> Price looks good if it includes the RAM. Am I out of touch? Isn't 16 cores of almost Cortex-A78 level for $120 insanely cheap? (5% slower on specint2k6 but 5

> Price looks good if it includes the RAM. Am I out of touch? Isn't 16 cores of almost Cortex-A78 level for $120 insanely cheap? (5% slower on specint2k6 but 50% less area than A78 [0]) > I also wonder how advanced is the compiler support. The compiler support for RV64GCB is pretty much on par with other architecturs, as far as I can tell. The vector extension has autovectorization support from both the latest gcc and clang. There are already cases where compilers can vectorize code for risc-v but not for x86 [1], but the rvv codegen isn't optimal yet, but it's steadily improving. From the top of my head: gcc doesn't choose the best lmul yet, although the experimental code for this can be already enabled. It also doesn't perfectly register allocate and sometimes generates redundant moves (clang is really good at this though). Clang generates more setup code then needed, because it doesn't yet fully use the rvv strip mining. I think once we can actually get the hardware next year, most shortcommings wil be fixed. Oh and the NPU they plan to add to the x280s also already has compiler support, because it uses Sifives VCIX interface, which is already aupported by llvm [2], so you can write assembly with it and use c intrinsics. [0] https://www.youtube.com/watch?v=C4DcZyB6i1E https://www.youtube.com/watch?v=C4DcZyB6i1E [1] https://godbolt.org/z/f1jdbTbE9 https://godbolt.org/z/f1jdbTbE9 [2] https://llvm.org/docs/RISCVUsage.html https://llvm.org/docs/RISCVUsage.html