We are sad to hear of the passing of Gordon Bell, a pioneer in high-performance and parallel computing and the visionary behind the ACM Gordon Bell Prize. His dedication to innovation inspired countless breakthroughs. Our deepest condolences to his loved ones.
In the process of debugging a NUMA first-touch problem, I accidentally found my simulation becomes significantly faster when it's running on garbage data without memset() - even on non-NUMA systems... What?! Does the kernel provide a fast-path for uninitialized memory that I've never heard of?
"a read from a never-written anonymous page will get every page copy-on-write mapped to the same physical page of zeros, so you can get TLB misses but L1d cache hits when reading it."
New to the Linux Australia #jobs board: the lovely folks at National Computational Infrastructure are looking for a #HPC#Linux administrator.
#NCI is the leading national provider of high-end computational and data-intensive services. It forms an integral part of the Australia Government’s #research#infrastructure strategy.
--
For more details, or to apply, please visit the listing on the ANU Jobs portal at:
Preliminary results of the CPU memory bandwidth micro-benchmark: If you're jumping in memory randomly, try doing number-crunching on at least 2-4 KiB of contiguous segments of data before jumping again to amortize the latency penalty down to an acceptable level (throughput is ~70%-80% compared to a sequential pattern). #hpc
@asltf@thijs_lucas Das Simulationsprogramm heißt MATSim und kann im Prinzip alle möglichen Verkehrsträger berechnen. Im gezeigten Bild sind Auto, Fracht, Fahrrad und zu Fuß enthalten.
Normalerweise haben wir auch noch öffentlichen Verkehr mit drin, hier aber nicht, weil es eher um den Informatikaspekt und weniger um die Simulationsstudie geht.
Wir können dann auch noch Dinge wie Demand Responsive Transport (Taxi, Uber) abbilden.
#HPC#CUDA#OpenCL#LAPACK
If you had to do a lot of linear least square solves, with potentially rank-deficient matrices, what would you use on a GPU? On CPUs, LAPACK's DGELSY does work, but most GPU libraries seem to not implement routines for rank-deficient matrices.
ProTip: Do NOT set OMP_PROC_BIND and OMP_PLACES globally. At least for GOMP, it breaks multiprocessing in many non-OpenMP applications. I wrote them into /etc/environment then I started wondering why Gentoo's code compilation is only able to use 1 CPU core. :woozypad: #hpc
@niconiconi Reminds me that OMP_WAIT_POLICY has to be PASSIVE for ngscopeclient to work properly and I never figured out why. I think it has to do with multiple threads spawning OpenMP tasks at once and getting confused, idk.
I'm gradually moving the project away from OMP and towards application-managed threading or GPU processing anyway so that might end up being the fix.
Today, Intel celebrated the opening of Fab 9 in Rio Rancho, New Mexico
The milestone is part of Intel's $3.5B investment to equip its New Mexico operations for the manufacturing of advanced semiconductor packaging technologies including Foveros
Finyally understood how to generalize the diamond tiling algorithm from 1D+1T spacetime to 2D+1T spacetime for stencil code. This mysterious diagram now makes perfect sense. But to the researchers at Keldysh Institute of Applied Mathematics, this algorithm was already obsolete for 20 years. It was already in use in Russian HPC code of the late 1990s and was known as "ConeTur". In the mean time, they invented 3 generations of newer algorithms, which are even more incomprehensible. Keldysh is at least 10 years ahead of the rest of the world... #hpc
@gorplop A related field is polyhedral compilers, which can automatically do loop transformations in this manner, and is the mainstream of research today. GCC's Graphite optimizer is an example, and there are many HPC-specific ones. The idea is that they are general-purpose, given an unmodifed loop, they can apply extremely complicated patterns beyond human understanding. But the heavy focus on automatic code generation means that if your code doesn't match the pattern the compiler already know, it likely won't perform very well. On the other hand, this Keldysh research team's algorithms are designed by hand for specific algorithms and all have geometric interpretations, and are meant for for human use - which appears to be rarely studied or implemented today by anyone else as they're just too difficult to reason.
