This is a popular forum for CFD people (and OpenFOAM users) www.cfd-online.com/Forums/openf...

Used long time ago - basically when someone uses OpenFOAM compared to commercial solvers (Ansys, StarCCM+) - they are looking to add their own source terms or are writing a solver of their own based on some examples. The actual solver depends on the type of application e.g. flow properties.

Pro tip, AI generated resumes go through AI filters more easily (use the Job Description in the prompt and your CV and ask it to generate your tailored CV for that job). Let AI fight AI. 😀

Hiring process is more humane in National Labs compared to Big Company - but the pay difference is high. For national labs, you can reach out to relevant people and just talk. For Big Cos, it is broken by design I think.

@glennklockwood.com Thanks for sharing your thoughts and best of luck for your future. Lots of your personal reflections resonated with me - Hope you find the balance you are looking for 😄

you know feeling - "this meeting could have been an email 😀"

In some science you might be ok with less precision (depends on the range of floating point numbers you need and accuracy of their operations), but some fields need high precision. In the end developers will have to work with the silicon they get.

You are more experienced than me :) so I won't be as accurate as you - but I guess, the worry is that silicon will now be decided by the #AI needs which can work on lower bits (half precision etc.) and since its demand is so high #HPC workloads will not have a voice.

PI - principle investigator. basically they write the research proposal for their team and apply for funding. they are usually lecturers or professors in the university working with a team of PhD and Post docs in their research lab.

My previous idea for a #HPC on cloud training course was shot down by my company's legal department. So this will be just me volunteering in individual capacity.

Hi, I'm trying to find my place in the diverse and rich #hpc community - I bring knowledge from different backgrounds - CFD, performance engineering, software development, training, DevOps etc. Any ideas if any group is looking for some help where I can be helpful. If yes, please let me know by DM.

Sounds good :) Let me look into the docs for google cloud and create something, I can share a draft with you if you would like to provide feedback. It will just take me some time (few weeks) due to day job

Thinking about creating a course for HPC in Cloud, since many are forced to move to cloud due to budget cuts. It will be posted for free online. Would anyone here be interested in it? your feedback would help me steer it, and give me motivatation to actually implement it :) #HPC

Getting ready for ISC 2025. Let's meet in person if you are attending as well (BTW early bird rates are open now) #HPC

Nice article in CACM - Meta’s
Hyperscale Infrastructure: Overview and Insights - dl.acm.org/doi/pdf/10.1...

#hpc

Linus on his workstation

Linux Headquarters

Universities charge upto 40-50% of research grant for admin. I will not comment if this right or wrong, I don't know enough on it. I'm just saddened that this will ultimately affect researchers and the students they hire and the research community which is already strained by the system #hpc

If you need better speeds you have to sacrifice portability - writing custom data structure/algorithms for your hardware, or in extreme cases custom kernels written in assembly to bypass compiler because they can make mistakes sometimes in optimizations #HPC

I guess it is just a nature of things that scripting languages are slow compared to compiled languages, because of Just In Time compilation and need for portability. E.g. Julia will compile the script on the first run and will be slow, but on the subsequent runs it will be faster (similar to C)

One point to note is that scripting languages don't slow down because deepcopies (complex objects e.g. in python are not deepcopies but just references upon copies)

I thought this was commonly known. You are right, compute stopped being expensive a long time ago - around 2004ish I guess. The main bottleneck is always data transfer - different speeds at different parts of the computer. Removing one bottleneck, exposes another.

I once tried to play with Julia for teaching #HPC. Didn't like that it was not possible to compile before first run. Left a bad taste. Learning Rust, single core performance seems fast (feels steeper learning curve than C++ for some reason), but not sure about MPI support.

I guess it is an opinionated terminal emulator and the new shiny thing. BTW, for iterm2 there is an advisory to upgrade it due to a security bug, if you haven't seen it before - iterm2.com/downloads/st...

in the naughty list of employees. I ended up on it and had a meeting with my VP since HR complained.

😂 I always thought who are these people, reading posts on LinkedIn - they just seem weird, I called some posts out for promoting "work at all costs", even over life and family kind of way. Now, I know why this happens. The HR of the company is monitoring the posts, and people don't want to end up

I have made a theory that the higher someone goes in management, he/she becomes fluent in keeping opposing thoughts in the mind at the same time. These thing make an engineer mad, because for him/her only one can be true. Either the company wants to increase productivity or it is all a show.

And the special part was that there was no accounting for this time. Removing that All hands meeting would be the most productive thing the company could do. But that would be Blasphemy.

Ahh. you bring back dark memories. In one of my employers we had meticuluos time sheet system for every project with time accounted for working, travelling, meetings etc. to account for cost associated with each project. But on the other hand we had weekly All hands for the whole company.

healthcare is a policy issue, so it might not be solvable by technological advances - it is very difficult for people to agree on some topics. #hpc and #ai are being used for better drug discovery for example. If you mean LLM, try med42 models. But they can only enhance doctor, not suitable for user

Someone is calling for the church of Emacs :) www.gnu.org/fun/jokes/go...