Hot Chips 2025 is less than three months away, and so we now have the preliminary schedule. There are new CPUs, new networking talks, lots of AI chips, and a lot more.
Hot Chips 2025 Preliminary Schedule Released
Monday kicks off with various CPU sessions. Hopefully Intel’s talk is Clearwater Forest (since that is after Sierra Forest) and we also get a next-gen IBM POWER talk. I have to say I am quite excited about the IBM POWER side after I saw the response to the IBM z17 articles and video we did in April for its launch. We then get a security session. Google has the day 1 keynote. After that we get a graphics session where interestingly Meta is talking about AR SoCs. On the networking section, I cannot wait for all of it. Intel’s next-gen IPU, Mount Morgan, is getting detailed, and we get big presentations from AMD, NVIDIA, and Broadcom as well.
Monday Aug 25 |
|
| Title | Affiliation |
| CPU Session | |
| Cuzco: A High-Performance RISC-V RVA23 Compatible CPU IP | Condor Computing |
| PEZY-SC4s: The Fourth Generation MIMD Many-core Processor with High Energy Efficiency and Flexibility for HPC and AI Applications | PEZY Computing |
| IBM’s Next Generation Power Microprocessor | IBM |
| Introducing the Next Generation Intel® Xeon® Processor with Efficiency Cores | Intel |
| Security Session | |
| Presto: A RISC-V-Compatible SoC for Unified Multi-Scheme FHE Acceleration over Module Lattice | Tsinghua University |
| Azure Secure Hardware Architecture: Establishing a Robust Security Foundation for Cloud Workloads | Microsoft |
| Keynote: Predictions for the Next Phase of AI | |
| Graphics Session | |
| AMD RDNA 4 and Radeon RX 9000 Series GPU | AMD |
| RTX 5090: Designed for the Age of Neural Rendering | NVIDIA |
| Specialized SoC enabling low-power ‘World Lock Rendering’ in Augmented and Mixed Reality Devices | Meta |
| Networking Session | |
| Intel Mount Morgan Infrastructure Processing Unit (IPU) | Intel |
| AMD Pensando™ Pollara 400 AI NIC Architecture and Application | AMD |
| NVIDIA ConnectX-8 SuperNIC: A Programmable RoCE Architecture for AI Data Centers | NVIDIA |
| Tomahawk-F1: A High-Bandwidth, Low Latency Ethernet Switch Chip for High-Performance Computing | Broadcom |
After a few quick hours of sleep, and a coffee, it will be time for day two. Kicking off is the optical session where it looks like AI fabrics and chip applications are going to be the focus. Fabric8Labs is kicking off the Power/ Methodology Session. Rapidus is doing the keynote. Machine Learning 1 has sessions from Marvell and d-Matrix that appear to focus on memory subsystems. Then, the Hot Chips folks made the wise decision to have the AI chips at the end. We get a look at the NVIDIA GB10 which I am extremely excited about, then the AMD CDNA 4/ MI350, followed by Google Ironwood.
Tuesday Aug 26 |
|
| Title | Affiliation |
| Optical Session | |
| Celestial AI Photonic Fabric Module (PF Module) – The world’s first SoC with in-die Optical IO | Celestial AI |
| A UCIe Optical I/O Retimer Chiplet for AI Scale-up Fabrics | Ayar Labs |
| Passage M1000: 3D photonic interposer for AI | Lightmatter |
| Co-Packaged Silicon Photonics Switches for Gigawatt AI Factories | NVIDIA |
| Power / Methodology Session | |
| ECAM Enabled Advanced Thermal Management Solutions for the AI Data Center | Fabric8Labs |
| Everactive Self-Powered SoC with Energy Harvesting, Wakeup Receiver, and Energy-Aware Subsystem | Everactive |
| Taping Out Three Class Chips per Semester in Intel 16 Technology | UC Berkeley |
| Keynote | Rapidus |
| Machine Learning 1 Session | |
| Memory: (Almost) the Only Thing That Matters | Marvell |
| UB-Mesh: Huawei’s Next-Gen AI SuperComputer with A Unified-Bus Interconnect and nD-FullMesh Architecture | Huawei |
| Corsair – An In-memory Computing Chiplet Architecture for Inference-time Compute Acceleration | d-Matrix |
| Machine Learning 2 Session | |
| NVIDIA’s GB10 SoC: AI Supercomputer On Your Desk | NVIDIA |
| 4th Gen AMD CDNA™ Generative AI Architecture Powering AMD Instinct™ MI350 Series Accelerators and Platforms | AMD |
| Ironwood : Delivering best in class perf, perf/TCO and perf/Watt for reasoning model training and serving | |
Usually it is quite warm in the August afternoons but those afternoon sessions should be exciting.
Final Words
Hot Chips is easily one of the nerdiest and most fun conferences I get to attend every year. If you have been a longtime STH reader, you will notice that we have been covering the event for many years and will continue to do so this year.
“Hopefully Intel’s talk is Clearwater Forest (since that is after Sierra Forest) …”
Yes, next-gen Xeon with e-cores will be Clearwater Forest
Finally, maybe we’ll see something on POWER11. That will be neat to see. I would be interested to see how much bigger those CPU packages will be compared to the current POWER10 ones. Ever bigger and ever more pins, I suspect. Rent’s Rule is very much alive and well.