A100

⚡️Accelerators @ 3x NVIDIA H200 perf, Made in the USA - Thomas Sohmers + Mitesh Agrawal, Positron AI

Aug 18, 2025 · 47:56

Thomas Sohmers and Mitesh Agrawal of Positron AI argue memory bandwidth, not compute, is the true bottleneck in AI inference, and their accelerator achieves 93% memory bandwidth utilization—triple NVIDIA's efficiency—enabling 70% faster token generation at 150W. The founders, both Lambda Labs veterans, shipped an FPGA product 15 months after founding, then raised a $51M Series A for an ASIC in late 2026. Their hardware requires no recompilation: it ingests raw binary weights from NVIDIA training and outputs an OpenAI-compatible API. Positron focuses on the decode phase of transformers, where memory-bound matrix-vector multiplication dominates, and already counts Cloudflare and Parasail as customers. The company sells systems directly, prioritizing capital efficiency and ROIC over operating its own cloud.

The Shape of Compute (Chris Lattner of Modular)

Jun 13, 2025 · 1:18:18

Chris Lattner of Modular explains how his company is breaking the CUDA monopoly with Mojo and MAX, matching NVIDIA's best inference performance on AMD and NVIDIA GPUs. After three years in R&D, Modular open-sourced its stack and now delivers state-of-the-art Llama 3 serving at over 800 tokens per second. Mojo, a Python-family language, runs kernels faster than Rust and can extend Python without bindings, while MAX provides a full inference framework with automatic kernel fusion and cluster management. Unlike VLLM or SGLang, Modular's container is just a gigabyte, fully open source, and not reliant on proprietary CUDA blobs. DeepSeek's low-level PTX work validated the approach, but Lattner emphasizes that Mojo's portability avoids rewriting kernels for each new GPU architecture. He also details his daily routine, using Cursor for coding, and hiring "elite nerds" to grow the team.

SF Compute: Commoditizing Compute

Apr 11, 2025 · 1:12:02

Evan Conrad, co-founder of SF Compute, argues that GPUs behave like a real estate business, not a traditional cloud, because price-sensitive customers value every incremental GPU and will switch for a 10% margin. CoreWeave succeeded by selling locked-in long-term contracts to low-credit-risk customers like Microsoft and OpenAI, ignoring short-term demand. He predicts hyperscalers and providers like Together and DigitalOcean will lose money on GPU clusters because software margins cannot match the hardware costs. SF Compute started as an AI lab forced to sublease its cluster monthly to avoid bankruptcy, then evolved into a market where anyone can buy H100s by the hour via dynamic pricing—often below $1/hour for short bursts. Utilization stays near 100% as prices adjust. Future plans include cash-settled futures to reduce financial risk across the industry, while the brand deliberately stays anti-hype and calm.

Beating Google at Search with Neural PageRank and $5M of H200s — with Will Bryk of Exa.ai

Jan 10, 2025 · 55:54

Will Bryk, CEO of Exa.ai (formerly Metaphor), details building a neural search engine from scratch using link prediction as 'Neural PageRank' — predicting documents rather than keywords. Exa's new product offers near-perfect lists (e.g., 'startups working on hardware in SF') by scaling compute per query, from milliseconds to a day, like o1 for search. Bryk argues LLMs will become the interface to search, while Exa provides the 'super knowledge' that even AGI will need. He contrasts Exa's neural approach with Google's keyword-based system and Perplexity's reliance on Bing. The company recently purchased a $5M H200 cluster and maintains a culture of nap pods and first-principles thinking.

[Paper Club] BERT: Bidirectional Encoder Representations from Transformers

Nov 27, 2024 · 53:12

Eric Ness walks through the BERT paper, explaining how its bidirectional encoder architecture with masked language modeling and next sentence prediction pre-training enabled state-of-the-art results on 11 NLP tasks in 2019. He details the 110M parameter BASE and 340M parameter LARGE models, trained on 2.5B Wikipedia words and 800M book words, and how a simple logistic regression on BERT embeddings achieves 82% accuracy on IMDb sentiment classification, far above the 50% baseline. Swyx and Eric discuss how BERT’s pre-training objectives—masking 15% of tokens and predicting sentence order—teach small models word relationships for efficient classification, though they lose data compared to decoder-only next-token prediction. They note that while scaling laws favor decoder models, BERT-style encoders remain cost-effective for edge deployment, with recent work showing full retraining in 24 hours for under $500 or even 1 hour for $20 on 8 A100s, matching original GLUE scores.

Segment Anything 2: Memory + Vision = Object Permanence — with Nikhila Ravi and Joseph Nelson

Aug 7, 2024 · 1:00:44

Nikhila Ravi from FAIR and Joseph Nelson from Roboflow discuss Segment Anything 2 (SAM 2), which extends zero-shot object segmentation to video with memory and real-time interactive tracking. The model is one-third the size of SAM 1 (224M vs 630M parameters) and six times faster, using a novel memory attention mechanism with six-frame spatial memory and longer-term object pointers. Ravi explains the three-phase data engine that built the SA-V dataset of 51,000 videos, enabling SAM 2 to track arbitrary objects like a T-shirt or an octopus even when occluded. At Roboflow, users labeled 49 million images with SAM in its first year, saving an estimated 35 years of manual annotation time. The demo features swim lanes that show object visibility and allow refinement clicks to correct tracking mistakes, a key improvement over prior video segmentation models. SAM 2 also handles out-of-distribution domains like underwater footage, though screenshots remain challenging and may require fine-tuning.

Why Google failed to make GPT-3 -- with David Luan of Adept

Mar 27, 2024 · 49:27

David Luan, co-founder of Adept and former early OpenAI leader, explains why Google failed to build GPT-3—its 'brain credit marketplace' prevented critical mass—and why Adept builds enterprise AI agents prioritizing reliability over generality. He recounts the GPT-2 demo that helped secure Microsoft's $1B investment and details Adept's goal: an AI teammate that can do anything a human does on a computer, targeting 'nines of reliability' for workflows like dispatching a physical truck. Luan contrasts Adept's vertical integration—training fast multimodal models (Fuyu) for charts and UIs—with pure-play foundation model companies that sell tokens, predicting commoditization. He notes Adept is sold out for Q1 and raised $420M, and explains how an augmentation focus creates a data flywheel from human oversight.

Building an open AI company - with Ce and Vipul of Together AI

Feb 8, 2024 · 1:15:18

Together AI co-founders Vipul Ved Prakash and Ce Zhang explain why openness is core to their mission, detailing their journey from Apple (Vipul) and Stanford research (Ce) to building an open AI platform. They discuss RedPajama’s evolution into a modular dataset with 40 quality signals, the need for 5,000 tokens/second inference speed, and their investment in state space models like Mamba and Hyena as alternatives to transformers. The company, which runs 7,000-8,000 GPUs (mostly H100s), sees training as a larger workload than inference, with fine-tuning driving top models. They advocate for independent inference benchmarks, publish open research like FlashAttention, and keep some software proprietary. With 38 employees and 45% researchers, they are hiring across the stack, from CUDA to DevOps.

The State of Silicon and the GPU Poors - with Dylan Patel of SemiAnalysis

Dec 5, 2023 · 1:07:55

Dylan Patel of SemiAnalysis argues the 'GPU poor vs GPU rich' divide defines AI, with Google's TPU ramp (millions of units) dwarfing others and Nvidia selling over a million H100s this year. He explains that training costs are irrelevant compared to inference costs, and that memory bandwidth is the key bottleneck for LLM inference (e.g., Llama 70B needs ~2.1 TB/s for human reading speed). Patel advises GPU-poor players to focus on on-device innovations like speculative decoding (Medusa), asynchronous training, or fine-tuning for edge use—not on gaming benchmarks or fine-tuning small models. He details why AI hardware startups (Cerebras, Graphcore) bet wrong on on-chip SRAM, while Google’s TPU/Broadcom partnership and Nvidia’s yearly cadence make alternatives tough. He also warns that rebuilding the semiconductor supply chain in the US is unfeasible due to its fragmentation, and that safety through obscurity doesn't work—open innovation is better.

RWKV: Reinventing RNNs for the Transformer Era

Aug 31, 2023 · 1:56:51

Eugene, CTO of UILicious, introduces RWKV, a recurrent neural network that matches transformer-level performance while scaling linearly with context length, solving the quadratic cost of attention. He explains how RWKV uses Attention-Free Transformer layers and parallelizable training, achieving competitive results on 7B and 14B models. The community-driven project prioritizes multilingual support, with a tokenizer that handles languages without spaces. Eugene recounts his journey from GPU.js to finding RWKV, and highlights use cases like long HTML analysis where transformers fail. He also discusses the token crisis, diffusion models for text, and how the AI waifu community drives optimization and alignment research. Finally, he advises AI engineers to focus on practical prompting and data curation, noting that deep architecture knowledge is optional.

FlashAttention-2: Making Transformers 800% faster AND exact

Aug 3, 2023 · 1:04:06

Tri Dao, creator of FlashAttention and FlashAttention-2, explains how his I/O-aware algorithm makes attention 2x faster by fusing kernels and using online softmax, achieving near-matrix-multiply efficiency. He argues that transformer alternatives like state space models and RNNs (e.g., RWKV) could surpass transformers for long sequences and high-throughput generation, though attention still dominates. Dao discusses the hardware lottery, where NVIDIA's CUDA ecosystem entrenches transformers, and advocates for open-source AI, praising Meta's Llama 2 for shifting enterprise adoption despite its restrictive license. He emphasizes that understanding both algorithms and systems is key to scaling AI, and that academia should pursue risky bets that industry cannot.