3 ms·
Additional note, the L4 inferencing board is also still on PCIe-4 (like the L40, very sadly), so NVIDIA doesn't seem to foresee the need for more bandwidth on i
by touisteur 3y ago
Additional note, the L4 inferencing board is also still on PCIe-4 (like the L40, very sadly), so NVIDIA doesn't seem to foresee the need for more bandwidth on inferencing workloads in the next 3 years (until they maybe get a H30 out, with the binned not-fully-functional H100s? Maybe? Hopefully?) and/or will make you pay for a full H100.