Found 42 total tags.
2 items with this tag.
Artificial intelligence (AI) encompasses a bunch of topics, but I use it as shorthand for generative artificial intelligence (Gen AI) which involves using transformers to generate stuff—text (including chat), images, videos, audio, and other media.
This page is a collection point for all the pages I’ve written that are pertinent to AI and the AI industry.
Government-sponsored Canadian HPC is funded through ISED and, as of 2025, has two major components:
As of September 2024, Canada is the only G7 nation without a Top 25 supercomputer, and Canada’s public supercomputers are not available for industrial uses.[^2]
Unlike the US national HPC efforts, Canada favors an arm’s-length nonprofit model of funding national cyberinfrastructure. Whereas DOE or NSF might sponsor FFRDCs or universities to deploy and manage HPC infrastructure, Canada sponsors non-profits (like DRAC) which are created and established explicitly to do this. Government Canada does not take as much of a hands-on approach in these nonprofits as DOE or NSF would for their investments.
Government Canada maintains a list of organizations that have signed its Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems which is a good starting point of Canadian parties interested in AI.1
Formerly Compute Canada. See Digital Research Alliance of Canada.
Mitacs is a Canadian arm’s-length nonprofit that promotes partnerships between academia, industry, and government through technology. While not exclusively focused on HPC or AI, it sounds like they route government funding to industry to promote collaboration with public researchers.
Established in 2023, PAICE is an initiative led by a coalition of2
Like the Canadian sovereign AI efforts, it is funded by ISED, and is the organization that is the steward of three AI supercomputers:
| Name | Location | Specifications |
|---|---|---|
| TamIA | Mila (Québec City) | 22x 4-way H100 |
| Killarney | Vector Institute (Toronto) | 168x 4-way L40 + 10x 8-way H100 |
| TBA | Amii (Edmonton) |
Who is evaluating SCIP SOIs and proposals?
ISED is, but they have been receiving education on how other nations do these evaluations.
What is the balance of public/private partnership expected for this effort?
SCIP is similar; companies like Hypertec or Denvr cloud are unlikely to “prime” since the structure is likely to take one of a nonprofit with a large university-led component. Expect the SCIP systems to be large, on-prem systems.
2 items with this tag.
4 items with this tag.
5 items with this tag.
11 items with this tag. Showing first 10 tags.
European HPC is divided into a couple notable segments:
5 items with this tag.
3 items with this tag.
20 items with this tag. Showing first 10 tags.
20 items with this tag. Showing first 10 tags.
5 items with this tag.
14 items with this tag. Showing first 10 tags.
3 items with this tag.
17 items with this tag. Showing first 10 tags.
Let’s try keeping notes of the interesting papers and articles I read in Obsidian so that I can reference them in my other notes.
1 item with this tag.
1 item with this tag.
1 item with this tag.
6 items with this tag.
In HPC, there are two ways to look at the reliability of a supercomputer.
Top-down reliability is where you start with what a full-scale system job would experience in practice and begin breaking that down. Top-down reliability is governed by metrics that characterize job reliability.
Bottoms-up reliability is where you start with individual components and build a reliability model by connecting those components in series and in parallel. Bottoms-up reliability is governed by metrics that characterize component reliability.
The statistics that govern how parts of a system affect the whole thing are described in MTBF.
3 items with this tag.
Seedlings are a tag I use to indicate that a page has not been tidied up. I likely just pasted in some notes while I was in the middle of something else and will come clean it up later.
Seedling pages do not show up in the Explorer sidebar, so if you stumble upon a page but can’t find a way to navigate back to it, it’s probably a seedling.
I spent much of my career in HPC focused on storage and high-performance I/O. Specifically,
It is a complicated and not-well-understood aspect of HPC, so the community of researchers and practitioners around HPC storage is small and dedicated. I stopped working on storage when I joined Microsoft Azure’s AI infrastructure team, where the biggest challenges arose from compute, not storage.
I don’t miss working on storage, as I felt I had caught up to the cutting edge of it and there wasn’t enough innovation to warrant my focusing on it as a full-time job. Being “the storage guy” also required that I pay attention to some uninteresting aspects of HPC storage such as data management, compression, and topics like that. However, I miss the community and the smart people who dedicated their careers to I/O.
20 items with this tag. Showing first 10 tags.
2 items with this tag.
1 item with this tag.
2 items with this tag.
The U.S. Department of Energy is the largest sponsor of basic science research in the United States. It’s sometimes jokingly called the “Department of Everything” because of how broad its scope is.
ASCR (pronounced like the name “Oscar”) is the part of the U.S. Department of Energy dedicated to open-science research. By definition, ASCR does not touch nuclear weapons.
4 items with this tag.
NNSA is the part of the U.S. Department of Energy responsible for nuclear weapons.
2 items with this tag.
3 items with this tag.