Mastodawn

In my second substack post I talk about some of the challenges of building an ecosystem, but I end with a surprising twist: everything involved in building an ecosystem in a new tech venture is actually things you should be doing anyway.

https://codeandrew.substack.com/p/my-brilliant-technology-needs-an

My brilliant technology needs an ecosystem. Now what do I do?

So many new technologies need an ecosystem to be viable, but no-one wants to build one for me

Andrew Richards

Show thread

Andrew Richards Jan 7

SYCL is now truly turning into a cross-platform high performance software development platform with Qualcomm's upcoming support. This started when Qualcomm the UXL Foundation.

Show thread

Andrew Richards Jan 7

"We’ll also be introducing native OpenCL 3.0 support, also as used by our other products. And then in the first quarter of 2026 we’d like to introduce SYCL support, and SYCL is a higher-end compute-focused API and shading language for a GPU. It’s an open standard, other companies support it, and it helps us attack some of the GPGPU use-cases that exist on Windows for Snapdragon."

Show thread

Andrew Richards Jan 7

Read the interview with Eric Demers here for lots more interesting details on Qualcomm's GPU: https://chipsandcheese.com/p/diving-into-qualcomms-upcoming-adreno?utm_campaign=post&utm_medium=web It's a fascinating interview for me because it covers the benefits and challenges of low-power GPUs. We hear lots about very high-end GPUs and Gigawatt datacentres, but bringing those GPUs to the consumer and much lower power consumption is going to make GPU-accelerated applications like AI much more useful for those of us without our own nuclear power station.

Diving into Qualcomm's Upcoming Adreno X2 GPU with Eric Demers

Hello you fine Internet folks,

Chips and Cheese

Andrew Richards Jan 7

What's interesting is how SYCL is being used for such a wide range of applications. We've seen SYCL used in supercomputers for science, medical systems for diagnostics and now for new applications targeting Windows PCs. This is a real proof-point of the benefits of cross-platform development: it lets developers target high performance for a very wide range of use-cases because of the freedom to choose the right hardware for the job.

Andrew Richards Dec 14

I spent a lot of my time at Codeplay wondering why it was so hard to build a disruptive technology company. A lot of what I ended up focusing on was the ecosystem. In technology, ecosystems are critical, especially for businesses. But they're not a direct revenue-generator, so they get overlooked. I thought I would start writing about it here: https://codeandrew.substack.com/p/why-is-disruptive-technology-so-hard This is going to be a mix of tech and entrepreneurship. I hope it's both helpful to tech founders & also starts a good discussion

Why is disruptive technology so hard?

I mean: you've invented something that can change the world, so why aren't people throwing cash at you?

Andrew Richards

Andrew Richards Jan 8, 2025

Phoronix Jan 8, 2025

Intel's Clang Code Begins Landing For OpenMP Offloading To SPIR-V For GPU Execution

Intel software engineers have been working on allowing OpenMP offloading to their Intel GPUs by way of targeting generic SPIR-V, the common intermediate representation used across Vulkan / OpenGL / OpenCL drivers. The initial patches to that work have now landed in upstream LLVM/Clang 20 for OpenMP offloading to SPIR-V...
https://www.phoronix.com/news/Intel-Clang-OpenMP-To-SPIR-V

Intel's Clang Code Begins Landing For OpenMP Offloading To SPIR-V For GPU Execution

Andrew Richards Nov 13, 2024

I'm going to be at Supercomputing #SC24 next week in Atlanta. Really looking forward to going as I've had to miss it for the last few years. I'll be meeting people to talk about getting them involved in the UXL Foundation and really growing the open ecosystem for accelerated computing, such as GPUs

Andrew Richards Nov 13, 2024

The way we achieve peak performance in SYCL and oneAPI is very practical: we actually use hand-optimized libraries underneath standard APIs. Here's a talk from the recent UXL/oneAPI DevSummit on how we do this in oneMKL:
Portable SYCL code using oneMKL on AMD, Intel and Nvidia GPUs by Rafal Bielski of Codeplay:
https://oneapi.io/event-sessions/portable-sycl-code-using-onemkl-on-amd-intel-and-nvidia-gpus-uxl-q324/

Portable SYCL code using oneMKL on AMD, Intel and Nvidia GPUs UXL Q324 - oneAPI

oneAPI

Andrew Richards Nov 6, 2024

This is a great talk by Jeremy Bennett of Embecosm about using the open-source tools the UXL Foundation provides to support PyTorch on new processors based on RISC-V. Embecosm is a valued member of the UXL Foundation that specializes in building tool chains for new processors designed for their customers. The benefits of an open ecosystem is it enables exactly this kind of rapid innovation. See how it works in this talk: https://www.youtube.com/watch?v=DqNWF26A8Io

Building Tool Chains for RISC-V AI Accelerators - Jeremy Bennett, Embecosm

Building Tool Chains for RISC-V AI Accelerators - Jeremy Bennett, EmbecosmOur client is developing a massively parallel 64-bit chip for AI inference workload...

YouTube