3 ms·
IIRC I measured something like 10-50% performance difference (don't remember exactly, but it was somewhere in there), vs a reasonably well-regarded blas impleme
by moonchild 4y ago
IIRC I measured something like 10-50% performance difference (don't remember exactly, but it was somewhere in there), vs a reasonably well-regarded blas implementation. This was for dgemm specifically; I don't know if the story changes for smaller floats.
- saboot 4y agoNot too bad, I do scientific computing and choose Intel/Nvidia as their APIs for accelerated math operations are documented and supported for developers. I've been paying attention to what Apple has been pushing with their M1/M2 chips, and I'm pretty tempted to try it out, but unless these features are documented and supported I can't feel comfortable writing programs relying on them.
- mhh__ 4y agoThe API apple wants you to use is documented and presumably is here to stay. That doesn't help if there's some edge case you'd need access to the raw ISA but still.
- Someone 4y agoAlso, BLAS is part of that interface (https://developer.apple.com/documentation/accelerate/blas https://developer.apple.com/documentation/accelerate/blas) Of course it is a black box in that you can’t (realistically) try and speed it up. You still run the risk of Apple’s priorities being different from yours.
- kergonath 4y agoOTOH the interfaces are standard. Using Accelerate instead of the standard BLAS is just a compiler switch.
- kergonath 4y agoThey are not supported directly, but you can ue them through the Accelerate framework, which has an optimised BLAS and FFT implementations (amongst many other things).