@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250324

Episode Date: March 24, 2025

- GTC25 Quantum Day - Post-Quantum Cryptography guidelines - Quantum Blockchain? - NVMe-enabled HDD storage with SSD cache [audio mp3="https://orionx.net/wp-content/uploads/2025/03/HPCNB_20250324.mp3"][/audio] The post HPC News Bytes – 20250324 appeared first on OrionX.net.

Welcome to HPC News Bytes, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Welcome to HPC News Bytes. This is Shaheen Khan. Doug is away. NVIDIA dominates the fabulous semiconductor market as well as the AI revolution, so a lot of the news last week was at or linked to the GTC conference. One thing we did not cover was the so-called quantum day at GTC with two fireside panels

with quantum computing execs from in alphabetical order, Alison Bobb pursuing cat qubits, atom computing, building neutral atom, AWS also cat qubits, D-W build a neutral atom. AWS, also cat qubits. D-Wave, which has an analog quantum annealing system, as well as a separate project to build a gate-based system. Inflection with a queue, with neutral atom technology looking at sensing as well as computing. Ion queue, with trapped ion approach. Microsoft, topological qubits.

Pascal, also with a Q, also neutral atom, PsiQuantum, silicon photonics quantum computing, Quontinium trapped ion, quantum circuits building a modular superconducting system, Quera computing also neutral atom, Rigetti superconducting, and CQC supering, but also a digital interface for quantum classical integration. But the day included talks by other organizations, too, like Quantum Brilliance, the MITRE Corporation, Quantum Machines, Hewlett Packard Enterprise, QBraid Company, and various US and international universities.

And of course, NVIDIA, who covered hybrid quantum classical systems and running quantum applications on GPUs using CUDA-Q and Q quantum libraries. Think of CUDA-Q as an environment for hybrid quantum classical computing, which then uses Q quantum libraries to simulate the backend and run quantum code on GPUs.

As you recall, Jensen had been asked about the space, and he reflexively shared his opinion that he thought it'd be another 15 to 20 years for the technology to do quote, useful work. That got reported widely and led to a surprisingly large drop in quantum stocks. Presumably this session was planned before that, and simply having it on the calendar had helped raise the stock price's sum, but the session didn't really change the sentiment, and stocks dropped again a little bit. As you've heard me say here, the product roadmaps for quantum computers start looking very interesting in the 2030 to 2035 time frame. So 10 years is a more reasonable horizon to account for unforeseen progress or

setback. A lot is going on. The progress is real, but so is the hype. So you have to be excited and careful. And that's hard to do. There were several announcements tied to this event, notably NVIDIA's plan to form a hybrid quantum classical research lab in Boston, together with Harvard and MIT, and including Quantinium, Quantum Machines, and Quera Computing. The lab is called the NVIDIA Accelerated Quantum Research Center, or NVAQC, I suppose, and also announcements by Inflection and NVIDIA about Contextual Machine Learning, CML, and by CQC and NVIDIA on quantum classical connectivity. A few years ago, I emphasized the need to look at quantum advantage as not only quantum performance advantage, but also quantum energy advantage and quantum accuracy advantage. I was

happy to see that language show up with some panelists talking about energy savings, which is the big issue with GPUs. All energy savings, which is the big issue with GPUs. All in all, the sessions drew attention to quantum computing and provided technical training. The main stage panels were good and informative, but if you are a regular listener to this podcast or track quantum computing, you didn't miss much from the executive sessions. A big attraction of quantum computing is its expected ability to crack existing encryption. This week, the UK government published guidance about the implementation of quantum resistant encryption

algorithms towards so-called PQC, Post Quantum Cryptography. They want UK organizations to identify their needs by 2028, upgrade by 2031, and complete it by 2035. This is very similar to guidance issued by NIST in the US a few months ago, and the legislation that's making its way

through the regulatory system. NIST is the National Institute for Standards and Technology. It's a non-regulatory agency within Department of Commerce that was founded in 1901 as the National Bureau of Standards and changed names in 1988. NIST has been driving the agenda for post-quantum cryptography with a focus on identifying new encryption algorithms, phasing out weaker algorithms, and defining standards for key encapsulation mechanisms, KEMs, and digital

signature algorithms, DSAs. You hear the word FIPS a lot in government IT, and that stands for Federal Information Processing Standards. NIST publishes those too. Somewhat related to this, D-Wave announced a quantum blockchain framework that it describes as, quote, proof of quantum algorithm that uses quantum computation to generate and validate blockchain hashes. There's another company called Quantum Blockchain Technologies in the UK that provides proprietary mining hardware and

software also using quantum inspired algorithms. Storage is rapidly becoming a fierce battleground. Progress in technology is making room for new products and the sheer volume and velocity of data demands new ways to get speed at reasonable costs. One recent walk away is the need for hard disk drives to carry on and all SSD storage system is looking too expensive and even unnecessary. And hard disk drives, HDDs, are coming to the rescue, and they aren't going away anytime soon. Even tape drives are still in the data center picture, albeit not part of high-end production systems

like AI applications. Fun fact, global tape storage market was estimated at about $7 billion in 2024 and growing at about 8% per year. HDDs are electromechanical platters, spinning anywhere from 7200 to 15,000 revolutions per minute. Probably one of the most complicated technologies anywhere. While SSDs, solid-state drives, are all electronics and use flash memory. So the whole software and control and read-write transfer models are different. So a combination of HDDs and SSDs seems to be the way to go,

but they have different protocols and as you configure many of them and install various software, you have to avoid incompatible or conflicting policies, protocols, and practices. protocols and practices. Notably, modern SSDs rely on NVMe, non-volatile memory express protocol, but HDDs don't. NVMe was designed for SSDs to connect via the PCIe bus. NVMe is faster and lower latency than older storage protocols like SAS for serial attached SCSI, which is faster but more expensive than the even older and fading but more flexible SCSI, which is faster but more expensive than the even older and fading but more flexible SCSI, small computer system interface, which itself is faster than SATA, Serial Advanced Technology Attachment, which is good and

cheap and common and the way to go if it fits the needs. They're all liable to show up on the large menu of how storage devices connect to a computer. Seagate is working on HDDs with NVMe support, which can make it better, faster, cheaper. On the storage side, there will be a DPU for data processing unit that connects to the server via NVMe over Fabric and connects to a PCIe storage on the other side that includes HDDs and SSDs as cache. The DPU consolidates the work of and eliminates traditional controllers and host-based adapters, HPAs. DPUs promised such applications from their initial concepts, so it is interesting to watch as it looks more real.

Alright, that's it for this episode. Thank you all for being with us. HPC News Bytes is a production of OrionX in association with Inside HPC. Shaheen Khan and Doug Black host the show. Every episode is featured on insidehpc.com and posted on orionx.net. Thank you for listening.