mkbosmans

7 ms·

31.
by mkbosmans 4y ago
Reasoning from first principles: When a problem is linear, a small pertubation in your input will result in a similarly small difference in the results, bounded by some constant factor. When a problem is non-linear, there is no such constan
32.
by mkbosmans 4y ago
To be fair, most analytical models working with data already incorporate a lot of 'data science' techniques. That does not have to be ML or NN, but could be as trivial as a least squares regression to fit you model to a set of obs
33.
by mkbosmans 4y ago
Yes, I think ML could be useful at places where the current physics modelling falls short. The nowcasting rain example from DeepMind is in some respects pretty comparable with the Next Ocean wave prediction. In the wave prediction case, wa
34.
by mkbosmans 4y ago
My guess would be: not that much. But I work in the field of numerical mathematics and computational physics, so I could have some bias. :-D The more nuanced answer would be that taking the raw radar data as input to e.g. a neural network a
35.
by mkbosmans 4y ago
Sure, but tide and wave models generally cover a much larger area, such as whole coastal area's and over larger time spans. An example of such a model would be: https://www.youtube.com/watch?v=eN6CDaoMZ7U In the Neth
36.
by mkbosmans 4y ago
This is a commercially available product for wave prediction on ships: https://nextocean.nl/wavepredictor.php Not specifically meant for rogue waves though, but rather to predict calm streaks in the waves affecting the ship
37.
by mkbosmans 5y ago
Your life will be a lot easier if you decide not to chase bit-level reproducibility when doing floating point calculations. You can try setting CFLAGS to get the same result on all hardware, but that is a lot of effort, especially if you d
38.
by mkbosmans 5y ago
Sure, but that doesn't change the fact that you cannot simply plug that into a simple formula to get the e.g. the optimal block size for cache-blocking. To be more precise: you can try to figure out how cache size would affect optimal
39.
by mkbosmans 5y ago
It is an overly strict interpretation of that rule to assume it prohibits any randomized algorithm. Aren't these the same steps for any given input? 1) Roll dice to get random number N 2) Get the Nth permutation of the input sequ
40.
by mkbosmans 5y ago
L is the latency of the instruction, so 1 clock cycle after the INC r8 instruction another instruction can operate on the result in r8. T is the reciprocal throughput, meaning in this case that the CPU can sustain 2 INC instructions per clo
41.
by mkbosmans 5y ago
The issue with that approach is that renting is a lot more expensive than owning a car long-term. Something like 4x. This means that the small car you buy should be at least 40% cheaper than the car you would originally have bought and are
42.
by mkbosmans 5y ago
Vectorizing by hand is not very difficult, but even for moderately complex loops it gets tiresome very quickly. And that is before considering loop peeling for vector alignment, adding both vector and scalar remainder loops and loop multiv
43.
by mkbosmans 5y ago
When I needed to work with a subversion the last time, I always found myself using git and git-svn when interacting with the version control system. It's not just the familiarity of the commands, but also the enormous flexibility you g
44.
by mkbosmans 5y ago
That doesn't sound right. Actual corruption of the repository can happen, but not by simply using another client than your team members. May be you mean something more alike commiting non-standard line endings? That can easily happen i
45.
by mkbosmans 5y ago
Assume there is a solution to the equation. Then if you take the mod n of both sides the new equality will also hold, for any n. Therefore, if you can prove there are no solution for the mod n equation, then the original equation is also n
46.
by mkbosmans 5y ago
At least the Marvin Gaye track is not a good example of just the Atmos difference. It is a completely remastered version that sounds nothing like the original. It also is 3.5 dB louder, which naturally would make the listener think it sound
47.
by mkbosmans 6y ago
I doubt they are short on inspiration for further improvements. A big part of the speed advantage of ifort is the aggressive loop unrolling, pipelining, splitting and multiversioning. Coming up with these transformations is not the difficul
48.
by mkbosmans 7y ago
Your links in the article to the HN discussion point to a different submission, not this one.
49.
by mkbosmans 7y ago
What are you talking about? Intrinsics don't force the compiler to use a specific instruction. It's not even a 1:1 mapping. I think all compilers will fuse separate add and mul instrinsics to fmadd. Both GCC and Clang will convert
50.
by mkbosmans 8y ago
So how much is that 1MP retina display covering of your field of view?
51.
by mkbosmans 8y ago
Can you share more details about this CPU? The page mentions a custom AMD EPYC, and indeed, your topology suggests this is not a standard 24-core CPU. A standard 24-core EPYC would have 3 enabled cores per CCX and 4 zeppelins (NUMA nodes) p