PM
Same. Not even close.
My 13yo showed me this passage in a book he was reading the other day. My response: "yeah, I've got one of those (blades) in my office". 😄
October 4, 2025 at 6:03 PM
My 13yo showed me this passage in a book he was reading the other day. My response: "yeah, I've got one of those (blades) in my office". 😄
Yeah, this has been a go-to for years. Becoming especially important lately trying to find facts in the AI hype cycles.
September 20, 2025 at 5:53 PM
Yeah, this has been a go-to for years. Becoming especially important lately trying to find facts in the AI hype cycles.
See also: fs.blog/two-types-of...
Two Types of Knowledge: The Max Planck/Chauffeur Test
Charlie Munger identifies the two types of knowledge and shows us how to distinguish between those with Chauffeur knowledge and those with Planck knowledge.
fs.blog
September 20, 2025 at 4:53 PM
See also: fs.blog/two-types-of...
wut
The only office in the Twin Cities was for educational software and Office trainings, last I looked.
That would have been super valuable
The only office in the Twin Cities was for educational software and Office trainings, last I looked.
That would have been super valuable
September 10, 2025 at 10:24 AM
wut
The only office in the Twin Cities was for educational software and Office trainings, last I looked.
That would have been super valuable
The only office in the Twin Cities was for educational software and Office trainings, last I looked.
That would have been super valuable
The extra fun trivia is that I posted that message while sitting at the project kick-off meeting in Tokyo. :-)
September 9, 2025 at 8:20 AM
The extra fun trivia is that I posted that message while sitting at the project kick-off meeting in Tokyo. :-)
That's a DFMA target number. Not GEMM.
September 8, 2025 at 10:01 PM
That's a DFMA target number. Not GEMM.
To be chosen to participate in the design of the next iteration is a massive honor for us.
Through this partnership, we aim to develop a scientific discovery ecosystem based on the full integration of simulation and AI. We hope this work will benefit not only Japan, but also researchers worldwide.
Through this partnership, we aim to develop a scientific discovery ecosystem based on the full integration of simulation and AI. We hope this work will benefit not only Japan, but also researchers worldwide.
August 22, 2025 at 3:20 AM
To be chosen to participate in the design of the next iteration is a massive honor for us.
Through this partnership, we aim to develop a scientific discovery ecosystem based on the full integration of simulation and AI. We hope this work will benefit not only Japan, but also researchers worldwide.
Through this partnership, we aim to develop a scientific discovery ecosystem based on the full integration of simulation and AI. We hope this work will benefit not only Japan, but also researchers worldwide.
Those numbers were from their feasibility study and not necessarily where the design will land.
August 22, 2025 at 2:15 AM
Those numbers were from their feasibility study and not necessarily where the design will land.
Not enough parallelism to saturate either.
August 8, 2025 at 4:17 AM
Not enough parallelism to saturate either.
For some reason, my three day weekends don't produce things like this...
August 2, 2025 at 12:49 PM
For some reason, my three day weekends don't produce things like this...
Congrats, Glenn! Hope you enjoy the new gig! 🥳
July 19, 2025 at 1:07 AM
Congrats, Glenn! Hope you enjoy the new gig! 🥳