Pat Gelsinger on LinkedIn: 5 years ago, the world of computing was a boring. Add some cores, speed up… | 61 comments
Summary
Add some cores, speed up IO, increment memory performance ... then, the AI/LLM explosiong and the world is anything but boring today. With the advent of reasoning models, which require memory-bound generation of thousands of output tokens, the limitations of existing hardware roadmaps has compounded. Im pleased to share that Ive recently invested in Fractile, a UK founded AI hardware company who are pursuing a path thats radical enough to offer such a leap. Their in-memory compute approach to inference acceleration jointly tackles the two bottlenecks to scaling inference, overcoming both the memory bottleneck that holds back todays GPUs, while decimating power consumption, the single biggest physical constraint we face over the next decade in scaling up data center capacity. In the global race to build leading AI models, the role of inference performance is still under-appreciated.