Recap of Super Computing 24

Back to Atlanta, GA, for this new edition of Super Computing, hat has been organized in a big expo and sport zone. The conference took place in the area of the Centennial Olympic park and the main sport stadiums. Impressive by its location, the event was also a very good edition by a few numbers: 18,104 attendees which is more than 4,000 in comparison with the 2023 edition and 494 exhibitors from 29 countries.

The conference is now centered on High Performance Computing, Networking, Storage, and Analysis and AI as AI eats everything present on every lips, present in every vendors’ message and asked all day by end-users. We live in a genuine tsunami that relegates other topics to the background. Imagine the opposite, a company without any mention of AI, it would be immediately ignored. Now for sure for end-users it’s a real nightmare, a bit like a search for a needle in a haystack. But the real reason is really based on the fact that AI and HPC shared some challenges, foundations and solutions’ approaches. For storage, it makes sense for HPC storage vendors to extend, adapt and design new solutions tailored to AI workloads all over the data process chain.

At such an event, with a good mix of exhibitors, we found vendors, public agencies and organizations, universities, data centers services or end-users, and thus open source in addition to commercial solutions are actively represented. We were surprised to see some booths even if it’s a large independent industry conference that appears to be a must go show.

The show is also the place where each year the Top500 is unveiled and this year marks El Capitan, the super computer at Lawrence Livermore National Laboratory is #1, followed by Frontier and Aurora, the 2 followers in that order. These 3 systems are Cray machines powered by AMD EPYC processors and the 3rd by Intel Xeon processors. The top 4 systems are based and controlled by the US, the 5th and 9th from Italy, the 6th is Japanese, the 7th from Switzerland, the 8th Finland, and the 10th from the US again. So clearly the US dominates this American ranking.

The companion list, the IO500, has been updated as well and 4 systems use DAOS, 4 rely on DDN ExaScaler, 1 on Lustre and 1 on Weka. And the rest of the list confirms the domination of parallel file systems, both from commercial or open source flavor.

AI was everywhere and all players articulated their message with AI/ML in mind for different phases. Therefore flash media and SSDs seem to be the preferred choice for obvious performance reasons snd many of them had booths like Micron, Phison, Samsung, Seagate, SK Hynix, Swissbit or WDC…. plus of course systems, platform, storage subsystem or divers arrays.

As an anecdote, visiting the Seagate booth, we realized that the logo of the company has been changed with different proportions, any other moves behind…? On the other side of the street, we didn’t get any more news regarding the WDC split.

S3 and file storage were also very present in different flavors with Cloudian, DataCore, DDN with its Infinia product, Dell, Hammerspace with their Hyperscale NAS and Tier 0 model, Hitachi Vantara and the recent partnership with Hammerspace in addition to the WEKA one, IBM, the return of NetApp, Pure Storage and their recent GPUDirect addition, Quantum, Qumulo, Vast Data and Weka or small players like MooseFS or Tuxera among others. Many of them announced recently or even during the show some Nvidia GPU support extensions like GPUDirect. The only strong player absent for geo-political reasons was Huawei. The parallel aspect of object and file storage have been clearly adopted, pretty easy for HPC actors having this kind of approach in their parallel file system model with some open source Lustre roots, other similar data service layers, pNFS, extensions to NFS or commercial parallel animals.

The new Panasas team tried to change the company image and business trajectory with a new name, Vdura, pushed by a new management team with several heads coming from Seagate and a special attraction at their booth.

MinIO introduced, during the KubeCon conference the week before, AIStor and should join the pack in SC25.

Same thing for unstructured data management with real pressure as data volumes grows very fast, we met Arcitecta, Atempo, iRODS, Komprise, PoINT Software and Systems, QStar, Starfish or Versity even if some of them didn’t have booth but they found a way to be seen.

Graid Technology continues to grow their partnership and ecosystem solving a real problem from small config to very large one, levering GPU to run faster data protection algorithms. Viking Enterprise Solutions, a division of Sanmina, exposed multiple of its solutions and we recognized many systems used by other exhibitors. They also displayed the Kove shared memory approach with the Kove:SDM Memory Tower.

A CXL pavilion with 16 pods pushed CXL and all its developments and applications. Among them, ZeroPoint Technologies and Micron. The latter presented a Fabric Attached Memory File System aka FAM-FS to give perspective to CXL.

Click to enlarge

As HPC and AI are essentially compute activities, the vast majority of players exposed their CPU, GPU, DPU… architecture, designs and integrations. We count NextSilicon, Intel, AMD, of course Nvidia but also specific vertical players like Groq, Cerabras and SambaNova. And outside of our scope, a few quantum computing instances with IBM, IonQ, Quandela or Quantinuum.

It also means that energy was central with many booths promoting new energy efficiency and cooling systems, a serious challenge for new intense computing environments.

Some tape library vendors exhibited as well like HPE, IBM, Quantum and Spectra Logic. We even met BDT and we saw an IBM Diamondback at the ATTO booth.

We also attended several side events where we had the opportunity to find non official SC24 exhibitors, such as Cerabyte, Grau Data, Leil Storage, OSNexus or Versity for instance. We noticed also Nexsan, QStar, Scaleflux… present at partners booths.

Many companies have issued some press releases during the show, among them we select:

• Arcitecta for their immersive demonstration during the show,
• ATTO for the Ethernet-to-SAS bridge integration with IBM’s Diamondback tape library,
• Bacula with the new 18.2 release,
• Cloudian for the GPUDirect acceleration support,
• DDN with the new iteration of AI400X2 and AI200X2 appliances,
• Hammerspace with their Tier 0 approach,
• Komprise with their HPC survey,
• MaxLinear partnering with QCT to embed Panther III Storage Accelerator,
• Phison demonstrating their Pascari product line and the leading 128TB model,
• Pliops boosting LLM Inference with their XDP LightningAI model,
• PoINT Software & Systems with a new flavor of S3-to-Tape solution,
• Qumulo with Cloud Native Qumulo on AWS performance passing 1TB/s and 1M IOPS,
• Viking Enterprise Solutions with the Kove:SDM Memory Tower,
• WEKA with the preview of Nvidia Grace CPU Superchips integration.

SC25 is scheduled November 16-21, 2025, in Saint Louis, MO.