Raghu Prabhakar

AI @SambaNovaAI

[Raghu Prabhakar @raghup17]

@Hishinuma_t Congratulations @Hishinuma_t !

[Intel News @intelnews]

Today, Intel and SambaNova, announced plans to build a multi-year strategic collaboration to deliver high performance and cost-efficient AI inference solutions for AI native start-ups, model providers, enterprises, and government organizations worldwide, built around Intel Xeon-based infrastructure. It was also announced that Intel Capital is participating in SambaNova’s Series E financing round.
https://t.co/0F9TixleRG

[Annie Shea Weckesser @asheaw]

Excited to see @intel and @SambaNovaAI collaborate to advance cloud-scale AI inference and enable more scalable, efficient AI infrastructure. Having spent time at both companies, it’s especially meaningful to see this shared vision take shape. 🤝 Let's go!
https://t.co/5GszUJgrvt

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.

My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently.

So, here goes.

[Raghu Prabhakar @raghup17]

It's not just about raw memory bandwidth. Peak bandwidth utilization requires the architecture to overlap compute, memory, and cross-chip comms. And, minimize overheads from kernel launch and sync. Doing this needs PTX gymnastics on NV - which makes it hard to scale.

For e.g., see this study https://t.co/vDXC9pv8P2

[Raghu Prabhakar @raghup17]

As @omarsar0 says: code is out for ACE, check it out! Joint work with @SambaNovaAI and @Stanford

[elvis @omarsar0]

Agentic Context Engineering

The code for the paper is finally out!

I had built an implementation for this (not exactly the same) that already boosted performance for my agents.

Evolving context for AI agents is a great idea.

Official implementation out now!

https://t.co/xpYWqnqvGc

[Raghu Prabhakar @raghup17]

Nice read. Well written, @jukan05

[Jukan @jukan05]

It’s honestly such an honor. Jensen Huang read the Chinese article I translated, and they even shared it internally.

[Raghu Prabhakar @raghup17]

This is not lying.
This counting.
Counting is math.
Math is science.
Nvidia is science company.

Love it!

[James Bradbury @jekbradbury]

opus 4.5 is really good at GPU programming, but somehow it’s even better at GPU programming jokes (h/t @Si_Boehm)

[sogabe @sog4be]

OpenAI/Anthropicのモデル群から選定すること多かったけど、SambaNova/Groqで使えるLlama4とかも簡単なタスクならいけるし、爆速かつ安価なのいいな

[Octave Klaba @olesovhcom]

L'inférence haute performance en "token / seconde / flow" exécuté sur les RDU de SambaNova (au lieu de GPU de Nvidia). En savoir plus: https://t.co/WmrYgDjHSA

[LeMagIT @LeMagIT]

Inférence IA à haute vitesse : OVHcloud choisit SambaNova https://t.co/Adm9xfTxj1

[Kenta Tanaka | WillRobotics | Physical AI @neka_nat]

ハッカソンで紹介された独自チップを使ったLLM推論サーバのSambaNovaを試してみた。
groqよりも高速でllama4が0.3秒とかで推論できる。
https://t.co/jVGSFdCUXU

[Raghu Prabhakar @raghup17]

I'm attending @Supercomputing tomorrow. Looking forward to meet friends and @SambaNovaAI partners from around the world!

[Raghu Prabhakar @raghup17]

Intelligence is Power, but also costs Power! This paper from @HazyResearch @JonSaadFalcon @Avanika15 quantifies the true cost of intelligence. Nice work!

Why the disparity? Comes down to utilizing the underlying hardware better. Dataflow's great at that - Go @SambaNovaAI !

[Jon Saad-Falcon @JonSaadFalcon]

Data centers dominate AI, but they're hitting physical limits. What if the future of AI isn't just bigger data centers, but local intelligence in our hands?

The viability of local AI depends on intelligence efficiency. To measure this, we propose intelligence per watt (IPW): intelligence delivered (capabilities) per unit of power consumed (efficiency).

Today’s Local LMs already handle 88.7% of single-turn chat and reasoning queries, with local IPW improving 5.3× in 2 years—driven by better models (3.2×) and better accelerators (1.7×).

As local IPW improves, a meaningful fraction of workloads can shift from centralized infrastructure to local compute, with IPW serving as the critical metric for tracking this transition.

(1/N)

[Annie Shea Weckesser @asheaw]

New independent study of energy usage for AI inference from local to cloud computing. To better normalize AI value across models, they introduce two key metrics : Intelligence per Watt (power) and Intelligence per Joule (energy). Research by @Avanika15
@HazyResearch @StanfordAILab @JonSaadFalcon 👇

[SambaNova @SambaNovaAI]

What is model bundling? 🧐

It's like having all your AI models hanging out on one rack, ready to tag in like a relay race. Now spread that across multiple racks, where every system pitches in to handle requests at warp speed.

Read more in our blog: https://t.co/4fCDSN6tdt

[Annie Shea Weckesser @asheaw]

🔍 @BoLi1550313 @UrmishThakker @raghup17 nice work 🙌

[Raghu Prabhakar @raghup17]

@KyleLiang5 Agentic IR!

[Raghu Prabhakar @raghup17]

Can't wait to be bandwidth-bound on everything on the #dgx_spark