Modern AI systems require dynamic, adaptable software layers built to handle changing computation needs.
I had the pleasure of hosting renowned computer architect and Tenstorrent CEO Jim Keller, on the latest episode of Baya Systems’ Tech Threads podcast. If you haven’t already, listen to get his insights on the need for “open” intelligence architectures and what would be needed to drive the semiconductor industry forward.
What is an “open” intelligent architecture and ecosystem? As the proverb goes, “If you want to go fast, go alone. If you want to go far, go together.” In other words, open architecture helps every player work both individually and collaboratively for (in the long term) more comprehensive innovations, even though initially it might be slower. Having said that, open ecosystems in numerous markets over the last two decades have shown that the pace improves much faster once they reach critical mass. Our conversation peeled this onion back layer by layer.
We began our conversation with a theme that sits at the heart of modern compute innovation: simplicity through modularity. Referencing “The Systems Bible,” Jim reminds us that “you can’t design complicated things” and of the corollary, “You can’t fix broken, complicated systems.”
Instead, true progress in computing comes from layering and modularity: “By making everything have the right abstraction layers and modularity, we can actually build really complicated things out of simpler components.”
This principle defines the chiplet revolution. By deconstructing large, monolithic chips into smaller, specialized components, engineers can overcome the limits of traditional single die designs in favor of:
From there, our discussion turned naturally to how CPUs and AI processors are learning to collaborate. Tenstorrent’s work on the Open Chiplet Atlas architecture, built around RISC-V, embodies this idea.
“We picked RISC-V so we can build our own CPU and connect it with AI in a really flexible, novel way,” Jim explained. “The way that the processor integrates with the AI is novel, and I think it’s really going to unlock a lot of creativity on how to build future software stacks.”
Having been through the processor world myself for the last 3 decades, both at AMD and then at Arm, I see the wisdom of having an extensible CPU architecture. While extensible architectures can be a double-edged sword, as they sometimes go counter to maintaining standards, they make perfect sense for innovation in AI. Meaningful AI systems require not just powerful hardware but also optimized software stacks built around well-defined interfaces. That same philosophy drives Baya’s work on scalable interconnect fabrics and performance modeling to ensure heterogeneous systems are cohesive from the start.
The semiconductor industry is in the middle of a major rethink to meet today’s demands, largely driven by AI. How do we open the industry’s mindset, as well as our architecture? Jim has long been an advocate for open design and community-driven innovation, and at Tenstorrent, he’s continuing to prove the benefits of a collaborative ecosystem.
Going back to Jim’s comment about abstraction layers, this same philosophy is what he cites as Tenstorrent’s genesis for developing their Open Chiplet Atlas architecture. Named after the mythological Titan holding up the world, Atlas is a comprehensive ecosystem for chiplet design, providing:
One of Jim’s strongest convictions is how open ecosystems like RISC-V and Atlas are what create real progress. In his words, “Committees are good for standards that stabilize, not for speed. You can’t meet once a quarter to solve a thousand problems.”
While not entirely the same, we share a similar mindset at Baya: true innovation happens when ideas, data, and designs are shared, and there is an agile framework for innovation, design and deployment. Our view is that system architecture and design can and must be simplified to enable the scale that innovations in AI and data movement will soon require. The future belongs to companies that can collaborate quickly, iterate often, and treat openness as a competitive advantage.
This industry transformation not only requires us to rethink our hardware approach but also to adopt more adaptable software solutions for modern AI systems’ shifting computational needs.
It’s a concept that resonates deeply with us at Baya. Whether analyzing cache performance or designing interconnect fabrics, we’ve seen how much speed and efficiency come from making invisible system-level dynamics visible to the architect.
Modern AI systems require dynamic, adaptable software layers built to handle changing computation needs. By simplifying software layers, developers can build and deploy AI models without getting caught up in the complexity of the underlying hardware. Simplified software layers make it easier to accommodate new demands—and can even reduce the complexity of the hardware needed. That’s exactly what Baya’s WeaverPro platform enables: simplifying the software layer to help teams build, optimize and scale AI systems efficiently across multi-vendor hardware architectures.
As Jim put it, in his own hard-hitting way, “Humans aren’t getting smarter. We can’t solve really big, complicated problems. But we’re pretty good at breaking them down.”
The discussion reinforces some of the key philosophies that drive both Tenstorrent and Baya: simplify complexity, enable openness, accelerate innovation and amplify scale.
Solving the future of compute is complex, and Baya is helping architects bridge the gap between concept and silicon, allowing the architectural exploration of the next implementation with our solutions that enable rapid system architecture exploration with our WeaverPro platform and fabric development with WeaveIP, because the future of computing is modular, open, and collaborative.
Closing out our chat, Jim and I spoke on a final, yet crucial, detail: the next generation of engineers. His advice for the industry’s up-and-coming?
“Work on open technologies, focus on speed, and get hands-on experience in smaller companies where you can truly make an impact.” The future of the industry lies in artificial intelligence and innovation, and we’re seeing incredible speed in both areas today. As Jim points out, new architects starting their careers today need to be focused on writing code and doing commits daily, and most importantly, hone their skills in AI.
To hear more of Jim’s thoughts on how the next generation of engineers can set themselves apart and catch our discussion on data movement and democratization, listen to Episode 7 of Tech Threads: The Architecture of ‘Open’ Intelligence on the Baya website, or wherever you get your podcasts.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply