تغريدة مثبتة
dela°🧸
5.7K posts
dela°🧸
@vxt0r_
in pursuit of the ultimate ideal, actively ridding oneself of every local maxima.
¯\_(ツ)_/¯ انضم Ağustos 2011
309 يتبع340 المتابعون
@semidoped @vikramskr fun episode. it was really technically dense. definitely would need to rewatch.
English
A masterclass on Google's TPU v8 Networking.
Two TPU chips? Pssh. We already knew workload-specific silicon was here.
But two scale-up networking topologies?
That's the actual Google TPU news.
Workload-specific interconnects. Think about that.
New Semi Doped with @vikramskr and @austinsemis.
Copper? Yep. Optics? Yep.
What we cover:
- TPU splits in two: 8t training, 8i inference.
- Virgo: 47 Pb/s scale-out fabric, 100% OCS.
- Boardfly scale-up: copper PCB + AECs inside racks, OCS between groups. 16 hops → 7.
- Training uses 3D torus (Rubik's Cube).
- Inference doesn't. Workload-specific topologies now.
- Dragonfly traces to a 2008 paper by Kim, Dally, Scott, Abts. Abts went on to build Groq's interconnect before Nvidia.
Chapters:
0:00 Intro
0:21 Two TPUs for two workloads
2:31 HBM, SRAM, and Axion CPUs
7:22 Why networking is the new bottleneck
17:14 Virgo: rebuilding scale-out on optics
25:24 3D torus Rubik's Cube scale-up for training 34:50 Boardfly: scale-up for MoE inference
42:07 Workload-specific everything
$GOOGL
English
dela°🧸 أُعيد تغريده
Hiring 1,000 new grad builders — because emerging AI talent is the experienced AI talent. Technical foundation required. Builder mindset non-negotiable. ⬇️
Marc Benioff@Benioff
I’m locked on, @DavidSacks! We’re hiring 1,000 new grads & interns right now to ride the AI exponential. You are right they said AI would kill entry-level jobs. Meanwhile these grads & interns are building it — powering Agentforce & Headless360 at Salesforce. 🚀 New grads: Drop your resume to @salesforcejobs or futureforce@salesforce.com #FutureForce #AI
English
dela°🧸 أُعيد تغريده
I’m locked on, @DavidSacks! We’re hiring 1,000 new grads & interns right now to ride the AI exponential. You are right they said AI would kill entry-level jobs. Meanwhile these grads & interns are building it — powering Agentforce & Headless360 at Salesforce. 🚀 New grads: Drop your resume to @salesforcejobs or futureforce@salesforce.com
#FutureForce #AI
David Sacks@DavidSacks
Narrative violation: Hiring of new college graduates is up 5.6% over last year. Youth unemployment for degreed 20–24‑year‑olds fell to 5.3% from 8.9%. Weren’t we told that 50% of entry-level jobs were going away?
English
dela°🧸 أُعيد تغريده
dela°🧸 أُعيد تغريده
China keep distilling those fucking models thank you for the hard work I salute you 🫡
English
dela°🧸 أُعيد تغريده
A Chinese company facing chip restrictions can train this
But xAI can’t even get SOTA
With a million H100 equivalents
Xiuyu Li@sheriyuo
Welcome DeepSeek V4 Pro Max huggingface.co/deepseek-ai/De…
English
dela°🧸 أُعيد تغريده
Funniest part about this is xai has had access to a significantly larger compute pool and can't seem to train anything decent to save their lives
Lisan al Gaib@scaling01
GPT-5.5 was trained on a 100k GB200 cluster probably the Stargate in Abilene Texas with 2 out of 8 completed buildings and ~112k operational GPUs
English
dela°🧸 أُعيد تغريده
He wanted to show the world what an intelligent black man looked like.
Many whites didn’t believe he authored his writings and he understood the link b/w racial intelligence and aesthetics in western society. This is why his portraits all emphasize his head. He was playing on assumptions in ethnology and phrenology.
The fact be did this despite being a fugitive makes it a bigger statement imo.
English
dela°🧸 أُعيد تغريده
So we are just cannon fodder for teams in the league now @ChelseaFC
English
dela°🧸 أُعيد تغريده
@reinerpope @itsclivetime @dylan522p Great conversation. I turned the key ideas into a clear note for easier reading: lilys.ai/digest/9191726…
English
Intelligence per picojoule, with @itsclivetime and @dylan522p
(0:00) Intro
(1:22) What is codesign?
(2:49) Codesign example: Swish vs ReLU
(4:22) Are DeepSeek papers codesign?
(6:45) Predicting where ML research will go
(8:06) Should researchers hate your chips?
(9:34) Can you codesign too much?
(13:23) Picking the right grain size for specialization
(16:22) How much hardware flexibility for The Age of Research?
(20:05) Did reasoning and RL disrupt hardware roadmaps?
(23:09) Cerebras/Groq: unexpected wins on reasoning and RL
(25:34) Disaggregating MLP and attention
(29:06) The right metrics for quantization and codesign papers
English
dela°🧸 أُعيد تغريده
Today we're announcing LevelUp: a free, four-week training program that takes people with no prior experience and prepares them to work as fiber technicians on data center construction sites across the US.
We built this program with CBRE because the fiber technician field, and the broader construction industry, is facing a nationwide shortage at a time when data center demand is higher than ever.
How it works:
🔧 Classroom instruction, hands-on labs + team activities covering transferable technical skills
🎓 Graduates have the opportunity to work at Meta's US construction sites through our contractor network
🤝 Open to everyone from recent high school grads to mid-career professionals
Since 2010, Meta's data center projects have supported 30,000+ skilled trade jobs during construction + 5,000+ permanent operational roles. LevelUp is about building the pipeline to keep that going.
Learn more: go.meta.me/0eb3f6
English
dela°🧸 أُعيد تغريده
but what if we made an LLM from scratch?
Reese Chong@_reesechong
Behind the scenes of mni-ml: January 4th 2026 - my roommate @MankyDankyBanky and I wanted to do a big project together. ”maybe we should try to build pytorch from scratch” We found @srush_nlp's minitorch curriculum and committed to grinding through it Jan to April. February - autodiff and tensor internals done. lots of late night PR reviews, stacked diffs, Kinton ramen runs to Toronto when I'd visit Aadi at Shopify. We started posting on X to keep ourselves accountable. March - the month of parallelization: Aadi shipped tiled matmul using the same algo @nvidia teaches in their CUDA guide, wrapped by end of month - pooling, conv1d/2d forward+backward, softmax, dropout. March 22-23 — @socraticainfo symposium & we see the tinytpu team on the stage which filled us with determination 🫡 cc: @evanliin @XanderChin @suryasure05 @kennykgguo March 24 - chose the mni-ml brand and started the educational blog March 30 - minitorch is DONE ahead of schedule. now we build on top of the framework. April 5-6 - cuBLAS matmul via koffi FFI. buffer pooling, strided batched GEMM, kernel optimizations. CUDA backend takes shape. April 7 - huge day. cross-platform CI pipeline, prebuilt npm binaries, v0.3.0 — CUDA live on @npmjs. flatten the monorepo, add @WebGPU + Windows CUDA build targets by eod. April 12 - flash attention CUDA kernel ships. we caught a bug where head dim > 32 was truncating. April 14 (during exam season), we recorded the demo in @Shopify recording studio during Aadi’s lunch break. Everything over the last 4mo finally came together. Cc: @fnthawar @tobi @alspee April 17: launch post and bought the domain mni.ml and we’re just getting started. We have so much in store for this summer, stay tuned 🫡 cc: @sundeep @GavinSherry
English