Cerebras founder and chief architect Michael James walks through the CS-3 system and its wafer-scale engine, a single 300 mm die integrating around a million AI-optimized compute cores on one piece of silicon. Built in 5 nm with roughly 4 trillion transistors, WSE-3 delivers on-chip memory, interconnect and compute in one monolithic device, targeting high-throughput AI inference and data-intensive HPC workloads in a compact rack-scale node. https://www.cerebras.ai/chip
—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—
He explains the extreme power-delivery and packaging needed to run this chip at roughly 25 kW: front-side AC/DC modules, 3D power distribution and dense arrays of regulators positioned close to the wafer to manage around 30,000 amps of current. Because all compute and 44 GB of SRAM sit on a single wafer, the system minimizes off-chip traffic and uses control logic that smooths power ramps with dummy operations when workloads switch off, avoiding destructive current spikes while preserving energy efficiency.
On the architecture side, James describes the WSE-3 as a proprietary dataflow processor designed for strong scaling. Loop induction variables, data movement and network behavior are encoded directly into the instruction set, so a single matrix operation can be spread spatially across the full grid of cores with minimal software overhead. That allows Cerebras to map full transformer layers over the wafer and reach very high inference throughput, with customers reporting large speedups over Nvidia GPU clusters on latency-sensitive language-model serving.
The discussion then shifts to real workloads, including a global shallow-water-equation simulation of an asteroid impact off California, run at about 200 m resolution over the entire planet. By exploiting the dense on-wafer memory and mesh interconnect, a cluster of CS-3 nodes achieved exascale-class performance for this tsunami scenario at a fraction of the power draw of traditional exascale systems, while still supporting large language models such as Llama and DeepSeek on the same architecture.
Filmed at Supercomputing 2025 in St Louis, the interview also touches on manufacturing yield and roadmap. Cerebras overprovisions identical cores across the wafer and then uses automated defect mapping plus constraint solving to reroute communication around faulty regions, guaranteeing at least 900,000 working cores per device and turning the rest into pass-through fabric. James hints that future generations will continue this wafer-scale path, pushing AI inference and physics-based HPC further by co-designing architecture, packaging and dataflow software as a single system.
I’m publishing about 60+ videos Supercomputing 2025 #SC25 I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Supercomputing 2025 SC25 videos in my playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvihnaq98TO55Cbe2VMD9mk8
This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK
Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁
Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY
