Divyesh Peshavaria

Forever Curious

[Lulu Cheng Meservey @lulumeservey]

You forgot to cut the crust off your slop

[Ryan Fleury @rfleury]

@nearX86 The way to revert back to this mindset is to actually understand that the goal is not to write as much code as possible, by hand or not

[Indian Strategic Studies Forum @ISSF_India]

Did the Reserve Bank of India (RBI) secretly sell USD 12 BILLION of its gold reserves to defend the Rupee?

A recent Bloomberg report claimed exactly that.

The RBI and Indian Govt stepped in, calling the report "FAKE". So, who is right?

Let’s break down the exact math that Bloomberg and other so called Economist used to fool the public. 🧵👇

[Charles 🎉 Frye @charles_irl]

New articles in the GPU Glossary for CuTe DSL, CUTLASS, and CuTe -- the tools used to write some of the highest-performance kernels on contemporary data center GPUs.

https://t.co/hhdyvAO092

[Daniel Wortel-London @dlondonwortel]

It’s not just new, it’s newspeak

[Nikita Bier @nikitabier]

@CodingNoobie Remove everyone on this list from the revenue sharing program @allegrajacchia

[Tim Zaman @timzaman]

[dusting off my blog] - I'd love for more people to get into AI Infra, and so I wrote about my path there from NVIDIA - Tesla AI - X - DeepMind and now OpenAI. I only see our industry growing, so this is for those wanting to gain a solid physical-to-sw foundation of the field.

[Divyesh Peshavaria @ParallelDiv_]

"a fast follower strategy requires the fast follower to follow fast"

[Nicholas Wilt @CUDAHandbook]

One of the great paradoxes of AMD is that their CPU division owes its existence and recent great performance to a successful fast follower strategy; yet the folks running their GPU division don’t understand that a fast follower strategy requires the fast follower to follow fast.

[Simo Ryu @cloneofsimo]

You guys have ML researchers in white house?

[Divyesh Peshavaria @ParallelDiv_]

@Hi_Mrinal @GPU_MODE @gordic_aleksa #365DaysOfUpskilling
Day 26 - Read through chapter 9 and learned about Service Discovery in Kubernetes. Two important points I learned today:
1. Automatic service registration with CoreDNS
2. Service to Pod IP translation via the Kernel trap through IPVS rules

[Divyesh Peshavaria @ParallelDiv_]

#365DaysOfUpskilling

Day 1
- Read this awesome explanation by @Hi_Mrinal on Time complexity - https://t.co/wy1GtUnhUV
- Watched @mattgodbolt's video on modern CPUs - https://t.co/qF6iunDW8O
- Started working on my personal portfolio website (Coming Soon!)
...

[Divyesh Peshavaria @ParallelDiv_]

#365DaysOfUpskilling
In 2026, I want to get deeper into ML Infra and Low level systems. I will be sharing my learning journey here (and hopefully make some like-minded friends along the way)

[Divyesh Peshavaria @ParallelDiv_]

@Hi_Mrinal @GPU_MODE @gordic_aleksa #365DaysOfUpskilling
Day 25 - I could not make much progress due to a turbulent schedule and had to pause posting for ~10 days.

Today I went through Chapters 1-8 of Nigel Poulton's The Kubernetes Book. It is the 2024 version, so a little dated but a good refresher still.

[Divyesh Peshavaria @ParallelDiv_]

@o_v_shake Amazing!

[Nicholas Wilt @CUDAHandbook]

It is out of print, but I negotiated a rights reversion w the publisher and will be moving the contents to the Web site.

[ℏεsam @Hesamation]

bro casually walks and explains 5 GPU performance optimization methods for LLMs. one of the most simple and intuitive explanations for beginners.