CUDA/PTX matmul kernel reports 6% gain on B200
A post published on April 27, 2026 describes a pure CUDA/PTX matrix multiplication kernel that reportedly outperformed cuBLAS by 6% on a B200 with M=N=K=8192.
Source: Kernel de matmul em CUDA/PTX supera cuBLAS no B200 (paulwillchan.com). Text prepared with AI from this source.
What happened and what to do
In a post published on April 27, 2026, the author describes a matrix multiplication kernel written in pure CUDA/PTX for the B200. The post reports that, in a test with M=N=K=8192, the kernel was 6% faster than cuBLAS. It points to a technical blog and a repository containing the implementation.
A company running linear algebra workloads can evaluate the implementation in its own environment, reproduce the test with representative data and configurations, and monitor performance, accuracy, and resource use. A benchmarking pipeline and comparative dashboard can help determine whether the optimization justifies its adoption and maintenance costs.
How the consultancy can help
Wendelmaques can assess the opportunity, define benchmarks and validation criteria, and propose implementation and operation of a testing and monitoring pipeline suited to the client’s infrastructure.
Next step
Send a short description of your workload and infrastructure to receive a proposal with a scoped diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.