Feed - Buzzberg

02:30
Jun 03

SemiAnalysis

Why next-gen AI scale-up needs CPO

22:15
May 27

SemiAnalysis

How Makora Generates CUDA Kernels That Beat Hand-Tuned Code | Researcher Conversations at GTC

19:00
May 18

SemiAnalysis

Why a former oven factory runs Claude Code

SMH 1ST ▾

Dylan Patel

Semiconductor stocks only go up.

Compute for AI is still a limited resource, driving sustained demand for semiconductor stocks. AI models are increasingly restrictive with tokens and downgrading to save compute, and the latest frontier models require massive compute. This backdrop supports continued upside for semiconductor stocks.

SMH LONG

MED

00:01
May 14

SemiAnalysis

Analog's Mechatronics Engineer: Hardware Powering AI Inference | Researcher Conversations at GTC

19:30
Apr 30

SemiAnalysis

AWS Trainium: How Amazon Built Their Own AI Chips | Researcher Conversations at GTC

17:01
Apr 24

SemiAnalysis

Why 99% Can't Use Semiconductor Data

21:00
Apr 19

SemiAnalysis

Memory Cycle Timing Is Everything

19:58
Apr 16

SemiAnalysis

Why Positron AI is Choosing LPDDR over HBM for Next-Gen LLM | Researcher Conversations at GTC

HBM 1ST NVDA 1ST SOXX 1ST ▾

Thomas Sohmers

Commodity LPDDR and substrates over HBM and advanced packaging.

Positron AI uses LPDDR memory and organic substrates instead of HBM and advanced packaging, avoiding supply chain constraints and achieving higher memory capacity for scalable AI inference with commodity technologies.

HBM AVOID SOXX LONG

Thomas Sohmers

NVIDIA has poor matrix-vector performance for inference.

NVIDIA's GPUs have a worsening ratio of matrix-matrix to matrix-vector performance from Hopper to Blackwell, making them inefficient for AI inference workloads that rely heavily on matrix-vector operations, which Positron's architecture solves.

NVDA AVOID

HIGH

21:30
Apr 10

SemiAnalysis

Claude Code Psychosis: How SemiAnalysis Is Token Mogging Meta | Ep. 008

Feed — Buzzberg