Florian
@oook.bsky.social
High Performance Computing for climate & weather models at NASA's GSFC.
Got my name on more monkey movies than average.
Got my name on more monkey movies than average.
Je vis aux US. C'est une bonne hypothèse mais probablement plus pour les Miller et autres Vougth de son administration. Pour lui, il est malheureusement beaucoup plus probable que ce soit la peur de sa mort imminente qui le pousse à des actions "glorieuses". C'est un être de pur égo avant tout
January 11, 2026 at 2:31 PM
Je vis aux US. C'est une bonne hypothèse mais probablement plus pour les Miller et autres Vougth de son administration. Pour lui, il est malheureusement beaucoup plus probable que ce soit la peur de sa mort imminente qui le pousse à des actions "glorieuses". C'est un être de pur égo avant tout
Been driving a 2017 Bolt in the US. Best car I ever had and we live in the rural cold parts where the battery takes a nasty hit in winter.
Of course, they discontinued it...
Of course, they discontinued it...
January 3, 2026 at 12:55 AM
Been driving a 2017 Bolt in the US. Best car I ever had and we live in the rural cold parts where the battery takes a nasty hit in winter.
Of course, they discontinued it...
Of course, they discontinued it...
"This paper is so good that I have decided to join the author in his research as he has clearly outshined any other endeavor in our domain"
December 23, 2025 at 7:05 PM
"This paper is so good that I have decided to join the author in his research as he has clearly outshined any other endeavor in our domain"
Allow me to ask: is it the logical conclusion of the cultural mantra: "public good or common good doesn't exist, the individual reign supreme and any wealth or power acquired is inherently moral, because the individual success is always good regardless of the means used"
December 20, 2025 at 6:49 PM
Allow me to ask: is it the logical conclusion of the cultural mantra: "public good or common good doesn't exist, the individual reign supreme and any wealth or power acquired is inherently moral, because the individual success is always good regardless of the means used"
I was in a call today with Europeans colleague. The very first topic before we could even get into the agenda: what's going on with NCAR?
December 18, 2025 at 11:27 PM
I was in a call today with Europeans colleague. The very first topic before we could even get into the agenda: what's going on with NCAR?
Reposted by Florian
They’re not even pretending that it’s for any kind of legitimate reason. Instead, they’re trying to dismantle NCAR to suppress knowledge.
December 17, 2025 at 2:12 PM
They’re not even pretending that it’s for any kind of legitimate reason. Instead, they’re trying to dismantle NCAR to suppress knowledge.
The Age of Morons is in full swing. The damage to the scientific community will be unmeasurable. I cannot count how many pieces of software we use that originated or are maintained at NCAR. Taking this burden on would literally stop dozens of teams like mine from continuing research.
December 17, 2025 at 3:06 AM
The Age of Morons is in full swing. The damage to the scientific community will be unmeasurable. I cannot count how many pieces of software we use that originated or are maintained at NCAR. Taking this burden on would literally stop dozens of teams like mine from continuing research.
Contractors and feds alike... My team is nearing a state of the tech where we can deploy, we were hoping to go gather feedback at AGU. Got denied. Next year? Who knows...
December 16, 2025 at 3:41 PM
Contractors and feds alike... My team is nearing a state of the tech where we can deploy, we were hoping to go gather feedback at AGU. Got denied. Next year? Who knows...
That's moving up, arguably. We are running 300+ GB on a SoC in HPC grade hardware with integrated LPDDR5.
Far from consumer electronics, yes, but a step in that direction
Far from consumer electronics, yes, but a step in that direction
December 3, 2025 at 2:24 PM
That's moving up, arguably. We are running 300+ GB on a SoC in HPC grade hardware with integrated LPDDR5.
Far from consumer electronics, yes, but a step in that direction
Far from consumer electronics, yes, but a step in that direction
That absolutely doesn't solve the capacity issue the AI bubble has created in RAM. It's a knock-on good side effect at best...
December 3, 2025 at 2:22 PM
That absolutely doesn't solve the capacity issue the AI bubble has created in RAM. It's a knock-on good side effect at best...
A small silver lining is that the GH/BH all on one die silicon AI chips that exists alongside the classic GPUs are pushing development of LPDDR5 for performance. Those yield good results and could be a significant step in consumer electronics if they generalize in chip architecture.
December 3, 2025 at 2:22 PM
A small silver lining is that the GH/BH all on one die silicon AI chips that exists alongside the classic GPUs are pushing development of LPDDR5 for performance. Those yield good results and could be a significant step in consumer electronics if they generalize in chip architecture.
Unemployed engineers, but no budget to hire. Unused GPUs, but no electrical capacity or used GPUs with only a few years left in them.
We are running those 128 cores x86 for a loooooong time ;)
We are running those 128 cores x86 for a loooooong time ;)
December 3, 2025 at 2:11 PM
Unemployed engineers, but no budget to hire. Unused GPUs, but no electrical capacity or used GPUs with only a few years left in them.
We are running those 128 cores x86 for a loooooong time ;)
We are running those 128 cores x86 for a loooooong time ;)
Giving a seminar soon on High Performance Computing for research, gotta update my slide titled "Hardware market instability: want some chips, roll a dice, maybe you get it this year, maybe you don't!"
November 28, 2025 at 7:55 PM
Giving a seminar soon on High Performance Computing for research, gotta update my slide titled "Hardware market instability: want some chips, roll a dice, maybe you get it this year, maybe you don't!"
It goes beyond that. When they are installed, they have to be tested, tuned, parts have to be replaced because of a bad switch that kills performance, code tuned, etc. It's a damn long road after the ribbon cutting to a purring HPC!
November 18, 2025 at 10:28 PM
It goes beyond that. When they are installed, they have to be tested, tuned, parts have to be replaced because of a bad switch that kills performance, code tuned, etc. It's a damn long road after the ribbon cutting to a purring HPC!
If I may, let's actualize the language: IS stripping it for parts. Present tense. NASA is being disassembled now, very much underway.
October 25, 2025 at 11:56 AM
If I may, let's actualize the language: IS stripping it for parts. Present tense. NASA is being disassembled now, very much underway.
Our travel budget was 1$ _before_ the shutdown...
October 24, 2025 at 9:24 PM
Our travel budget was 1$ _before_ the shutdown...
Well deserved, congrats!
October 18, 2025 at 2:04 PM
Well deserved, congrats!
Thank you! I am slowly going insane seeing a vast majority of talking heads going on about something that has no proper definition!
Is AI the moment where science/tech truly becomes co-opted to become a religious cult?
Is AI the moment where science/tech truly becomes co-opted to become a religious cult?
October 15, 2025 at 6:13 PM
Thank you! I am slowly going insane seeing a vast majority of talking heads going on about something that has no proper definition!
Is AI the moment where science/tech truly becomes co-opted to become a religious cult?
Is AI the moment where science/tech truly becomes co-opted to become a religious cult?
It's a good way to get your badge deactivated for the entire building
October 9, 2025 at 9:47 PM
It's a good way to get your badge deactivated for the entire building
The message should read "MPI Fatal error" followed by a nice 100 deep stack of unresolved symbols.
October 9, 2025 at 9:46 PM
The message should read "MPI Fatal error" followed by a nice 100 deep stack of unresolved symbols.
Been running with those principles:
- write code at 50% complexity, since debugging is twice as hard
- code for today's problem as understood today
- write an ADR (or any design docs) and include potential paths to generalization
- write code at 50% complexity, since debugging is twice as hard
- code for today's problem as understood today
- write an ADR (or any design docs) and include potential paths to generalization
September 20, 2025 at 1:39 PM
Been running with those principles:
- write code at 50% complexity, since debugging is twice as hard
- code for today's problem as understood today
- write an ADR (or any design docs) and include potential paths to generalization
- write code at 50% complexity, since debugging is twice as hard
- code for today's problem as understood today
- write an ADR (or any design docs) and include potential paths to generalization