Simplest matmul using PTX During the Chuseok (Mid-Autumn Festival) break I decided to build a clear mental model for how mma works at the PTX level. The example below is close to the simplest matrix multiply that mixes CUDA and PTX. The…
Recently, my first contribution to Triton ( PR #7918 ) was merged. This features a new matrix multiply operation introduced in sm_120 NVIDIA architectures (RTX 50XX series and RTX Pro 6000). Thanks to @masahi , and @mobiacham , I was able…
Work culture I’ve seen many people on LinkedIn saying that work-life balance is a lie. They claim that Europe, with its pro–work-life-balance culture, has been declining, while Chinese tech giants—famous for the 996 culture—are on the…
Introduction This is a super brief summary of SVDQUant . Please refer to the original paper if interested. Q X = round ( X / s X ) Q_X = \operatorname{round}(X / s_X) Q X = round ( X / s X ) s X = max ( ∣ X ∣ ) / q m a x s_X =…
다짐 I'm proud Nvidian. I work hard, ethically.
I have little experience with open source, but currently I’m neither associated with nor active in any open source project. I want to be, and I have a good plan for it. From my experience, I believe it’s a good plan, so other people who…