5 years ago I started the game bottom of the barrel Iron 1. 5000 hours later I hit Radiant. Thank you @CHUNK_VAL for being the goated duo.
Now vs then.
I chatted with @ysmulki about MatX, chip design and where silicon designed for LLMs is headed
(8:17) Tightly coupling SRAM and HBM on one chip
(14:03) More MoE FLOPS, smaller KV cache load
(16:08) Numerics: from 32-bit to 4-bit
(19:02) Targeting both training and inference
(22:14) Chip timelines
(27:15) Logic and memory scarcity
(29:42) Compute costs
(32:07) Latency: from 20ms to 1ms as the new table stakes
(40:50) Programming the chip
(43:00) Starting MatX
(47:11) Codesign without seeing the models
(51:57) Interconnect design
(55:44) Performance modeling philosophy
(1:07:02) Prefill vs. decode
(1:13:47) What's next
Anyone who had a ticket for any of this weekend’s playoff matches. Tarik and I are hosting a watch party at Sen HQ. Just bring your Riot ticket. Doors open at 1. Free Food free Red Bull!
Ethan doesn't deserve any hate for what happened and what hes saying is true, on breeze we found out the round wasn't live because we killed sova/kayo afk and we weren't told anything, we weren't told not to move so we played the round out expecting the roll back. Ethan moving around or standing afk not doing anything wouldn't have changed the outcome of the game so we can all move on from that. There needs to be better communication from the admins and clarification on future scenarios so this can be prevented in the future