Denmark has marked a significant milestone in computational research with the addition of 16 NVIDIA H100 GPUs to the DeiC Interactive HPC service, and 16 more are expected by the end of the year. The tender for the acquisition of the GPUs was awarded to Lenovo. This addition signifies a big leap forward for Danish researchers, who can now easily access the new GPUs through UCloud, and positions Denmark at the forefront of AI innovation. We recently had the privilege of talking with Dr. Valerio Rizzo, head of AI & SME at Lenovo’s EMEA (Europe, Middle East and Africa) division, about the AI revolution that is currently taking place and Lenovo’s role in the production of the world’s most powerful hardware for AI.
Collaboration between Lenovo and NVIDIA
NVIDIA is the company behind the industry-leading graphics processing unit (GPU), H100, currently the most sought-after hardware for AI and used extensively by some of the biggest AI companies in the world, including OpenAI, Microsoft and Google. H100 is generally 2-3 times faster than the previous A100 generation, and an impressive 30 times faster for AI workloads such as large language models (LLM). The recently announced supercomputer, Gefion, by Novo Nordisk Foundation (NNF) and the Export and Investment Fund of Denmark (EIFO) will make use of the same technology.
In the past, Danish researchers had to either search abroad or pay commercial cloud providers expensive fees to use these powerful machines. However, since January this year the 16 H100 GPUs have been available via the DeiC Interactive HPC service, provided by a consortium of universities consisting of Aarhus University (AU), Aalborg University (AAU) and the University of Southern Denmark (SDU). The H100 GPUs are hosted at the SDU data center and available via the user-friendly platform, UCloud, that powers the DeiC Interactive HPC service.
Lenovo’s role in the production of the H100 GPUs is to provide the liquid cooling system, which removes heat from the power-hungry GPUs. GPUs with this much power produce massive amount of heat, and the liquid glycerol used in the Lenovo’s “Neptune” liquid-to-air system takes the heat away from the GPUs, similarly to the way a radiator works in a car.
“Lenovo’s contribution to this machine is fundamental,” says Dr. Valerio Rizzo. “As the computational capabilities of the machine grows, the TDP (thermal design power) and heat generation also grows. The more heat you remove from the single node, the more nodes you can pile up into a rack, and the density of the GPUs will mean that you reduce the total amount of cooling devices needed in your data center.”
“Neptune” is used for the Lenovo ThinkSystem SR670, similar to the SR675 acquired by the consortium, which is currently leading the GREEN500 list.
“This makes it the most efficient single node server on earth,” says Dr. Valerio Rizzo.
Supercomputers and AI in general consume immense amounts of electricity, so reducing their carbon footprint is essential. This is a task which Lenovo continues to work on.
“We are working on the form factor and on the thermal envelope in order to host this ever-increasing TDP or power-hungry GPUs,” says Dr. Valerio Rizzo.
Democratization of technology
Valerio Rizzo himself has a background as a researcher before he joined Lenovo. He is incredibly excited about the advancements currently happening in his field thanks to AI.
“We are seeing things that were unimaginable just a few years ago. My background is in neuroscience, and what I’m seeing right now in the community is insane.”
Dr. Rizzo highlights a paper in which researchers used AI to reconstruct a song from the electrical signals recorded in the brain of the participants in the experiment. In this sense, the AI is reconstructing the experience of the participants listening to the song.
On top of this, AI is now reaching the point where it can reconstruct mental images from our brains.
“In the recent six months, I have seen at least three different papers on how you can decode or reproduce images out of magnetic resonance, so this is pretty exciting to me. I can imagine how this can have a beneficial impact on certain very dramatic conditions such as dementia or on neurological pathology in general,” says Dr. Valerio Rizzo.
It is examples such as this that underlines the current potential arising out of ensuring easy access to world-leading AI infrastructure for researchers.
“I come from biology and about 99% of biologist are not into IT stuff,” says Dr. Valerio Rizzo. “I was passionate about computer science and IT, so I developed myself as a midway between a biologist and a computer scientist, but this is not common. Jensen Huang (CEO of NVIDIA) has said that with Generative AI and Large Language Models, we are enabling domain experts to leverage IT tools to do a better work and to solve some of the most critical challenges of our time. This is the power of democratization of technology.”
Fostering the next AI generation
The democratization of technology is a vision that the DeiC Interactive HPC consortium share with NVIDIA and Lenovo. By making these powerful GPUs available via the user-friendly and interactive HPC platform, UCloud, accessing some of the most complex IT tools in the world has never been easier for the Danish research community. On top of this, Danish researchers will now be able to use this hardware for the most sensitive types of data, e.g. medical records or genomics data, something which has so far not been possible.
It also follows that the investment made by the DeiC Interactive HPC consortium is an investment in the future generation of AI researchers in Denmark.
“If you think about the variety of users, who will be able to access this type of system at a university – anything from students to full professors – you would expect that not all of them are experienced. Some of them will have to go through a learning curve and they will use the system inefficiently. Imagine translating this into a commercial cloud system – that would be a massive cost without a return of investment in the sense that once these people learn how to leverage this technology, the investment would probably already cover the cost of an on-premises alternative,” says Dr. Valerio Rizzo.
More and more applications for AI are available on the DeiC Interactive HPC. We just released a new app called “Chat UI” which provides a simple interface to powerful AI Large Language Models (LLM), similar to commercially available services like ChatGPT. The app, which is now available to all Danish users of the DeiC Interactive HPC, takes full advantage of the new GPUs to analyze texts and generate answers to complex prompts. The Chat UI app can use any of the many open-source LLMs available online, which are now competitive for most tasks with state-of-the-art commercial solutions.
Arrival of Gefion
Since January where the 16 H100 GPUs were made available via the DeiC Interactive HPC service, the Novo Nordisk Foundation (NNF) and the Export and Investment Fund of Denmark (EIFO) have announced that in collaboration with NVIDIA, they will establish a national AI innovation center featuring one of the world’s most powerful AI supercomputers. The new AI supercomputer will be equipped with more than 1500 H100 GPUs.
The initiative from NNF and EIFO complements the service available via DeiC Interactive HPC well. The service Gefion will provide is batch access to the H100 GPUs on a much larger scale in a similar way to traditional HPC systems. For interactive use, Danish researchers can benefit from the GPUs at the DeiC Interactive HPC and then use larger machines like Gefion for scaling up their research whenever needed.
We asked Dr. Valerio Rizzo how he thinks the new AI supercomputer might impact the Danish research community.
“It’s exciting to see how fast research is adopting AI technology nowadays, and some of the most critical problems today might find a solution by applying a certain kind of algorithm and methodology. The array of possibilities within the AI space is huge.”
More information:
We would like to thank Dr. Valerio Rizzo for taking the time to talk with us! If you are interested in reading more examples of how AI is revolutionizing neuroscience, Dr. Rizzo recommends the following papers:
https://medium.com/@VRizzo/non-omnis-moriar-5002e16dc95b
https://www.nature.com/articles/s41593-023-01304-9
https://arxiv.org/abs/2209.07162
https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002176
If you want to know more about Gefion, please see the NNF press release here.
You can find a link to an article recently published on our website about director of the SDU eScience Center, Prof. Claudio Pica’s, involvement in the establishment of the new national AI innovation center here.
If you are a Danish researcher, and you want to try out the H100 GPUs available via DeiC Interactive HPC, you can find the resources on UCloud now (select machine type u3-gpu). For further guidance, please contact your local Front Office.
