Tessera
Folgen

Inno4scale

@inno4scale.bsky.social

The project includes the take up of innovative algorithms created in innovation studies. If you want to learn more, give us a follow and stay tuned! https://www.inno4scale.eu/

35 Posts30 Follower31 Folgt
Inno4scale@inno4scale.bsky.social · 11mo ago

Inno4scale partners met today in Luxembourg for the project’s final review meeting with the @eurohpc-ju.bsky.social project officer and reviewers. We appreciated the constructive and positive feedback and look forward to the final report. Thanks to all involved for the strong collaboration!

Inno4scale Team in a meeting roomLuxembourg sign at the airportInno4scale Team in a meeting room
Inno4scale@inno4scale.bsky.social · last yr.

🆕 New from Inno4scale: A booklet with selected highlights from 22 innovation studies on algorithms for European exascale systems. Scalable solvers, mixed-precision CFD, sparse matrix kernels & more. Check it out ➡️https://www.inno4scale.eu/success-stories/

Cover of the Inno4scale Success Story Highlights Booklet
Inno4scale@inno4scale.bsky.social · last yr.

Matrix sparsity hides useful patterns for solver choice. Discuss the in the comments 💡🧠 🔬 What does the LimitX recommender predict from sparse matrix structure? 1️⃣ Solver memory usage 2️⃣ Spectral matrix properties 3️⃣ Eigenvector orientations Look for hints 👀🔍 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

We’re proud to share that Inno4scale was presented as part of the EuroHPC Joint Undertaking booth at this year’s ISC High Performance Conference in Hamburg! 🚀 A big thank you to everyone who stopped by the booth to learn more about Inno4scale, share ideas, and engage in insightful discussions 🧠 🖥️

BildBild
Inno4scale@inno4scale.bsky.social · last yr.

GPU stalls limit performance in exascale CFD solvers. Discuss the question below in the comments 💡🧠 🔬 What limits GPU scalability in coarse-grid solvers? 1️⃣ Too few elements per core 2️⃣ CPU driven communication stalls 3️⃣ Excessive matrix assembly costs Look for hints 👀🔍 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

Two way EL simulations struggle to scale due to stepwise synchronisation. SCALETRACK changes that. Discuss the following question in the comments 💡🧠 🔬 What enables scalable EL simulations? 1️⃣ Particle preaveraging 2️⃣ Async coupling 3️⃣ Interpolation caching Look for hints 👀 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

Sparse training for LLMs can reduce costs and emissions but hardware limits still apply. A new approach changes that. Discuss in comments on the question below 💡🧠 🔬 What enables full-range sparsity on GPUs? 1️⃣ Mixed precision 2️⃣ VENOM format 3️⃣ cuSparse Look for hints 👀 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

GNNs struggle with memory and speed. CBM4Scale compresses binary matrices by reusing similar rows. Discuss in the comments on the following question 💡🧠 🔬 Which approach supports performance portability? 1️⃣ OpenACC 2️⃣ Domain-specific libs 3️⃣ Manual CUDA Look for hints 👀🔍 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

‼️Exciting update‼️ Inno4scale Innovation Day goes hybrid! 💡 If you can't join us onsite in Barcelona, then watch the livestream on May 13 🚀 📺 rtmp://a.rtmp.youtube.com/live2

Bild
Inno4scale@inno4scale.bsky.social · last yr.

ML speeds up atomic simulations, but key challenges remain. Discuss in the comments! 💡🧠 🔬 What is the main challenge for MLPs? 1️⃣ Slow computation 2️⃣ Missing long-range effects 3️⃣ Limited training data Look for hints 👀🔍 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

Simulating space plasma at ion-kinetic accuracy demands exascale computing! But checkpointing is a challenge. Discuss in the comments: 🔬 What is AI’s advantage in checkpointing? 1️⃣ Less simulation noise 2️⃣ Higher plasma density 3️⃣ Faster storage & recovery Look for hints 👀🔍 www.inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

GW calculations reveal materials insights, while scaling remains diffucult. Discuss in the comments on the following question 💡🧠 🔬 What is the main bottleneck in large-scale GW? 1️⃣ Polarizability calculation 2️⃣ Wavefunction storage 3️⃣ Screened Coulomb W Look for hints 👀🔍 www.inno4scale.eu

inno4scale

Inno4scale is a European initiative, which was started to support the development of innovative algorithms for exascale supercomputers, so their efficient use can be fully exploited. Currently…

inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

HPC advances simulations but communication overhead limits scalability. Discuss in the comments on the following question 💡🧠 🔬 What improves CG solver efficiency on GPUs? 1️⃣ Kernel launch overhead 2️⃣ Floating-point precision 3️⃣ Memory bandwidth More information and hints ▶️

inno4scale

Inno4scale is a European initiative, which was started to support the development of innovative algorithms for exascale supercomputers, so their efficient use can be fully exploited. Currently existing codes for high-performance computing will not be able to function efficiently on upcoming exascale...

inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

Lattice QCD pushes precision physics, but scaling remains difficult. Discuss in the comments on the following question 💡🧠 🔬What is a major challenge in Lattice QCD? 1️⃣ Critical slowing down 2️⃣ Low lattice resolution 3️⃣ No experimental proof More insights and hints ▶️

inno4scale

Inno4scale is a European initiative, which was started to support the development of innovative algorithms for exascale supercomputers, so their efficient use can be fully exploited. Currently existing codes for high-performance computing will not be able to function efficiently on upcoming exascale...

inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

HPC accelerates simulations, but scaling remains a challenge. Discuss in the comments on the following question💡🧠 🔬 How do Neural Operators improve PinT? 1️⃣ No need for fine solvers 2️⃣ Faster with high accuracy 3️⃣ Replaces all numerical models More information and hints ▶️

inno4scale

Inno4scale is a European initiative, which was started to support the development of innovative algorithms for exascale supercomputers, so their efficient use can be fully exploited. Currently existing codes for high-performance computing will not be able to function efficiently on upcoming exascale...

inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

HPC drives CFD forward, but scalability remains a challenge.  Discuss in the comments on the following question 💡🧠 🔬 How does ML improve CFD scalability? 1. Replaces Navier-Stokes 2. Speeds up solvers & coupling 3. Removes computational grids More information and hints ▶️

inno4scale

Inno4scale is a European initiative, which was started to support the development of innovative algorithms for exascale supercomputers, so their efficient use can be fully exploited. Currently existing codes for high-performance computing will not be able to function efficiently on upcoming exascale...

inno4scale.eu

Inno4scale@inno4scale.bsky.social · last yr.

Now, you have had the chance to learn a lot about our innovation studies 🧠   Over the next few weeks, you can look forward to some exciting quizzes on our innovation studies and related topics! 🖥️ 🔢  You are invited to visit our website to refresh your knowledge ▶️ https://www.inno4scale.eu/

Bild