Aaryaman "Jam" Vasishta

ML @ AMD

Former ML+3D Engineer @ Stability AI

Ex. AMD Research Engineer, RT & Neural Rendering

2021 Graduate, Computer Graphics Group @ University of Tokyo.

[Aaryaman "Jam" Vasishta @adyaman]

Super happy to see this :) all those late nights fixing and getting ROCm + PyTorch running on Strix Halo with @scottttw
on TheRock have finally paid off 🥲 https://t.co/UtANhHk8Hf

It's now ready and widely available. This is just the beginning, and we're just getting started!

[Anush Elangovan @AnushElangovan]

Local [Superintelligence + Supercomputing] + Signed by @LisaSu 💻 🚀🚀🚀

Ryzen AI Max+ PRO 395 (Strix Halo) with ROCm
* runs GPT-OSS locally
* runs Battlefied 6 like a desktop
* 16x Zen5 cores for builds

@sama @gdb Sarah As promised please tag me if you run into any issues 🤙

[Aaryaman "Jam" Vasishta @adyaman]

@doublemover @AMDRadeon @AnushElangovan Can you please report any bugs you experienced in https://t.co/4oThAUsleF ? We'll take a look at it.

[AMD @AMD]

Generative AI gives answers. Agentic AI executes work but it needs more than just fast answers.

That shift changes the infrastructure equation. As agents scale, customers need CPU-rich infrastructure that can support more orchestration, retrieval, APIs, databases, caches, containers, VMs and workflows at rack scale.

That is where AMD EPYC processors are built to deliver the highest performing CPUs for the next era of AI: https://t.co/kWZXaYSYSR

[Saša Marinković @SasaMarinkovic]

192GB Unified Memory: @AMD Ryzen AI Max 400 Series supports models that are too big for N1X silicon

[Aaryaman "Jam" Vasishta @adyaman]

@wild_zones @AnushElangovan @AIatAMD 🙏

[AMD @AMD]

AMD Ryzen AI Max+ is available now.

16 cores. 32 threads. 40 compute units. 50 TOPS NPU. Up to 128GB unified memory.

Built on x86, it brings breakthrough local AI performance to the enterprise and developer workflows teams already use. No translation layer, no performance penalty, no workflow disruption.

[Anush Elangovan @AnushElangovan]

🚀🚀🚀

[Patrick Moorhead @PatrickMoorhead]

@raywccf @AnushElangovan Anush is based.

[Anush Elangovan @AnushElangovan]

Open is a Philosophy !!! Speed is the moat

[Aaryaman "Jam" Vasishta @adyaman]

@DavidBennett__ @hara_shimpei @AMD @aiand_ Considering buying some AMD GPU’s for your cluster in Tokyo?

[Aaryaman "Jam" Vasishta @adyaman]

Glad to have helped with one of those upcoming playbooks :)

[AI at AMD @AIatAMD]

Introducing AMD AI Playbooks, your shortcut to building AI on AMD.

Step-by-step guides to help you run real AI workloads, customize them for your environment, and get up and running faster.

New playbooks drop every month.

Start building: https://t.co/eA1h5ePtF0

[Anush Elangovan @AnushElangovan]

Tell us how we can make ROCm better. We are listening and we will grind it down.

[Aaryaman "Jam" Vasishta @adyaman]

@wild_zones @AnushElangovan @AIatAMD https://t.co/oORjfNF5nv has been merged. Please try a build and let me know if you face issues.

[Aaryaman "Jam" Vasishta @adyaman]

@AnushElangovan @wild_zones @AIatAMD Build commands shared here https://t.co/fB4jAYhsID much simpler than before.

[Aaryaman "Jam" Vasishta @adyaman]

@AnushElangovan @wild_zones @AIatAMD @wild_zones with the above PR also applied to flash-attention itself, we have it working fine on windows. I'll post a short list of commands that makes it easy to run with comfyui.https://t.co/oORjfNF5nv

[Anush Elangovan @AnushElangovan]

7.13 has multi-arch packaging so you can pull only the arch you need at runtime -- get the benefits of native ISA targets but also don't need the bloat (or jit to a lower level ISA).

[Anush Elangovan @AnushElangovan]

The challenge has now concluded. Please tag me here if you have eligible submissions by EOD 4/20.

[Aaryaman "Jam" Vasishta @adyaman]

@xennygrimmato_ @mike64_t Agreed. I tend to find older games running smoother on wine than on windows 11

[the tiny corp @__tinygrad__]

And now it's documented. For people serious about performance, this is a real AMD advantage. Thanks @AnushElangovan for pushing this through! Code is here: https://t.co/WEEj9EVZe8

[Anush Elangovan @AnushElangovan]

BS. Right on target for H2 2026.

[Anush Elangovan @AnushElangovan]

Speed is the moat.
MI455X is both fast, and is the fastest I have seen in execution for bringup of a complex GPU platform. MI455X is right on target for shipments in 2H2026 - Irrespective of what @SemiAnalysis_ says - rev your engines because speed is coming.