3 ms·
> In my microbenchmark1 the single triangle approach was 0.2% faster than two. Sounds like something that would be within the margin of error? Seems especially
by nsajko 4y ago
> In my microbenchmark1 the single triangle approach was 0.2% faster than two.
Sounds like something that would be within the margin of error? Seems especially meaningless because it's just the average of the timings, instead of something that would visualize the distribution, like a histogram or KDE.
- zamadatix 4y agoProbably quite far outside margin of error given the number of runs and simplicity of the test but you'd need the measured variance to be sure. The AMD study goes into more metrics and why's of it being faster.
- gpderetta 4y agoI know very litte of graphics programming and GPUs, but I would expect this particular shader to be very deterministic with very little jitter, right?