Eric Quinnell (@divBy_zero) profile picture

Founding Chip Architect, @river_ai_inc Fmr AWS Tranium, Tesla Dojo, CPUs (x86+ARM). PhD Computer Arithmetic

Joined December 2022
Photo by Eric Quinnell (@divBy_zero)Photo by Eric Quinnell (@divBy_zero)Photo by Eric Quinnell (@divBy_zero)Photo by Eric Quinnell (@divBy_zero)Photo by Eric Quinnell (@divBy_zero)Photo by Eric Quinnell (@divBy_zero)
NOP is the most important assembly instruction
5
4
1
22
5,055
FAA ARP flood perhaps? Layer2 only internal network? L3/IP would have fenced something only blasting 1000 transactions, yes? Everyone get their DNS haikus ready. Even $50bn can’t fix DNS
2
11
960
Thrilled to have @FelixCLC_ join @river_ai_inc . Amazing projects bring in amazing talents
Hi friends! Super happy to share i've joined the incredible @divBy_zero, @aaronsrogers and @ibab at @river_ai_inc! Vision is simple: Intelligence that belongs to you.
2
3
51
2,165
Eric Quinnell retweeted
Introducing River Recipes: our open-source initiative for turning promising research into vetted, reproducible training recipes. Our first recipe tackles text-to-SQL, where our trained open models beat GPT-6 Astra Pro and Claude Opus 5.5 on Arcwise-Plat for less than 1% of the cost. Read the full breakdown and explore the recipe ↓ river.ai/achieving-frontier-…
10
24
6
207
42,860
Eric Quinnell retweeted
It's beneficial for a software engineer to learn chip design so they understand how numerics flow through circuitry each cycle.
17
30
4
459
881,721
mxfp8 x mxfp4 at higher perf than normal mxfp8 matmul is a magic bullet worth pointing out. Kudos to Richard Ho and the team, this is good engineering. That’s 2x activation fetches plus scale sidecar concurrently on the fp8 to match the fp4 native element sizes. Not trivial
Jalapeno beats VR200 in A0 silicon with a vibe-ported, non-specdec implementation of DeepSeek 🌶️ 🌶️ And with much faster program execution than VR200, and with a B0 update landing imminently Congrats to the Jalapeno team!!!
4
12
96
27,105
SW terms decoded for HW ppl: BigO = they need more compute EP = they’re out of RAM Ephemeral = the L1 is too small Sparse Attention = mythical systolic array that is faster with partial outputs Tokenizer = cpu scrabble Batch norm = previous trauma
5
8
49
2,792
Eric Quinnell retweeted
The River API was tested in this blog post and outperformed Tinker on reinforcement learning runs with identical training code. We spent a lot of effort to get details like routing replay right so you get the best possible results with the API.
We can now RL large MoEs with 0 train-infer mismatch! And doing so can improve performance (pictured task: teach Qwen3.6-35B-A3B to play Wordle). Everything is open-source and we did a bunch of ablations. 🧵
7
15
1
185
40,161
Eric Quinnell retweeted
Excited to share I've joined @river_ai_inc to work on personal AI! As AI products become pervasive in everyday life, it's clear this technology needs to be private & owned by the users. I'm looking forward to a future with great products and alternatives to centralization.
11
2
2
133
13,580
Eric Quinnell retweeted
Today, we're sharing that River AI has raised $1.1 billion, led by @generalcatalyst and @amppublic with strategic investment from @nvidia and @AMD. Additional investors include @ycombinator and @Temasek. We imagine a future where your AI works entirely for you and deeply aligns with your values. Compared to the corporate chatbots of today, it will feel completely different: responsible, curious, and truly yours. Most importantly, you'll own everything: the hardware it runs on, the data it learns from, and the intelligence itself. Our training API is already live and powering incredible work today. This new funding accelerates our push to bring the full personal AI stack to life. Read River AI CEO Igor Babuschkin’s (@ibab) conversation with the New York Times to learn more: nytimes.com/2026/08/11/techn…
79
74
68
868
789,564
Eric Quinnell retweeted
River API is now available to everyone. Fine-tune and run RL on frontier open models. Fast, cheap, and scalable. Go to console.river.ai/ to try it out!
4
19
7
205
115,515
Important IEEE-754 reminder: sqrt(-0) = -0
7
5
45
3,516
Pick a matmul shape you’re going to stick to for a while. I’ll wait.
2
1
1
8
2,361
Eric Quinnell retweeted
First you have to understand that modern LLM inference already disaggregates weights as models outgrew single chips years ago. You shard either by layer (pipeline parallelism) or by slicing every layer (tensor parallelism), and the two do very different things. As an example, let’s look at Llama 3.3. It has 70B of weights and at FP8 that’s 70 GB of memory which is enough to fit on a single H100. Now that H100 has 3.35 TB/s of HBM, so the fastest it can ever decode for one user is 70/3.35 ≈ 21 ms/token or ~48 tok/s while using under 1% of its FLOPs. Now if we pipeline it across 8 chips: each chip holds ~8.75 GB, which means it only needs 1/8th the bandwidth and 1/8th the FLOPs to sustain the same aggregate throughput. Now crucially the token/sec a user gets is limited by the amount of data that crosses the link. In current LLMs all that is a small amount of activations for LLama 3.3 it’s ~8 KB per token…. Yes, you read that right it’s 8 KILOBYTES we are sending over a <900 GB/s link. That’s only 9 ns of serialization time but the overhead of 224G PAM4 SerDes adds ~100 ns per link traversal with RS-FEC which is 11x longer than the payload itself. And then you have the NVSwitch adding ~300 ns per hop and you need to pay twice. That’s ~600 ns of just hardware latency wrapped around 9 ns of data making a 98% tax before software even shows up. Then NCCL’s collective stack turns 600 ns into 10-20+ us… all to move 8 kilobytes lol. For comparison 8 KB serializes over 10 Gigabit Ethernet NRZ, in just 6.6 us. Pipeline parallelism however doesn’t make a single user faster as the token still needs to visit every layer in the sequence, so per-user speed is still weights / per-chip bandwidth. To get more speed per user token you need to use tensor parallelism and have all the chips work on the same layer simultaneously. TP costs you 2 all reduce OPs per layer, 160 per token on llama 3, that’s still kilobytes of traffic but with NVLink overhead it’s a massive tax and why pipeline parallelism on most models still gives more interactivity per user. However, this gives you a huge latency lever to pull that scales tokens per second with interconnect speed instead of memory BW. The clever amongst you might have also realized that sharding doesn’t just cut memory bandwidth per chip it also cuts FLOPs per chip and is why we have such bad MFU on decode. So once you’ve sized the link for the memory, you need to size the compute for it too. This is called “balancing the pipeline”, and currently no shipping chip does it because they were all designed as standalone monsters. Remember Tokens/sec = ~aggregate memory BW / bytes touched per token. At batch 64 in FP4 you need ~250 FLOPs per byte, and Blackwell ships 1,250. Provisioned 5x more than the narrow pipe of HBM. Nobody saturates shit cause they are all building around HBM. So now it all comes full circle. Parallelism reduces memory bw pressure and thus FLOPs but increases interconnect latency pressure. Despite having HBM and GigaSERDES we aren’t actually doing more work lol. But if you really wanted to balance the pipeline you need to match the memory bandwidth, the flops, and most importantly the interconnect. So what does that look like ? Well if you build around LPDDR’s lower bandwidth, lower your interconnect latency, you actually can beat Nvidia on decode with a fraction of the silicon.
131
188
56
3,639
1,512,393
Eric Quinnell retweeted
Sign up for the River API if you want to try it out. We’re open ing up access incrementally.
We've received plenty of applications for the API waitlist and have opened up access to a number of enterprises and researchers already. We're improving the API based on your feedback, and we expect to open up to the public in the next few weeks.
6
3
118
25,017
Hiring all silicon disciplines for @river_ai_inc. Plus free cookies every tapeout
River AI is building personal AI owned & shaped by you. We are hiring exceptional talent across the stack: * Research * Software Engineering * Product Development * Data * Hardware, RTL Design Engineer * Hardware, Design Verification Engineer * Hardware, Physical Design Engineer * Hardware, Performance Engineer * Hardware, Compiler Engineer * Open Application, Exceptional Talent river.ai/careers We are a small, elite team of researchers, builders, and pioneers from the world's leading AI labs. If you want to do the most ambitious work of your career alongside, apply today!
23
5,787
Hardware engineers: come build something new.
We are incredibly excited to announce River AI. Our mission is to create personal AI that is owned and shaped by you. Today’s best AIs are controlled by a few large corporations. We are building the alternative: a new, personal stack for AI that works entirely for you, shares your values, and operates on your terms.
3
24
4,613
CPU thread def: PID+context+mem GPU thread def: whatever you feel like
1
5
1
40
5,705
To be fair, at low enough FP quantization it does indeed return to associative behavior. Bc it’s INT
Floating point math is not associative! And many of the highest performance kernels split the workload among SMs and accumulate partial results in a nondeterministic order. Many AI labs just accept this, or pay a huge performance penalty for determinism. DeepSeek decided to do neither. (1/4) 🧵
2
9
3,155