3 ms·
Its not a big deal. In a nutshell, Intel found that customers want to aircool the 600W liquid cooled SKU, instead of buying a separate 450W aircooled SKU. And t
by brucethemoose2 3y ago
Its not a big deal. In a nutshell, Intel found that customers want to aircool the 600W liquid cooled SKU, instead of buying a separate 450W aircooled SKU. And they are making a neutered version for export to China.
...But is anyone on HN actually using these things? I haven't seen a single paper, HF model or GitHub repo even mention a Xeon Max. All papers seem to use A100s or the rare H100 for experimentation, and the occasional GitHub project tests on Arc GPUs.
- c0balt 3y agoWe recently got some in for testing at work. The adoption of intel GPUs, and AMDs for that matter, however will pose a challenge as the inertia in many fields is high for CUDA. This is also supported by big CUDA ecosystems for, e.g., MD, Fluid Sim and Tensorflow. Further we very rarely see HIP and even less OpenCL for any serious workloads. It's usually CUDA or CPU (AVX-{2,512}).
- washadjeffmad 3y agoDid a department request them, or do you know if Intel initiated? We (oddly) haven't had any reps pushing Xeon Max at all despite it not being out of alignment with our purchase history.
- c0balt 3y agoAfaik it was intiated by an intel partner. There also has been some interest by internal sources as the NVIDIA pseudo-monoculture is not fully welcome.
- VHRanger 3y agoHistorically that sort of hardware was used in the physical sciences in HPC, stuff like fluid dynamics