@AWScloud leading keynotes, demos, launches ☁️. prev @wolframresearch . CS board @HofstraU. New Yorker. ‘Pinions my own.
Joined April 2015
- Tweets5.3K
- Following2.4K
- Followers2.4K
- Likes15.6K
Came for the Trainium3 benchmarks, stayed for the Neuron Profiler art
Amazon’s Trainium Takes a Strike at the Training King
Trainium Best for Inference?
But now good for Training too,
Native Trainium PyTorch, High Quality CI,
TorchTitan, GPU Impossible Profiler
newsletter.semianalysis.com/…
Nick Walsh retweeted
Introducing TorchNeuron, a new open-source tool that lets developers take their PyTorch code and run it on @awscloud Trainium chips, no rewrite required.
85% of AI researchers and developers write their code in PyTorch, and now, with TorchNeuron, the tools developers already use to test, debug, and scale their models all work on Trainium without modification. Teams building everything from voice AI to drug-discovery models are already using it. ⬇️
What started with components (DRAM, GPUs) is now coming for entire systems: “stranded compute/capacity”. Increasingly hard to price physical consoles for gamers (where losses on hardware are subsidized by profits on games) when the alt buyer is now someone generating AI tokens!
NOT EVEN THE PLAYSTATION 5 IS SAFE 🚨
Someone jailbroke a PS5 and got Qwen 3.8 27B running on its GPU at 20 tok/s. That's a real 27B AI model running locally on a games console.
The same model on the PS5's CPU did about one token a minute.
Meanwhile, DDR4 prices are up 24% in 3 months, AMD just raised GPU memory prices, and 5090s are basically gone.
PS5 coming to a datacenter near you.
Nick Walsh retweeted
AWS CEO Matt Garman with a16z's Raghu Raghuram on the biggest cloud provider in the world being rebuilt for the agentic era:
$220B in CapEx for 2026, two million NVIDIA GPUs on order, AWS's AI chips sold out through next year, and a new 30-second AWS account with no credit card required so agents can get rolling.
0:55 $170B in revenue, growing 37%
2:40 30-40% of revenue started as startups
4:55 Startups want a cloud for agents
9:10 An AWS account in 30 seconds
12:25 Agents don't need bulletproof databases
16:30 Why AWS won't sell every GPU to the frontier labs
22:00 Why Matt isn't worried about a bubble
24:45 Why AWS builds its own power plants
27:45 Planning for the next supply shortage
30:25 Saving one county $5K per resident
33:10 How AWS built its own chips
40:05 What's holding enterprise agents back
45:35 Why Bedrock never sends your data to the model
49:30 AWS's new AI security tool
51:50 Why AWS teams of 10 are now 3-4
YouTube: youtube.com/watch?v=rn_afJaP…
@mattsgarman @awscloud @RaghuRaghuram
Nick Walsh retweeted
Strands Box: an open source sandbox for AI agents. OS-level isolation plus Dogwood policies that can depend on what the agent has already done, e.g. "no outbound HTTP after reading customer data." Credentials stay out of the agent. macOS, developer preview.
go.aws/472amYe #AWS #OpenSource
Sat down with AWS's Matt Wood @mza and asked for hot takes. He delivered:
- Roll out AI like a political campaign, not a mandate
- <1% of work AI can do today hasn't even been tried
- per-token pricing is on its way out?(he bet us a beer 🍺)
Full ep👇
youtube.com/watch?v=Kmellvol…
Even Dante Alighieri could not have depicted a level of hell that rivals the experience of true purgatorio moderno - being sick in the back of an EV uber in stop and go NYC traffic.
Nick Walsh retweeted
Trainium has by far the best profiler of any accelerator. As you can see we have nanosecond-accurate traces of our programs, allowing you to write a program that generates images reliably in the profiler.
With all the recent discussion on WALs, figured it was fitting to bring this one back -- the log is the database.
From 2023 Monday Night Live with Peter DeSantis, where we talked Aurora Serverless v2, Aurora Limitless Database, and more.
youtu.be/pJG6nmR7XxI?si=OXHC…
The voice of the Toonami ads that will echo in my head forever. RIP, Peter Cullen - truly one of the greatest English VAs to ever do it.
youtu.be/OW7GYCbu6Os?is=eZLO…
Nick Walsh retweeted
昨天 DuckLabs 宣布加入 AWS。真正让我有感触的，反而不是收购本身，而是 AWS 那篇文章的标题：
The changing physics of analytics.
很多系统架构，其实并不是永远正确。
只是它在当时的 CPU、Memory、Network 成本下是正确的。
以前一台机器不够，就天然想到 distributed。
所以最近越来越觉得：
做基础架构最大的风险，可能不是技术落后。
而是把上一代硬件条件下的 trade-off，
当成了永远不变的 invariant。😂
Architecture is a function of physics.
allthingsdistributed.com/202…
Nick Walsh retweeted
I wrote an article for All Things Distributed about some of the background on today's DuckDB announcement and why I think it's so exciting.
allthingsdistributed.com/202…
"In Plato's cave we may still be alone, but at least we can be having fun with this weird distorted and alien thing that is now deep in the cave with us."
pantel.is/projects/ai-gaming…
Nick Walsh retweeted
Zak van der Merwe has spent 14 years building control planes at AWS. His blog is a candid look at what happens when your bookkeeping layer has to keep up with the scale of the cloud, and why some of the most interesting problems in distributed systems often hide in the most "boring" parts. allthingsdistributed.com/202…
Nick Walsh retweeted
Kiro now runs the same agent harness across your IDE, terminal, mobile, and web sessions. We rebuilt our entire agent architecture on an open protocol to make that work
Here's a technical deep dive into what we did: kiro.dev/blog/one-agent/
Nick Walsh retweeted
Pretty cool to see @awscloud firecracker microvms used in kimi k3:
> By running isolated microVMs with Firecracker [3], AgentENV provides a level of isolation and fidelity that container-based runtimes cannot match.
