(pretty long) report on the 16th Annual Swiss Conference of the HPC•AI Advisory Council

(pretty long) report on the 16th Annual Swiss Conference of the HPC/AI Advisory Council

The event took place from 7th to 9th of April, 2025. in Lugano, in lovely Ticino. (This is the Italian-speaking part of Switzerland. Sometimes, this caused some problems because my Italian is non-existent, but many people in Lugano speak German, e.g. at the hotel.)

The HPC/AI Advisory Council is an organization/association which was originally established and funded by Mellanox (now part of Nvidia). It should initially offer the opportunity to test Infiniband. Now, they offer some more services in the HPC and AI space. Therefore, they operate a few (small) clusters. Today, the AC has >400 members (companies, universities, research institutes…). More informtion can be found here: https://www.hpcadvisorycouncil.com

The agenda of the event is available here: https://www.hpcadvisorycouncil.com/events/2025/swiss-conference/agenda.php

On Sunday, the 6th of April, I travelled to Lugano by train. I was impressed how punctual trains can be: the trains I took were operated by SBB, the Swiss state-owned railway system – no comparison to the Deutsche Bahn, the German state-owned railway operator. The reason is that Switzerland spends four to five times more money (per citizen) on their railway system than Germany.

I arrived in the late afternoon at my hotel Pestalozzi which is very close to the Palazzo di Congressi, just across the street. In the evening, I had dinner at the hotel‘s restaurant (I love Italian food, it is almost as good as the French one ;-). I knew the hotel and restaurant from my first trip to the conference in 2019, so six years ago. The hotel has been renovated quite nicely, in the meantime. The rooms were/are simple but have everything one needs. And the dinner was again very good: veal escaloppes in White wine sauce with fried potatoes and fennel.

The next morning, after a good breakfast (Italians usually do not set value on a good „prima colazione“) I went to the congress center:

Before the event started I ran into Eric Schnepf, senior advisor fpr ISC Group: we had a „pre-conference coffee/cappuccino“.

The event started at 9am, with an opening/introduction by Brian Sparks from the AC and Hussein Harake, the local host from CSCS (Centro Svizzero di Calcolo Scientifico).

The first keynote was given by Siddharta Mishra of ETH Zurich. Its title had been changed from „AI for PDEs“ to „AI for Physics“. In his talk SM gave a lot of examples where AI „helps“ HPC applications.

The first „normal“ talk was given by Robert Henschel of Indiana University. He talked about „The Case for High Performance Research Desktop“. Later in a break, I learned that RH is German. I told him that I did not expect very much from his talk but that he outwent my expectations. Of course, he was delighted to hear this. 😉

I very much liked the following slide:

Later, I told him that I especially liked his comments on users but that he had generalized too much: not every user is „dumb“. He agreed.

Some more slides from his talk:

In the following break (great coffee prepared by a barista) I talked to several people:

Michael Hennecke:: he is with HPE now (before Intel, before IBM)
he is storage expert, since a few years working on DAOS

I know him from my early days at KSR:::

he showed in a paper that a random number generator

did not work on the KSR, i.e. that it did not generate „really random“ numbers

Marco Merkel (GM of doITnow Germany)
I know him since his time at Panasas
later he was with ThinkParQ, the company behind BeeGFS

a guy from Kaytus, a HPC hardware vendor
he is concerned about DJT‘s tariffs

a guy from E4, an Italian company with a focus on HPC
I asked him to send my greetings to Fabrizio Maguliani

After the break, the conference was continued with a talk by Ben Cumming of CSCS titled „Modern Software Deployment on a Multi-Tenant Cray-EX System“. Tbh: I do not remember a lot from this talk.

The following talk was titled „Facilitating Synergies Between AI and HPC: Update on HPE SmartSim Strategy“ and given by Alessendro Rigazzi of HPE. Here is a photo that I made of one of the slides:

I skipped to following talk „Building Switzerland‘s Open and Transparent Language Models“ by Imanol Schlag of ETH to answer some emails and be the first in line for lunch ;-): the buffet was delicious.

I also skipped the first talk after the break: „Alps Research Infrastructure at CSCS: Enabling World-Class ML Research in Switzeerland“ by Mohamed Fawzi Roberto of CSCS.

But I listened to the following talk „From Vision to Velocity: The VDURA Story and the Unveiling of the V5000“ by Kieran Penwill of Vdura, fka Panasas. (Btw: the company name comes from its slogan „where velocity meets durability“.) This was the worst presentation of the conference: the guy was reading his text from the screen of his laptop and obviously did not know a lot what he was talking about. But he was assisted by his colleague from presales: Sven Neirynck who gave a much better (short) presentation. From him I learned the main advantage of Vdura‘s new product V5000: the flexibility of having different percentages of a storage system implemented with HDDs and flash. In a break during the second day, I talked to SN at their „booth“ in the „mini exhibition“ (there were 8-10 companies presenting their products: Vdura, E4, Storj, doitnow, Hammerspace, Huawei, VAST Data, …). I said to him that a colleague from HP told me many years ago that hard disks won‘t go away. SN agreed and added that the main reason is the difference in price: a factor of 5 to 7. VAST however, which offers a flash-only solution, uses different types of SSDs on the different storage levels with different price tags.

After the following coffee break I skipped the rest of the presentations of the first day: „Using Natural Language Models to Support Pharmaceutical Application“ by Valentin Planes (Do IT Now) and „Practical Performance Modeling for Large-Scale Distributed GPU Workloads“ by Marcel Ferrari (ETH) and Nina Mujkanovic (CSCS).

Instead, I enjoyed a few minutes at fresh air on the HPCsquAIre 😉 in front of the congress center. Here are a few impressions:

The building in the background is my hotel (Pestalozzi).

Afterwards, I went back to the hotel and had dinner at the restaurant, again: Penne all‘arrabbiata – it was very good. I also had „una birra senza alcohol“.

The second day

started with a keynote by Taras Gerya (ETH Zurich): „Computational Biogeodynamics: Deciphering Influence of Plate Tectonics on Life Evolution and Habitability“.

This presentation was very interesting: biogeodynamics is a multi-disciplinary combination of biology, tectonics, climate research etc (a relatively new area of science). A better explanation gave the introductionary slide:

One key takeaway from this talk was that there is a strong correlation between tectonics activity and bio-diversity: the more tectonics activity are in an area of our planet the higher is the probability of a higher level of bio-diversity, i.e. the number of different species. Astonishing, isn‘t it?

Later, TG showed some findings about ACCs = actively communicating civilizations, i.e. he was talking about extraterrestrial life, i.e. aliens.

TG also used one of my favorite words/terms: Cambrian (explosion).

A new FLA I learned was COPT (planets): these are planets with continents, oceans, and plate tectonics, obviously a prerequisite for life, see also: https://ui.adsabs.harvard.edu/abs/2023AGUFM.P21B3003G/abstract

Again: this talk was very interesting, probably the most „innovative“ one of the whole conference.

I skipped the other two talks of the first session of the second day: „Developing an ML-ready Framework for Numerical Geodynamic Modelling“ by Marcel Ferrari (ETH) and „Evolving HPC: towards a Reconfigurable Paradigm for a Data-Intensive Era“ by Giordano Mancini (E4), again for answering emails.

Instead I spoke pretty long with Christopher Huggins (and two other guys) of VAST Data, namely about EU bureaucracy, SICOS, HPCsquAIre, … I know him since my time with T-Platforms:: we were looking for a partner who takes care of hardware support and met several times at SC and ISC but never came to an agreement.

I also talked to guys from ETH Zurich & U of Bergen, namely about the rise of AI and its implications:: regulations, automonous driving, Ray Kurzweil‘s book „Singularity“, …

And then I had a short chat with Axel Köhler, solution architect at Nvidia: I know him since he was with Sun (we were competitors as I was with HP then). I also was in concrete contact with him during my time at T-Platforms. This time, he somewhat ignored me – probably, he takes amiss that I mentioned him in a comment on Linkedin: I thanked a guy (Pavel Apanasovic of hydrograv) who shared a report on the unreliability of Nvidia ships and mentioned that Axel has denied this – well, I like my situation as a retiree: I don‘t have to be considerate of anything ;-).

After the break I skipped another talk: „Sarus and the Future of HPC Container Services at CSCS“ by Alberto Madonna (ETH Zurich / CSCS).

But then I listened to the talk of my „old friend“ Jan Heichler (presales manager at VAST). I know Jan also since my T-Platforms time, he then was the presales manager at Clustervision, so a co-worker of Christopher Huggins (see above). Tbh: I do not remember a lot from this talk, except that he mentioned that he will make some controversial statements. What I remember that it was not typical marketing bs.

Again, I skipped the following talk „FLASK: Foundation Learning AI for Synthesis Knowledge“ by Brian Van Essen (LLNL) to be one of the first in line for lunch. 😉

After the lunch break during which I talked again to the guy from E4.

Also: Axel Köhler was talking to me again, promised to roll out a red carpet for me the following day (which he did not, of course ;-). Also had a chat with Uli Michaelis of Nvidia and the “dream couple” from Megware: Tobias Pfennig and Peter Großöhme, and Holger Schwambeck (Ilmenau U).

After lunch I skipped all sessions talks (see the agenda) because I had heart problems (arrythmia) and also I had to prepare for / rest before the social event: dinner at the Hotel Splendide Royal (*****):

During the dinner I sat between Michael Hennecker (see above) and I guy from a Swiss company named Lake Solutions, a system integrator. Especially with the latter I had a very good conversation about all kind of topics: Swiss and German railways, tariffs, DJT, …

Whereas I went to the hotel by bus (the participants got the Ticino ticket, a free pass for public transportation) I ordered a taxi for the way back to the hotel because it was already pretty late and I could not walk the distance. I invited two guys to join me taking the taxi: one of them (his area of research: biomechanics) was also staying at the hotel Pestalozzi, the other one in a hotel close by. Btw: the taxi was impressive: a (very) big and (very) luxurious Mercedes, bigger than an S-class, probably a Maybach. And btw: the trip was pretty expensive:: 30 CHF for a drive of less than ten minutes (the same was true for the taxi drive from the railway station to the hotel on Sunday). In general, Switzerland is a very expensive country. However, my hotel rate was okay: ~120 CHF per night including breakfast.

The third day

started with a keynote by Torsten Hoefler of ETH Zurich: „Next Generation AI and HPC Networking with Ultra Ethernet“.

I am following this guy for many years now. The first time he took my attention was during an SC conference, about 15 years ago. I do not remember what he was talking about but what I remember it was a very good talk. I said to myself: this guy will make it, and he did. (I told him that during a break). In the meantime he is a professor at ETH and won several prizes. He published a lot of papers, on various topics of hpc, parallel programming, network technologies etc. He is still pretty young, around 40 I guess. In summary: he is one of the smartest guys on this planet I know.

Again, I skipped the other talks until lunch: food was why I went to Lugano. 😉

After the lunch break, Brian Sparks of the AC told the audience that the first talk had to be canceled because the speaker from Los Alamos did not get the permission to travel, thanks to DJT.

But the next presentation took place: Michael Hennecke (see above) talked about „Converging HPC and AI Storage: The DAOS Storage Architecture“. Tbh: I did not understand a lot. 😉

Afterwards, I skipped the rest of the conference because I was pretty tired after almost three days of intense input and communication.

In summary: the event was worth spending 150 CHF and three days.

Epilog:

In the evening, I had dinner at the hotel‘s restaurant. While I was waiting for my coffee after my dish (veal sausage with rösti and peas) Eric Schnepf and his girlfriend entered the restaurant. We had a short chat, then I went to my room and bed.

I stayed almost two more days in Lugano, some impressions: