Comments on: AMD ROCm 6.3 Has Goodies For AI Aficionados And HPC Gurus Alike https://www.nextplatform.com/2024/11/26/amd-rocm-6-3-has-goodies-for-ai-aficionados-and-hpc-gurus-alike/ In-depth coverage of high-end computing at large enterprises, supercomputing centers, hyperscale data centers, and public clouds. Wed, 04 Dec 2024 18:37:03 +0000 hourly 1 https://wordpress.org/?v=6.7.1 By: Slim Albert https://www.nextplatform.com/2024/11/26/amd-rocm-6-3-has-goodies-for-ai-aficionados-and-hpc-gurus-alike/#comment-241040 Sun, 01 Dec 2024 18:02:11 +0000 https://www.nextplatform.com/?p=145063#comment-241040 Glad to see that Fortran compiler in there! It harkens back (if I understand correctly) to those design decisions in the language, for storing matrices and higher-dimensional arrays, where Fortran relies on column-major order (colexicographic), while C uses row-major order (lexicographic). That basic difference makes performance-optimized computational libraries developed for one language (to enhance cache usage and multi-core partitioning that minimizes inter-core communication needs) roughly the “inverse” of what they are in the other language, and generally incompatible (mangling linear indexing).

Porting SGLang and Flash-Attention-2 to ROCm 6.3 and Instinct devices should also prove invaluable given the performance uplifts they gave to the A100s on which they were initially developed and tested. Overall, this is a great software update by AMD IMHO!

]]>