Torsten Hoefler 🇨🇭
@thoefler.bsky.social
Professor ETHZ, head of SPCL, Chief Architect ML at CSCS researching large-scale #HPC and #AI systems and #Climate computing - youtube: http://bit.ly/3h1VgIU
Very nice overview of the emerging UALink standard with nice features such as splitting packets in switches, in-network computing, high energy efficiency, and lowest silicon overhead: buff.ly/AgLvC1g
I'll be joining a panel at SC25 contrasting UALink and UEC next Wed: buff.ly/BeCMFcL Join us there
I'll be joining a panel at SC25 contrasting UALink and UEC next Wed: buff.ly/BeCMFcL Join us there
Introducing the UALink 200G 1.0 Specification Webinar
The Ultra Accelerator Link™ (UALink™) Consortium is an open industry standard group dedicated to advancing the UALink specification. The Consortium recently released the UALink 200G 1.0…
buff.ly
November 14, 2025 at 6:00 AM
Very nice overview of the emerging UALink standard with nice features such as splitting packets in switches, in-network computing, high energy efficiency, and lowest silicon overhead: buff.ly/AgLvC1g
I'll be joining a panel at SC25 contrasting UALink and UEC next Wed: buff.ly/BeCMFcL Join us there
I'll be joining a panel at SC25 contrasting UALink and UEC next Wed: buff.ly/BeCMFcL Join us there
Ram Velega (Broadcom) at OCP "The things that have confined us today that the scale-up domain is less than 100 are going to be changed very soon" (7:55 in buff.ly/9ioPH73)
I see this as a call to action for systems and software research.
UE 1.0 techniques such as LLR and CBFC play a crucial role!
I see this as a call to action for systems and software research.
UE 1.0 techniques such as LLR and CBFC play a crucial role!
2025 OCP APAC Summit Keynote - Broadcom
Ram Velaga (Broadcom)
Scale-up for AI: Balancing Compute, Memory and Networking
AI models are growing at an unprecedented pace, driving exponential increases in compute demands. As a single GPU…
buff.ly
November 12, 2025 at 6:00 AM
Ram Velega (Broadcom) at OCP "The things that have confined us today that the scale-up domain is less than 100 are going to be changed very soon" (7:55 in buff.ly/9ioPH73)
I see this as a call to action for systems and software research.
UE 1.0 techniques such as LLR and CBFC play a crucial role!
I see this as a call to action for systems and software research.
UE 1.0 techniques such as LLR and CBFC play a crucial role!
Ilya Sutskever "Three lines of math can prove all of supervised learning. That's nice" (4:33)
"I have not seen an exposition of unsupervised learning that I found satisfying" (7:50)
Its optimization objective has little relation to the actual objective you care about
Watch: buff.ly/Bnvazym
"I have not seen an exposition of unsupervised learning that I found satisfying" (7:50)
Its optimization objective has little relation to the actual objective you care about
Watch: buff.ly/Bnvazym
November 10, 2025 at 6:00 AM
Ilya Sutskever "Three lines of math can prove all of supervised learning. That's nice" (4:33)
"I have not seen an exposition of unsupervised learning that I found satisfying" (7:50)
Its optimization objective has little relation to the actual objective you care about
Watch: buff.ly/Bnvazym
"I have not seen an exposition of unsupervised learning that I found satisfying" (7:50)
Its optimization objective has little relation to the actual objective you care about
Watch: buff.ly/Bnvazym
Can we build an #AI #Climate Scientist? Asked at the ADIA Lab Symposium in Abu Dhabi last week - now online at buff.ly/6igSeyg :-).
Much work to be done - this is outlining some directions of indicative results with a lot of potential to accelerate AI for Science.
Much work to be done - this is outlining some directions of indicative results with a lot of potential to accelerate AI for Science.
November 9, 2025 at 9:24 AM
Can we build an #AI #Climate Scientist? Asked at the ADIA Lab Symposium in Abu Dhabi last week - now online at buff.ly/6igSeyg :-).
Much work to be done - this is outlining some directions of indicative results with a lot of potential to accelerate AI for Science.
Much work to be done - this is outlining some directions of indicative results with a lot of potential to accelerate AI for Science.
Collaborator and friend Dan Alistarh talks at ETH about using the new NvFP4 and MXFP4 block formats for inference.
Some going from "terrible" accuracy to acceptable using micro rotations to smoothen outliers in blocks.
arxiv.org/abs/2509.23202
Great collaboration and cool stuff
Some going from "terrible" accuracy to acceptable using micro rotations to smoothen outliers in blocks.
arxiv.org/abs/2509.23202
Great collaboration and cool stuff
November 5, 2025 at 8:32 AM
Collaborator and friend Dan Alistarh talks at ETH about using the new NvFP4 and MXFP4 block formats for inference.
Some going from "terrible" accuracy to acceptable using micro rotations to smoothen outliers in blocks.
arxiv.org/abs/2509.23202
Great collaboration and cool stuff
Some going from "terrible" accuracy to acceptable using micro rotations to smoothen outliers in blocks.
arxiv.org/abs/2509.23202
Great collaboration and cool stuff
Keren Bergman at the 2nd EFCL Workshop: "Huawei combines 3x less performant GPUs with a photonic scale-up network to build higher performance PODs than Nvidia."
Nvidia moving towards CPO 😀. Optics everywhere in scale-out.
Nice overview of optical networking at all distances.
Nvidia moving towards CPO 😀. Optics everywhere in scale-out.
Nice overview of optical networking at all distances.
November 4, 2025 at 12:49 PM
Keren Bergman at the 2nd EFCL Workshop: "Huawei combines 3x less performant GPUs with a photonic scale-up network to build higher performance PODs than Nvidia."
Nvidia moving towards CPO 😀. Optics everywhere in scale-out.
Nice overview of optical networking at all distances.
Nvidia moving towards CPO 😀. Optics everywhere in scale-out.
Nice overview of optical networking at all distances.
I was very honored to meet Carnegie Mellon University's President, Dean of the School of CS, and its famous founder Raj Reddy to present a lecture named after him.
I tremendously enjoyed speaking with young students and faculty and the evening with CMU's leadership. Thanks for the invitation Raj!
I tremendously enjoyed speaking with young students and faculty and the evening with CMU's leadership. Thanks for the invitation Raj!
November 3, 2025 at 6:00 AM
I was very honored to meet Carnegie Mellon University's President, Dean of the School of CS, and its famous founder Raj Reddy to present a lecture named after him.
I tremendously enjoyed speaking with young students and faculty and the evening with CMU's leadership. Thanks for the invitation Raj!
I tremendously enjoyed speaking with young students and faculty and the evening with CMU's leadership. Thanks for the invitation Raj!
MIT's Sandy Pentland at ADIA Lab symposium: "Modern companies need to structure their incentives to coordiate teams instead of using strict and siloed hierarchies." Enable team leaders to do "what they think is right" instead of inefficient political discussions with leadership.
October 30, 2025 at 6:01 AM
MIT's Sandy Pentland at ADIA Lab symposium: "Modern companies need to structure their incentives to coordiate teams instead of using strict and siloed hierarchies." Enable team leaders to do "what they think is right" instead of inefficient political discussions with leadership.
Reposted by Torsten Hoefler 🇨🇭
🎉 Uno accepted to SC25! Unified congestion control + reliable connectivity for intra- & inter-DC traffic to enable inter-DC AI training.
📄Paper: arxiv.org/abs/2510.15802
💻Code: github.com/spcl/Uno_SC25
🤝Collaboration with Microsoft
#SC25 #AI #SPCL @thoefler.bsky.social @csateth.bsky.social
📄Paper: arxiv.org/abs/2510.15802
💻Code: github.com/spcl/Uno_SC25
🤝Collaboration with Microsoft
#SC25 #AI #SPCL @thoefler.bsky.social @csateth.bsky.social
October 29, 2025 at 9:25 AM
🎉 Uno accepted to SC25! Unified congestion control + reliable connectivity for intra- & inter-DC traffic to enable inter-DC AI training.
📄Paper: arxiv.org/abs/2510.15802
💻Code: github.com/spcl/Uno_SC25
🤝Collaboration with Microsoft
#SC25 #AI #SPCL @thoefler.bsky.social @csateth.bsky.social
📄Paper: arxiv.org/abs/2510.15802
💻Code: github.com/spcl/Uno_SC25
🤝Collaboration with Microsoft
#SC25 #AI #SPCL @thoefler.bsky.social @csateth.bsky.social
One highlight of the ADIA Lab Symposium was Nobel Laureate Chu's talk towards Net-Zero emissions.
"China follows the US textbook from 100 years ago, when the US took products inventend in Europe, such as cars, industrialized and improved them. Now China takes things invented in the west..."
"China follows the US textbook from 100 years ago, when the US took products inventend in Europe, such as cars, industrialized and improved them. Now China takes things invented in the west..."
October 28, 2025 at 5:53 AM
One highlight of the ADIA Lab Symposium was Nobel Laureate Chu's talk towards Net-Zero emissions.
"China follows the US textbook from 100 years ago, when the US took products inventend in Europe, such as cars, industrialized and improved them. Now China takes things invented in the west..."
"China follows the US textbook from 100 years ago, when the US took products inventend in Europe, such as cars, industrialized and improved them. Now China takes things invented in the west..."
Just arrived at the ADIA Lab symposium in Abu Dhabi to listen to Horst Simon's introduction and Bjorn Stevens' keynote on how to compute the future climate! Featuring our Gordon Bell finalists 🌍🚀
Looking forward to speculating about how to create an #AI climate scientist 😀.
Looking forward to speculating about how to create an #AI climate scientist 😀.
October 27, 2025 at 6:25 AM
Just arrived at the ADIA Lab symposium in Abu Dhabi to listen to Horst Simon's introduction and Bjorn Stevens' keynote on how to compute the future climate! Featuring our Gordon Bell finalists 🌍🚀
Looking forward to speculating about how to create an #AI climate scientist 😀.
Looking forward to speculating about how to create an #AI climate scientist 😀.
Microsoft's Ultra Ethernet tutorial is now available on youtube 🎥!
Saurabh Dighe gives a brilliant motivation for Microsoft's "AI first datacenters" 🤖 followed by Abdul Kabbani and myself explaining the technical details and innovations of Ultra Ethernet supporting this goal 🫡.
buff.ly/IPN46ZR
Saurabh Dighe gives a brilliant motivation for Microsoft's "AI first datacenters" 🤖 followed by Abdul Kabbani and myself explaining the technical details and innovations of Ultra Ethernet supporting this goal 🫡.
buff.ly/IPN46ZR
October 20, 2025 at 5:00 AM
Microsoft's Ultra Ethernet tutorial is now available on youtube 🎥!
Saurabh Dighe gives a brilliant motivation for Microsoft's "AI first datacenters" 🤖 followed by Abdul Kabbani and myself explaining the technical details and innovations of Ultra Ethernet supporting this goal 🫡.
buff.ly/IPN46ZR
Saurabh Dighe gives a brilliant motivation for Microsoft's "AI first datacenters" 🤖 followed by Abdul Kabbani and myself explaining the technical details and innovations of Ultra Ethernet supporting this goal 🫡.
buff.ly/IPN46ZR
Reposted by Torsten Hoefler 🇨🇭
Next was an intriguing talk by @thoefler.bsky.social on computational architectures for more efficiently training large models at @scsatcmu.bsky.social www.youtube.com/watch?v=LnXp... (4/8)
The 2025 Raj Reddy Artificial Intelligence Lecture
YouTube video by CMU School of Computer Science
www.youtube.com
October 16, 2025 at 3:07 AM
Next was an intriguing talk by @thoefler.bsky.social on computational architectures for more efficiently training large models at @scsatcmu.bsky.social www.youtube.com/watch?v=LnXp... (4/8)
I'm excited to discuss whether we can build an "AI Climate Scientist" in my talk at the ADIA Lab Symposium 2025 🌎
Join us in Abu Dhabi or online from October 27–29!
Register here: buff.ly/gCP2K1z
#ADIALabSymposium2025
Join us in Abu Dhabi or online from October 27–29!
Register here: buff.ly/gCP2K1z
#ADIALabSymposium2025
October 15, 2025 at 10:47 AM
I'm excited to discuss whether we can build an "AI Climate Scientist" in my talk at the ADIA Lab Symposium 2025 🌎
Join us in Abu Dhabi or online from October 27–29!
Register here: buff.ly/gCP2K1z
#ADIALabSymposium2025
Join us in Abu Dhabi or online from October 27–29!
Register here: buff.ly/gCP2K1z
#ADIALabSymposium2025
I was shocked to see the first two people in my "masterclass" 🎓 on #AI networking with Ultra Ethernet at #HLF25: David Patterson and Bob Metcalfe 😅! Both Turing award winners - Bob being one of the inventors of Ethernet 🥹. Was great fun also with many enthusiastic students and great discussions 🚀.
October 13, 2025 at 5:00 AM
I'm honored to present the Raj Reddy Artificial Intelligence lecture at Carnegie Mellon University this Thursday. Join us in person at 5pm in Rashid Auditorium, Gates Hillman 4401!
buff.ly/1kj9LtN
This lecture series is honoring the Turing award winner's work in AI.
buff.ly/1kj9LtN
This lecture series is honoring the Turing award winner's work in AI.
Raj Reddy Artificial Intelligence Lecture - Torsten Hoefler | Carnegie Mellon University Computer Science Department
We will explore the fascinating evolution of Large Language Models (LLMs) and their transformative journey through the lenses of computation and optimization. We begin by tracing the origins of LLMs,…
buff.ly
October 8, 2025 at 5:00 AM
I'm honored to present the Raj Reddy Artificial Intelligence lecture at Carnegie Mellon University this Thursday. Join us in person at 5pm in Rashid Auditorium, Gates Hillman 4401!
buff.ly/1kj9LtN
This lecture series is honoring the Turing award winner's work in AI.
buff.ly/1kj9LtN
This lecture series is honoring the Turing award winner's work in AI.
Reposted by Torsten Hoefler 🇨🇭
Thanks to The New York Times for featuring the ACM A.M. Turing Prize! Indeed, beyond the Nobel Prizes, "there are plenty of other prizes scientists and mathematicians can compete for."
Learn more: buff.ly/NAhXVmy
Learn more: buff.ly/NAhXVmy
October 7, 2025 at 5:23 PM
Thanks to The New York Times for featuring the ACM A.M. Turing Prize! Indeed, beyond the Nobel Prizes, "there are plenty of other prizes scientists and mathematicians can compete for."
Learn more: buff.ly/NAhXVmy
Learn more: buff.ly/NAhXVmy
The first ADIA Lab transactions edited by Horst Simon arrive at my desk 📖.
An exciting mix of different science areas under the umbrella of advanced scientific computing and #AI 🎯. Congratulations to Horst and all co-authors 👏.
Onward 🚀!
An exciting mix of different science areas under the umbrella of advanced scientific computing and #AI 🎯. Congratulations to Horst and all co-authors 👏.
Onward 🚀!
October 6, 2025 at 5:00 AM
The first ADIA Lab transactions edited by Horst Simon arrive at my desk 📖.
An exciting mix of different science areas under the umbrella of advanced scientific computing and #AI 🎯. Congratulations to Horst and all co-authors 👏.
Onward 🚀!
An exciting mix of different science areas under the umbrella of advanced scientific computing and #AI 🎯. Congratulations to Horst and all co-authors 👏.
Onward 🚀!
Reposted by Torsten Hoefler 🇨🇭
Listen to 2024 ACM Prize in Computing recipient Torsten Hoefler share how he tries to bridge academia and industry for societal impact . Full episode here or wherever you get your podcast: learning.acm.org/bytecast/ep7...
#ACMByteCast #computing
#ACMByteCast #computing
October 4, 2025 at 2:05 PM
Listen to 2024 ACM Prize in Computing recipient Torsten Hoefler share how he tries to bridge academia and industry for societal impact . Full episode here or wherever you get your podcast: learning.acm.org/bytecast/ep7...
#ACMByteCast #computing
#ACMByteCast #computing
Cool and impactful project led by the MPI-M: coupled simulation of the full Earth system in 1.25km resolution. Unprecedented complexity, fidelity, and performance at 82.5 simulated days per day. Accelerated by DaCe on GH200 - next generation #HPC for #climate science is here!
Global climate simulations achieve 1.25 km resolution—team nominated for Climate Gordon Bell Prize | CSCS
For the first time, scientists have run global coupled climate simulations at a resolution of just 1.25 kilometres. Using CSCS’s “Alps” supercomputer, the team including researchers from ETH Zürich…
buff.ly
October 2, 2025 at 5:00 AM
I met many of my heroes at my first Heidelberg Laureate Forum #HLF25! Exciting discussions with 28 other laureates of the five highest awards in CS and Math and young researchers!
Watch my Spark talk at: buff.ly/SO6ntUb
Thanks to the HLF Foundation and Klaus Tschira Stiftung.
Watch my Spark talk at: buff.ly/SO6ntUb
Thanks to the HLF Foundation and Klaus Tschira Stiftung.
September 30, 2025 at 5:00 AM
I met many of my heroes at my first Heidelberg Laureate Forum #HLF25! Exciting discussions with 28 other laureates of the five highest awards in CS and Math and young researchers!
Watch my Spark talk at: buff.ly/SO6ntUb
Thanks to the HLF Foundation and Klaus Tschira Stiftung.
Watch my Spark talk at: buff.ly/SO6ntUb
Thanks to the HLF Foundation and Klaus Tschira Stiftung.
Happy 100th Birthday to Seymour Cray! 🎂
We took your relentless pursuit of speed and parallel processing, and built on your ideas to start the #AI revolution. Every neural network owes a debt to your chips.
Thanks for the gigahertz, Seymour! We're putting them to good use - your #LLM friend. 😉 #AI
We took your relentless pursuit of speed and parallel processing, and built on your ideas to start the #AI revolution. Every neural network owes a debt to your chips.
Thanks for the gigahertz, Seymour! We're putting them to good use - your #LLM friend. 😉 #AI
September 28, 2025 at 3:21 PM
Reposted by Torsten Hoefler 🇨🇭
Listen to 2024 ACM Prize in Computing recipient Torsten Hoefler discuss the power of High Performance Computing. Full episode here or wherever you get your podcast: learning.acm.org/bytecast/ep7...
#ACMByteCast #computing
#ACMByteCast #computing
September 26, 2025 at 1:26 PM
Listen to 2024 ACM Prize in Computing recipient Torsten Hoefler discuss the power of High Performance Computing. Full episode here or wherever you get your podcast: learning.acm.org/bytecast/ep7...
#ACMByteCast #computing
#ACMByteCast #computing
Apertus, the Swiss Fully-Open-Data model downloaded more than 379k times - ~13 downloads per minute of up to 140+ GiB! Trending for some weeks among the world's top LLMs.
The techreport is now on arXiv: buff.ly/bvSioQH
The techreport is now on arXiv: buff.ly/bvSioQH
September 24, 2025 at 5:00 AM
Apertus, the Swiss Fully-Open-Data model downloaded more than 379k times - ~13 downloads per minute of up to 140+ GiB! Trending for some weeks among the world's top LLMs.
The techreport is now on arXiv: buff.ly/bvSioQH
The techreport is now on arXiv: buff.ly/bvSioQH