Gabriel's Avatar

Gabriel

@dssgabriel.bsky.social

PhD candidate, HPC Software Engineering @cea.fr MSc HPC & Simulation from @univparissaclay.bsky.social Architecture, microbenchmarking & SIMD sorcery. Research on data structures & memory layouts at exascale. RTFM πŸ‘Ή

13 Followers  |  85 Following  |  16 Posts  |  Joined: 01.04.2025  |  2.2451

Latest posts by dssgabriel.bsky.social on Bluesky

Making sure you're not a bot!

#jj-vcs For Busy Devs, Part 2: "How Do I...?"

maddie.wtf/posts/2025-0...

02.08.2025 12:21 β€” πŸ‘ 45    πŸ” 8    πŸ’¬ 2    πŸ“Œ 0
Post image

I just finished my first meaningful #APX code path: #SHA3 / #Keccak1600 on 32 GPR registers: Only 130 uops + a few MOVs. I wonder if it can beat the latency of 35+ uop #AVX512 version on #Intel #DiamondRapids.
#Novalake , #PantherCove , #CoyoteCove

31.07.2025 15:13 β€” πŸ‘ 2    πŸ” 2    πŸ’¬ 1    πŸ“Œ 0

I see! Indeed, I can think of a few times where it would have been really useful to be able to do this.

Thank you very much for your explanation!

26.07.2025 21:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

"throw" expression has type "void", but has one special case if it is one of the cases in a ternary "?:" expression, where its void is "converted" to the other operand. This is the only situation where this happens, but it could be extended arbitrarily for return, break, continue, and goto.

26.07.2025 18:31 β€” πŸ‘ 1    πŸ” 1    πŸ’¬ 1    πŸ“Œ 0
Post image Post image Post image

Agner Fog updated his manuals with the #AMD #Zen5
agner.org/optimize/

26.07.2025 16:58 β€” πŸ‘ 1    πŸ” 2    πŸ’¬ 1    πŸ“Œ 0

I am not familiar with "throw’s weird void but special type". What is that and what would it allow for `return`? Do you have any resources I could read about it?

26.07.2025 16:53 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Final Benchmarks Of Clear Linux On Intel: ~48% Faster Than Ubuntu Out-Of-The-Box Last week Friday the unfortunate news came down that Intel was discontinuing their Clear Linux project effective immediately. For the past ten years Intel software engineers have been crafting Clear Linux as a high performance distribution that is extensively optimized for x86_64 processors via aggressive compiler tuning, various patches to the Linux kernel and other packages, and a variety of other optimizations throughout the operating system. For years Clear Linux has led Linux x86_64 performance not only on Intel desktop/mobile/server hardware but on AMD systems too. Here is a final look at the Clear Linux performance on the Intel side compared to the performance of the latest Ubuntu 25.04 release.

Final Benchmarks Of Clear Linux On Intel: ~48% Faster Than Ubuntu Out-Of-The-Box - https://www.phoronix.com/review/clear-linux-48p-ubuntu

25.07.2025 18:12 β€” πŸ‘ 9    πŸ” 2    πŸ’¬ 0    πŸ“Œ 0

πŸ’₯Spack v1.0 is out!πŸ’₯

This is a huge milestone. We reworked the core to add compiler dependencies, and we're introducing a stable package API.

πŸš€1.0 also adds concurrent builds, better includes, and much more -- read it all in the release notes!

github.com/spack/spack/...

20.07.2025 10:45 β€” πŸ‘ 41    πŸ” 16    πŸ’¬ 0    πŸ“Œ 5
Intel Announces It's Shutting Down Clear Linux The most depressing news of the week: Intel is ending their performance-optimized Clear Linux distribution. Over the past decade the Clear Linux operating system has shown what's possible with out-of-the-box performance on x86_64 hardware... Not just for Intel platforms but even showing extremely great performance results on AMD x86_64 too. But with the cost-cutting going on at Intel, Clear Linux is now being sunset...

Intel Announces It's Shutting Down Clear Linux - https://www.phoronix.com/news/Intel-Ends-Clear-Linux

18.07.2025 22:12 β€” πŸ‘ 10    πŸ” 3    πŸ’¬ 3    πŸ“Œ 0
Post image

#DΓ©fenseπŸ›‘| 🚨 La dixiΓ¨me newsletter du CEA est en ligne !
Tous les deux mois, la newsletter du CEA c’est un dossier d’actualitΓ© au ❀ de nos mΓ©tiers.
Aujourd’hui, zoom sur la propulsion nuclΓ©aire, un atout pour la France πŸ‘‰ bit.ly/CEANewslette...

11.07.2025 06:49 β€” πŸ‘ 8    πŸ” 1    πŸ’¬ 0    πŸ“Œ 0

Since Glenn is a storage guy… What about DDN? πŸ€”

11.07.2025 18:05 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Post image

Please join us in welcoming R‑CCS as our newest General Member πŸ™Œ

R-CCS brings deep expertise in high performance computing and optimization for Arm systems, and is the team behind the Fugaku supercomputer.

Read more: hpsf.io/blog/2025/hp...

08.07.2025 14:55 β€” πŸ‘ 4    πŸ” 3    πŸ’¬ 0    πŸ“Œ 0

#AMD #Instinct #MI300 Instruction Set Architecture Reference Guide:
www.amd.com/content/dam/...

09.07.2025 09:51 β€” πŸ‘ 6    πŸ” 7    πŸ’¬ 2    πŸ“Œ 0
Preview
Blackwell: Nvidia’s Massive GPU Nvidia has a long tradition of building giant GPUs. Blackwell, their latest graphics architecture, continues that tradition. GB202 is the largest Blackwell die. It occupies a massive 750mm2 of area…

Hello you fine Internet folks,

Today's article is on Nvidia's RTX PRO 6000 Blackwell and diving into the Blackwell architecture generally and more specifically into the GB202 GPU die in the RTX PRO 6000.

Hope y'all enjoy!

chipsandcheese.com/p/blackwell-...

old.chipsandcheese.com/2025/06/28/b...

29.06.2025 00:36 β€” πŸ‘ 36    πŸ” 4    πŸ’¬ 0    πŸ“Œ 1

Glenn Posted his ISC recap, and as always, it's excellent. The exert on Mixed precision and Ozaki did have me giggling whilst reading.

I think it's relevant to add some high level context on *why* we vendors decrease FP64; it's not just chasing AI, it's the shear *cost* of FP64.

The short ...

24.06.2025 14:51 β€” πŸ‘ 17    πŸ” 3    πŸ’¬ 1    πŸ“Œ 0
Post image Post image Post image Post image

Finishing some runtime system work; decided to try a #deskpi #super6c cluster board.

An ITX form factor beowulf cluster is amazing.

#slurm & #nfs worked out of apt.

#ucx, #openmpi, #openshmem, #openpmix, #gasnet, & #hpx needed custom compilation.

@raspberrypi.com #arm #hpc #supercomputing

23.06.2025 23:16 β€” πŸ‘ 18    πŸ” 3    πŸ’¬ 2    πŸ“Œ 1

Would be interesting to have CDNA 3 vs. 4 as well

23.06.2025 17:31 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

would greping for `\item` work?

22.06.2025 11:10 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I believe we should get more news on Alice Recoque by the end of summer.

I got to visit the room that will house the system this past Monday! Still under construction, but things seem to be progressing nicely. πŸ™‚

19.06.2025 10:04 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Thank you for your answer Andreas! I am looking forward to see the JUPITER system fully installed πŸ˜„

19.06.2025 06:48 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I know, we are also waiting on our samples to arrive, so I imagine that JΓΌlich haven’t received theirs either πŸ˜‚ SiPearl has been scrambling with Rhea1 for too long though. I don’t understand what’s taking them this much time to deliver on it… It’s not like they had to design the uarch from scratch.

19.06.2025 06:46 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Will the full HPL run also include the Rhea1s? Or is it only the score for the BOOSTER partition that you submit to TOP500?

18.06.2025 22:02 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Post image

@thoefler.bsky.social in the closing session of #ISC25 said his first and more sincere goodbye to our friend @hpcguru.bsky.social announcing his retirement! Torsten still hopes that HPC Guru plan will fail, but if that really happens we should thank him for such amazing years of #HPC together!

12.06.2025 16:30 β€” πŸ‘ 15    πŸ” 3    πŸ’¬ 1    πŸ“Œ 0
Post image

All things, with one exception, have to end

After years of sharing #HPC news and insights, it's time to log off

It was fun while it lasted and I remain grateful to all of you who followed and interacted

Stay curious, stay kind!

β€” @hpcguru.bsky.social signing off

#HPC #Farewell

11.06.2025 06:39 β€” πŸ‘ 110    πŸ” 14    πŸ’¬ 45    πŸ“Œ 3
NVIDIA Powers Europe’s Fastest Supercomputer NVIDIA today announced that the JUPITER supercomputer, powered by the NVIDIA Grace Hopperβ„’ platform, is the fastest in Europe β€” delivering a more than 2x speedup for high-performance computing and AI ...

JUPITER supercomputer, powered by the NVIDIA Grace Hopper platform, is the fastest in Europe β€” delivering a more than 2x speedup for #HPC and #AI workloads compared with the next-fastest system

nvidianews.nvidia.com/news/nvidia-...

#ISC25

10.06.2025 11:27 β€” πŸ‘ 3    πŸ” 2    πŸ’¬ 0    πŸ“Œ 0

At the #ISC25 vendor showdown, I think Bruno from Eviden captured a key difference between HPC and AI businesses. Paraphrasing, HPC people like to fuss with the system. AI customers don’t care about the system. Suggests why NVIDIA has been so successful. Meet the customers where they are.

10.06.2025 11:15 β€” πŸ‘ 11    πŸ” 3    πŸ’¬ 1    πŸ“Œ 0
Post image Post image

ISC 2025 HPSF Community Birds of a Feather (BoF) + marking one year since launch at last year's ISC πŸ”₯ with panelists:
πŸ’  Julien Bigot, CEA
πŸ’  Christian Trott, Sandia National Laboratories
πŸ’  Heidi Poxon, NVIDIA
πŸ’  Todd Gamblin, LLNL
πŸ’  Andy Warner, HPE
#HPSF #HPC #HighPerformanceSoftware #ISC25

10.06.2025 14:22 β€” πŸ‘ 7    πŸ” 2    πŸ’¬ 0    πŸ“Œ 1
Preview
CppNorth, The Canadian C++ Conference 2025: On coding guidelines, class invariants,... View more about this event at CppNorth, The Canadian C++ Conference 2025

CppNorth 2025: Write better C++!

Join Olivia Wasalski: "On Coding Guidelines, Class Invariants, and Special Member Functions." Master 5 guidelines & special member functions' interaction with invariants for cleaner, robust C++.
πŸ”— sched.co/21xRE
Tickets: CppNorth.ca
🍁 Toronto, July 20-23! #CppNorth

10.06.2025 17:09 β€” πŸ‘ 1    πŸ” 4    πŸ’¬ 0    πŸ“Œ 0
Preview
Intel Execs Launch AheadComputing, CPU Startup to Rival Giants In a surprising turn, top Intel researchers, led by CEO Debbie Marr, have left to launch a startup aiming to create β€œthe biggest, baddest CPU” ever. This bold move challenges semiconductor giants like...

AheadComputing: A group of researchers has departed from Intel to launch a startup with ambitions to create what they describe as β€œthe biggest, baddest CPU” the world has ever seen

www.webpronews.com/intel-execs-...

#HPC #AI

07.06.2025 11:09 β€” πŸ‘ 4    πŸ” 2    πŸ’¬ 1    πŸ“Œ 1
From Boolean logic to bitmath and SIMD: transitive closure of tiny graphs Let's say that we have a graph of at most 8 nodes (you can also think of it as a relation between 8 things), represented as an 8 by 8 Bo...

New blog post: From Boolean logic to bitmath and SIMD: transitive closure of tiny graphs

bitmath.blogspot.com/2025/06/from...

07.06.2025 17:20 β€” πŸ‘ 2    πŸ” 2    πŸ’¬ 0    πŸ“Œ 0

@dssgabriel is following 20 prominent accounts