trend: non-NVIDIA training
DeepSeek V3.1 was trained on Huawei Ascend NPUs
this one is a South Korean lab training on AMD
DeepSeek V3.1 was trained on Huawei Ascend NPUs
this one is a South Korean lab training on AMD
Motif 2.6B — compact model with long context
unique: trained on AMD GPUs
focus is on long context & low hallucination rate — imo this is a growing genre of LLM that enables new search patterns
huggingface.co/Motif-Techno...
unique: trained on AMD GPUs
focus is on long context & low hallucination rate — imo this is a growing genre of LLM that enables new search patterns
huggingface.co/Motif-Techno...
Motif-Technologies/Motif-2.6B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
August 24, 2025 at 2:26 PM
trend: non-NVIDIA training
DeepSeek V3.1 was trained on Huawei Ascend NPUs
this one is a South Korean lab training on AMD
DeepSeek V3.1 was trained on Huawei Ascend NPUs
this one is a South Korean lab training on AMD
Calling it now next gen consoles gonna get npus installed so games can run ai models on them 🙂↔️
December 6, 2025 at 9:57 AM
Calling it now next gen consoles gonna get npus installed so games can run ai models on them 🙂↔️
Windows 10 got a soft extention until October of this year.
There are only a handful of desktop systems with NPUs and almost no one has them.
The kind of changes to Windows they're talking about are so fundamental that they're nowhere close to a reality.
There are only a handful of desktop systems with NPUs and almost no one has them.
The kind of changes to Windows they're talking about are so fundamental that they're nowhere close to a reality.
March 4, 2026 at 8:35 AM
Windows 10 got a soft extention until October of this year.
There are only a handful of desktop systems with NPUs and almost no one has them.
The kind of changes to Windows they're talking about are so fundamental that they're nowhere close to a reality.
There are only a handful of desktop systems with NPUs and almost no one has them.
The kind of changes to Windows they're talking about are so fundamental that they're nowhere close to a reality.
Also, as a side-effect, the bursting of the AI bubble!
Nvidia is at the center of the bubble, most of their revenue is from NPUs for data centers.
But with no helium those NPUs can't be fabbed.
So Nvidia is inevitably going to crash hard, within 3-6 months.
Which takes down the AI bros.
Nvidia is at the center of the bubble, most of their revenue is from NPUs for data centers.
But with no helium those NPUs can't be fabbed.
So Nvidia is inevitably going to crash hard, within 3-6 months.
Which takes down the AI bros.
March 20, 2026 at 4:17 PM
Also, as a side-effect, the bursting of the AI bubble!
Nvidia is at the center of the bubble, most of their revenue is from NPUs for data centers.
But with no helium those NPUs can't be fabbed.
So Nvidia is inevitably going to crash hard, within 3-6 months.
Which takes down the AI bros.
Nvidia is at the center of the bubble, most of their revenue is from NPUs for data centers.
But with no helium those NPUs can't be fabbed.
So Nvidia is inevitably going to crash hard, within 3-6 months.
Which takes down the AI bros.
Later these would be sold as NPUs (No Op Processing Units). Adding a nop accelerator allowed the 68040 to offload nop processing allowing it to do nothing far faster than an unaccelerated Mac
March 1, 2025 at 7:08 PM
Later these would be sold as NPUs (No Op Processing Units). Adding a nop accelerator allowed the 68040 to offload nop processing allowing it to do nothing far faster than an unaccelerated Mac
Neural Processing Units (NPUs) in Embedded Applications
link.medium.com
December 8, 2024 at 3:03 AM
Neural Processing Units (NPUs) in Embedded Applications
Linux 7.1 adds power reporting for AMD Ryzen AI NPUs
Linux 7.1 adds power reporting for AMD Ryzen AI NPUs - VideoCardz.com
AMD Ryzen AI NPUs gain live power metrics in Linux 7.1 Linux 7.1 is set to add better monitoring for AMD Ryzen AI hardware. New kernel changes for the AMD
videocardz.com
March 16, 2026 at 12:33 PM
Linux 7.1 adds power reporting for AMD Ryzen AI NPUs
People are underestimating how much efficient machine vision systems will matter when it comes to drone warfare. They're making this mistake because they're thinking about power consumption and not heat rejection. I want tiny NPUs
This downed Ukrainian UAV has a Raspberry Pi 5 single board computer with a cooling system mounted on a wing.
Receiving telemetry from the flight controller, it could work as an autopilot when communications is lost, or provide terminal targeting and autonomy.
Receiving telemetry from the flight controller, it could work as an autopilot when communications is lost, or provide terminal targeting and autonomy.
April 3, 2025 at 6:25 AM
People are underestimating how much efficient machine vision systems will matter when it comes to drone warfare. They're making this mistake because they're thinking about power consumption and not heat rejection. I want tiny NPUs
First we got TPM modules and now there are NPUs and it's like how many Skynet chips do we need?
They literally made a movie where Skynet became a ubiquitous mobile app and no one took it to heart
February 17, 2025 at 1:07 AM
First we got TPM modules and now there are NPUs and it's like how many Skynet chips do we need?
A study shows an on-device KV cache reuse system that boosts LLM inference on mobile NPUs, cutting time-to-first-token by 60% while enhancing privacy. This method links dynamic algorithms to static execution environments, paving the way for efficient local-first AI. https://arxiv.org/abs/2609.34727
Dynamic Flow, Static Graph: KV Cache Reuse for Efficient LLM Serving on Mobile NPUs
ArXiv link for Dynamic Flow, Static Graph: KV Cache Reuse for Efficient LLM Serving on Mobile NPUs
arxiv.org
September 30, 2026 at 5:20 AM
A study shows an on-device KV cache reuse system that boosts LLM inference on mobile NPUs, cutting time-to-first-token by 60% while enhancing privacy. This method links dynamic algorithms to static execution environments, paving the way for efficient local-first AI. https://arxiv.org/abs/2609.34727
Microsoft hypes PCs with NPUs, still can't offer a good reason to buy one
Microsoft hypes PCs with NPUs, still can't offer a good reason to buy one
AI tech not on the hardware compatibility list for now. But future Windows will need it
Comment Microsoft has talked up the role played by neural processing units (NPUs) in making Windows more "intelligent," even though the silicon is not currently on the hardware requirements list.…
dlvr.it
October 10, 2025 at 7:33 AM
Microsoft hypes PCs with NPUs, still can't offer a good reason to buy one
Good news for those who want to run an open-weight LLM locally on their PC.
A Discrete NPUs from AMD, Intel, and Qualcomm are coming soon… Some are already here (like Qualcomm AI 100) and others are coming soon.
A Discrete NPUs from AMD, Intel, and Qualcomm are coming soon… Some are already here (like Qualcomm AI 100) and others are coming soon.
August 3, 2025 at 2:14 AM
Good news for those who want to run an open-weight LLM locally on their PC.
A Discrete NPUs from AMD, Intel, and Qualcomm are coming soon… Some are already here (like Qualcomm AI 100) and others are coming soon.
A Discrete NPUs from AMD, Intel, and Qualcomm are coming soon… Some are already here (like Qualcomm AI 100) and others are coming soon.
Oui, je veux être géré par l'assurance maladie et plus par la MGEN. Pitié, sauvez-npus de la MGEN !
📰🗞️
Le gouvernement envisage de supprimer la délégation de gestion du régime d'assurance maladie qui était donnée aux mutuelles des agents publics (MGEN pour les profs) pour une gestion directe par la Sécu.
Ce qui conduirait à des suppressions de postes.
www.ouest-france.fr/sante/securi...
Le gouvernement envisage de supprimer la délégation de gestion du régime d'assurance maladie qui était donnée aux mutuelles des agents publics (MGEN pour les profs) pour une gestion directe par la Sécu.
Ce qui conduirait à des suppressions de postes.
www.ouest-france.fr/sante/securi...
Enseignants, policiers… Les mutuelles de fonctionnaires bientôt privées du régime obligatoire de la Sécu ?
Le gouvernement envisage de supprimer les délégations de gestion du régime obligatoire d’assurance maladie, dont bénéficient encore une dizaine de mutuelles. Cela concernerait notamment les mutuelles ...
www.ouest-france.fr
September 12, 2026 at 8:58 AM
Oui, je veux être géré par l'assurance maladie et plus par la MGEN. Pitié, sauvez-npus de la MGEN !
>based on their hexagon NPUs
cooked on arrival
cooked on arrival
October 27, 2025 at 2:22 PM
>based on their hexagon NPUs
cooked on arrival
cooked on arrival
AMD preps SR-IOV support for next-gen Ryzen AI XDNA NPUs
AMD preps SR-IOV support for next-gen Ryzen AI XDNA NPUs - VideoCardz.com
AMD Medusa-era Ryzen AI NPUs may support SR-IOV AMD is preparing SR-IOV support for its next-generation Ryzen AI NPUs. According to Phoronix, a new Linux
videocardz.com
May 7, 2026 at 3:36 PM
AMD preps SR-IOV support for next-gen Ryzen AI XDNA NPUs
I started to talk with teams building the leading OSes about how AI changes (or does not change) their roadmap and what their OS looks like.
Ubuntu feels very grounded: massive focus on hardware support (GPUs, NPUs, DPUs)+local-first LLMs:
newsletter.pragmaticengineer.com/p/ubuntu-and...
Ubuntu feels very grounded: massive focus on hardware support (GPUs, NPUs, DPUs)+local-first LLMs:
newsletter.pragmaticengineer.com/p/ubuntu-and...
April 28, 2026 at 2:32 PM
I started to talk with teams building the leading OSes about how AI changes (or does not change) their roadmap and what their OS looks like.
Ubuntu feels very grounded: massive focus on hardware support (GPUs, NPUs, DPUs)+local-first LLMs:
newsletter.pragmaticengineer.com/p/ubuntu-and...
Ubuntu feels very grounded: massive focus on hardware support (GPUs, NPUs, DPUs)+local-first LLMs:
newsletter.pragmaticengineer.com/p/ubuntu-and...
Prozessoren werden noch komplexer: Außer separaten NPUs bekommen auch ihre CPU- und GPU-Kerne Matrix-Einheiten. Ein Linux-Projekt mit Snapdragon X scheitert. #Prozessoren
heise+ | Bit-Rauschen: Die CPU-Matrix-Rechenwerke kommen
Prozessoren werden noch komplexer: Außer separaten NPUs bekommen auch ihre CPU- und GPU-Kerne Matrix-Einheiten. Ein Linux-Projekt mit Snapdragon X scheitert.
www.heise.de
December 8, 2025 at 6:03 AM
Prozessoren werden noch komplexer: Außer separaten NPUs bekommen auch ihre CPU- und GPU-Kerne Matrix-Einheiten. Ein Linux-Projekt mit Snapdragon X scheitert. #Prozessoren
Liquid AI's LFM2.5-230M (open-weight and run fast anywhere (CPUs, NPUs, and GPUs)
> 230M parameters, built on the LFM2 architecture
> Pre-trained on 19T tokens, with a 32K context extension
> Post-trained with distillation from LFM2.5-350M
> 230M parameters, built on the LFM2 architecture
> Pre-trained on 19T tokens, with a 32K context extension
> Post-trained with distillation from LFM2.5-350M
June 25, 2026 at 6:11 PM
Liquid AI's LFM2.5-230M (open-weight and run fast anywhere (CPUs, NPUs, and GPUs)
> 230M parameters, built on the LFM2 architecture
> Pre-trained on 19T tokens, with a 32K context extension
> Post-trained with distillation from LFM2.5-350M
> 230M parameters, built on the LFM2 architecture
> Pre-trained on 19T tokens, with a 32K context extension
> Post-trained with distillation from LFM2.5-350M
Mesa Gallium3D Driver Merged For Arm Ethos NPUs - https://www.phoronix.com/news/Arm-Ethos-Gallium-Merged
Mesa Gallium3D Driver Merged For Arm Ethos NPUs
Arm has been working on an open-source Linux kernel accelerator "accel" driver for their Ethos NPUs. That kernel driver continues being revised and under review for inclusion into a future mainline Linux kernel release. Already though a Gallium3D driver for Mesa has been merged for leveraging the Ethos NPU...
www.phoronix.com
October 16, 2025 at 12:58 AM
Mesa Gallium3D Driver Merged For Arm Ethos NPUs - https://www.phoronix.com/news/Arm-Ethos-Gallium-Merged
i know its not fun to hear about the wet blanket on ragebait, but stop getting worked up about the fake reasons windows 12 sucks hallucinated by AI (NPUs being required to boot) and get worked up about the reasons it sucks, like the subscription and overabundance of AI slop, thank you
ai summaries of ai summaries of the best source, this pc world article, which was probably created with AI
it seems like NPUs are a requirement for the full facefukc of ai slop, but not to boot
it seems like NPUs are a requirement for the full facefukc of ai slop, but not to boot
March 4, 2026 at 5:35 PM
i know its not fun to hear about the wet blanket on ragebait, but stop getting worked up about the fake reasons windows 12 sucks hallucinated by AI (NPUs being required to boot) and get worked up about the reasons it sucks, like the subscription and overabundance of AI slop, thank you
So remember way back last week when Elmo's AI company bought X?
AI relies on NPUs.
Where are the majority of NPUs manufactured?
And which country is the orange guy threatening additional 50% tariffs on?
#justsaying
AI relies on NPUs.
Where are the majority of NPUs manufactured?
And which country is the orange guy threatening additional 50% tariffs on?
#justsaying
April 8, 2025 at 12:58 AM
So remember way back last week when Elmo's AI company bought X?
AI relies on NPUs.
Where are the majority of NPUs manufactured?
And which country is the orange guy threatening additional 50% tariffs on?
#justsaying
AI relies on NPUs.
Where are the majority of NPUs manufactured?
And which country is the orange guy threatening additional 50% tariffs on?
#justsaying
I built a "Trash Build" for just €300 total (i7-4930K + 2x GTX 1060) salvaging parts from eBay. Technologically, it's 10-year-old "e-waste".
The result? 🤯 On open-source LLMs (Ollama), it actually beats today's NPUs and €2000 AI laptops.
#LocalLLM #Ollama #HardwareHacking #AI #TechSalvage
The result? 🤯 On open-source LLMs (Ollama), it actually beats today's NPUs and €2000 AI laptops.
#LocalLLM #Ollama #HardwareHacking #AI #TechSalvage
December 13, 2025 at 7:36 PM
I built a "Trash Build" for just €300 total (i7-4930K + 2x GTX 1060) salvaging parts from eBay. Technologically, it's 10-year-old "e-waste".
The result? 🤯 On open-source LLMs (Ollama), it actually beats today's NPUs and €2000 AI laptops.
#LocalLLM #Ollama #HardwareHacking #AI #TechSalvage
The result? 🤯 On open-source LLMs (Ollama), it actually beats today's NPUs and €2000 AI laptops.
#LocalLLM #Ollama #HardwareHacking #AI #TechSalvage
Fuck NPUs
Bring back the Cell processor
Bring back the Cell processor
looking at new hardware, and it is /such/ a nightmare that everyone's added an onboard NPU without a single actual use case.
like literally MS has not even come up with a single NPU application other than studio? obscene. they're just waving their hands about how there's going to be in the future.
like literally MS has not even come up with a single NPU application other than studio? obscene. they're just waving their hands about how there's going to be in the future.
July 5, 2025 at 9:29 AM
Fuck NPUs
Bring back the Cell processor
Bring back the Cell processor
My son (11yo) likes to be given LEGO tasks. Yesterday, I told him to 'make me a philosopher in the style of The School of Athens for my office at the university.' I love how he even captured some Diogenes-like too-cool-for-school. And of course a couple of really excellent NPUs going on.
#philosophy
#philosophy
December 15, 2025 at 9:06 AM
My son (11yo) likes to be given LEGO tasks. Yesterday, I told him to 'make me a philosopher in the style of The School of Athens for my office at the university.' I love how he even captured some Diogenes-like too-cool-for-school. And of course a couple of really excellent NPUs going on.
#philosophy
#philosophy
As for citations, I'm not an ML researcher, they probably have better benchmarks of differences between training stuff on GPUs and NPUs. I am however a database guy, so I can tell you that there is research into how NPUs are much better for executing PageRank or fuzzy matches in databases.
dl.acm.org
May 7, 2025 at 12:29 PM
As for citations, I'm not an ML researcher, they probably have better benchmarks of differences between training stuff on GPUs and NPUs. I am however a database guy, so I can tell you that there is research into how NPUs are much better for executing PageRank or fuzzy matches in databases.