roofline

But not sure why comparing to the Panamera is a good example, because it has a low and sloping roofline, and is only a 4 seater.

Inference will be cheapest when run in a shared cloud environment, simply due to the LLMs roofline.

My house was reasonable, but my bedroom was tucked into the roofline on the top floor.

> (And actually during generation you can use speculative decoding to do better than this roofline anyways).

Assuming `k` is small enough, which in practice often is, the arithmetic intensity of this kernel is 25-90 Flops/Byte, way above the roofline knee of any modern CPU.

Analytical (85-92% accurate): Fast backup using roofline model.

(And actually during generation you can use speculative decoding to do better than this roofline anyways).

The slides with roofline analysis are great!

That's why chimneys all have a more or less minimum height above the roofline (and people can get away with little to nothing for a house on a ridge line or like an ice fishing shack or something).

When I rested my head on the headrest (which I almost never do in a car), I noticed my head brushed the sloping roofline, which is a common problem in sloped roof designs like the CLS and Panamera and A7.

I need more tokens not less because the available weight models aren't quite as strong, but I roofline sm_100 and sm_120 for a living: I get a factor of 2 on the spot arb, a factor of 2 on the utilization, and a factor of 4-16 on the quant.

This limit is often referred to as the "roofline model".

Wow - the title is "basic facts" - but it should be "key insights" You wouldn't believe how many PhDs I've met who have no idea what a roofline is.

AI-NWP models have fantastic levels of numerical intensity compared to physics-based NWP models since the former are "matrix-matrix multiplications all the way down." Traditional NWP models perform relatively little work per grid point in comparison, which puts them on the wrong (badly memory-bandwidth limited) side of the roofline plots.