Nvidia launches Cambridge-1, UK’s most powerful supercomputer, in Arm district • The Register
Nvidia has launched its Cambridge-1 supercomputer, focused on AI for healthcare, as part of larger plans to boost AI research in the University City itself and more broadly in the UK .
The Cambridge-1 is Britain’s top-ranked supercomputer according to the Top500 list, currently sitting in 41st place worldwide with maximum performance of 9,682 TFlop / s – although the fastest in the world, the Fugaku supercomputer, can handle 442,010.
The supercomputer was announced in October of last year when it was due to “go live by the end of the year”, so it’s somewhat delayed, but still being built quickly through use. of Nvidia’s modular module. SuperPOD DGX, described as a ‘turnkey AI data center’.
The actual deployment of the supercomputer took less than 20 weeks, Nvidia told us. The costs have risen a bit, however: the October post spoke of an investment of £ 40million, while at today’s launch the figure given was $ 100million (over $ 72million). of books).
The Cambridge-A1 consists of 80 DGX A100 systems connected by an Infiniband network. Each DGX A100 has 8 A100 GPUs, 320 GB of GPU RAM, two AMD EPYC 7742 processors (64 cores each), up to 2 TB of system RAM, up to 30 TB of NVME data cache and 2 SSDs 1.92TB NVME.
The A100 GPU is Nvidia’s high-end Ampere GPU, successor to the Tesla V100; it is described in detail here. There are 6,912 CUDA (Nvidia’s GPU programming platform) cores per GPU, to support massively parallel computing – and there are a total of 640 A100 GPUs in the system.
The only way is Essex
The physical location of Cambridge-1 (despite the name) is Harlow, Essex, at the Kao Data data center, chosen in part for its green credentials: The company claims to use 100% renewable energy, meet EU data center environmental guidelines, and use 100% free cooling, which means cool air comes from the outside atmosphere. Harlow is on the ‘Cambridge Innovation Corridor in London UK’, mentionned Kao Data, doing his best to link Harlow to Cambridge.
“This is our first supercomputer dedicated to healthcare, and possibly the world,” said David Hogan, director of Enterprise EMEA at Nvidia.
Aid to science and medicine
The initial setup involves five “founding partners”.
These are GSK, AstraZeneca, Oxford Nanopore, Kings College London (KCL) and Guy’s and St Thomas NHS Foundation Trust.
We spoke with Professor Sébastien Ourselin from KCL, responsible for biomedical engineering and imaging sciences. Ourselin is also deputy director of the AI center for value-based healthcare, a joint project mainly between KCL, Guy’s and St Thomas’, along with other NHS trusts, universities and technology providers, including Nvidia.
“For the past three and a half years, Nvidia has been our primary partner in our healthcare artificial intelligence program at Kings,” he said. The Reg.
Ourselin said Cambridge-1’s capability is needed for its healthcare AI to be fair and accurate. “With AI, there are two big challenges to solve. One is access to data at scale, and the second is access to computing power at scale. A lot of people can create AI models. The challenge is how to ensure that this can translate into an impartial and fair model, and in order to do that, especially in healthcare, you need access to a huge amount of data representative of your population.
“For Nvidia, being part of a consortium like the AI Center was a way to access a lot of data together with us and help us train these models,” he said.
KCL specializes in AI based on image data such as MRI scans. “If you try to predict or diagnose based on the image information, you are basing everything on training. So what you ingest is what the algorithm will be able to learn. If now your population is, say, only Caucasian, what if you now present non-Caucasian data? It might not work.
“What if your dataset only represents a healthy population? He might never find a disease. What if the dataset was only for males? If you want to increase the dataset, the model becomes much more complex to explain all the data. And the only thing is to have great computing power. This is where Cambridge-1 came in; it allowed us to train models on a scale that we weren’t able to do before, ”he said.
KCL has its own supercomputer which is “much smaller than Cambridge-1,” Ourselin said. This meant that testing a new model could take three weeks, limiting the number of tests that could be performed. “If you can get an answer overnight, you can speed up the search,” he said.
Confidentiality is an issue. “The data from the NHS will be transferred to the supercomputer in a fully secure enclave so that the only people who can access the data are the people in the hospital. In all of the experiments we are doing, the data will never be seen by anyone. ‘another It’s like the supercomputer is an extension of the hospital.
“The idea is to build what I call a smart NHS where you can send a lot of data without having to anonymize, to get all the richness of the data, and you are able to build multidimensional models,” a- he declared. says – words that may concern privacy advocates, as supposedly secure data leaks are not unheard of.
How much will KCL cost? “We are not paying for this resource. It is provided pro bono by Nvidia, which has invested over £ 40million in the supercomputer. We are not paying anything. We are not charged for the use of the supercomputer. let’s keep all the intellectual property that we generate.There is no financial model where we have to pay anything.
The initial access is one year, Ourselin told us, but he expects the grant to continue, “because we show them value.”
What is the value of Nvidia?
“CEO Jensen [Huang] was very keen to contribute, for society, not entirely on a financial model, but to demonstrate that AI can have a positive impact on society, ”Ourselin said.
“Nvidia offered to not only provide Cambridge-1 to these five founders, but to provide it to a much larger community in the future. In order to be able to do that, they first wanted to demonstrate through a set of concepts that l ‘adding such a capacity to be truly transformative. “
It may be true; but we note that Nvidia is in the process of acquiring Cambridge-based Arm, subject to regulatory approval, and keen to present this transaction in the best possible light.
Nvidia has a sub-site on the subject, including a page on “Our Vision for the UK” which states that “Nvidia intends to achieve its vision of creating the leading IT company in the AI age here in Cambridge – propelling Arm and the UK to leadership world of AI “.
Specific plans include a “new world-class AI research center” of which Cambridge-1 is a part. Cambridge-1 is based on AMD, but the company is planning a future supercomputer “powered by Arm processors and Nvidia GPUs … at the iconic Arm headquarters.” Cambridge-2 maybe?
Still, as Ourselin told us, Nvidia crashing a supercomputer in the Brexit-hit UK is a nice boost at a difficult time. “There are many reasons to celebrate why an American company comes to the UK and invests in the UK. It could have installed a supercomputer anywhere in the world,” he said. ®