Professional Documents
Culture Documents
Felipe F.A.Q
Felipe F.A.Q
Q)
Frequently asked questions and more
Disclaimer
This is based on my experience and expertise, that also means I am human and I can make mistakes. The
content is not just my authorship, I will probably have many inspirations from other people. SALT to some
exaggerations.
Windows 11
First and most important, all new versions of windows aren't new.. Surprise, they are all a Reskin of something
very old and the old stuff are TIED/DEPENDENT on the code. So it's always “Bloat”. Windows 11 is even worse
because it's not modular as his old versions, for example you disable Notifications and will brick Audio selection
and Network selection panels. There is Direct Storage that is new and promising but it will also come to Windows
10. As far as your compatibility needs, on Windows, the older the better.
GPU Architecture
Based on Tile-based Rasterization in Nvidia GPUs with David Kanter of Real World Tech, we can have a valid
example made by an expert. Modern GPUs tend to use more tiling rasterization and feel less responsive. Thanks
to KGCT more people know this. His, and other people's guesses are that NVIDIA Kepler(7xx) are responsive
and Maxwell(9xx) and higher aren't. AMD are all responsive, RDNAs will probably be the last low latency focused
AMD too, as the next generation have dual dies..
This quote is generalized since it depends on the apps and also AMD apparently interrupts all cores but it's an
example on how careful you should think about doing an affinity.
You can easily change for the worse if you don't do it correctly.
DON'T CHANGE DEFAULTS IF YOU ARE NOT SURE. REALLY SURE.
then it boils down to the quality of the testing scenario which means using applications that give reliable and
consistent results with little variation ============ then we have a 4th issue which is incorrect assumptions
one such assumption is that 0.1% lows have anything to do with input delay they do, but only represent a
component of it frametime variance matters more but it's still just a component a fifth issue the big "placebo"
issue these are actually issues 5 and 6 as reasons for placebo actually i will just keep writing issue 5 of placebo -
people usually say something it's placebo because they can't notice it that doesn't make them retards
automatically they simply haven't optimised the systems enough AND SET THEM UP PROPERLY in order to be
able to notice the differences of the "placebo tweaks" you can't tell 500hz vs 1000hz on a mouse when you play
at 30 fps with vsync on -----
issue 6, still tied to placebo just because something works for you doesn't mean it will for others, and the reason
is simple: TOO MANY VARIABLES --------- issue 7, not entirely placebo, it's more like 2 issues in one but it's still
a biggie people don't always know what to look at, and what to look for. Sometimes they may not even be
sensitive enough (sensitivity can sometimes be trained via experience). IF THAT WASN'T ENOUGH, that also
implies the REQUIREMENT OF KNOWLEDGE Specifically Vocabulary and technical repertoire for example if
you tell someone "look how the smoothness feels", they don't know what that is without some prior visual
example that illustrates stuttering if that wasn't enough just because you know how stuttering looks like doesn't
mean you are capable of imagining what microstuttering feels like --------------------
issue 8, which is more like a reminder: remember we are dealing with a BLACK BOX in the end and that we are
still going off of a lot of ASSUMPTIONS. THE TRAP IS THAT a lot of the times we ARE NOT AWARE of the
assumptions we make a simple example is assuming win32priorityseparation has definitive impact in real time -
when in actuality it has to be benchmarked with system restarts/reboots too p.s. i always assume "facts'' are just
"hey this is something that may work for you" "feeling" or 1000fps cameras are the only real deal because they
sample the entire pipeline no tools do that software* edit: some clarifications for potential questions (which come
from not understanding what's in the input-output pipeline): frametimes can't measure network stutters nor your
mouse click latency nor USB processing time.(..)