I think it does! RL needs the CPU to run the RL environment. Larrabee GPGPU allows this to scale efficiently.
Thanks Pat!
17 replies (most recent on top)
Laughable. 400Watt chip!
Larrabee? Can I just tell you how much overtime and sweat equity went into trying to duct-tape that P.O.S. Into a viable product. Everyone knew it was cr-p except Pat.
DeepSeek was trained & developed on AMD GPUs only. Sorry Intel and NDVIA
AI was, is, and will continue to be a hardware/software codesign issue. Intel learned it too late - or has not learned it. Worse yet, it is trying to solve it with the weakest AI/GPU/SW team in the industry.
Bah hah hah ha. No. Deep Seek has 50,000 H100 cards from nVidia.
Larrabee Ultra. Intel's 2025 flagship. lol
Larrabee was another product ki-led by Finance. The history books are going to write about how finance ki-led Intel innovation, risk taking and culture. There will be a dose of HR taking the hit also.
I worked on Larrabee. The hardware wasn't good and the software was worse. It was a money pit. What's your point? You want vindication that we had an Idea along time ago that would fit this new trend and muffed it?
Nope. We want you to but Xenon processors AND the Gaudi accelerators then try and achieve A.I. Whoops News out of china: You don't really need any of that. No vindication for a company with it's head in the sand.
Larrabee is the architecture, Xeon Phi was the product which was produced until 2020.
You can't keep a good Pat down.
How do you vindicate a product that never saw the light of day?
The game on learning was lost. Now with DeepSeek, new opportunities arise from running inference locally.
When it had the muscle, Intel should have push against running everything in the cloud.
68 cores running the RL on the same loaded model. Quite impressive! Sadly Intel has yet again dropped the ball while going for a layup.
What does reinforcement learning have more to do with Intel than AMD or any other CPU vendor?
Reinforcement Learning
What is RL?