Pathal @PathalEJ

The best egomaniac you'll ever meet. Joined September 2013

Tweets

187
Followers

42
Following

9
Likes

7

Pathal @PathalEJ

7 years ago

I recently discovered I never actually published the last of my Darkmantle code. I also recently discovered I hate how the final version was written 5 years ago. So I'm rewriting part of it and expect to be done in roughly a week. github.com/Pathal/Darkman…

0 0 2 0 0

View Details

Pathal @PathalEJ

8 years ago

@HamletEJ I mean, the C=P(E) conversion is not quite a sigmoid function, but I'm tempted to throw a function approximation at it and see what comes up one of these days.

1 0 0 0 0

View Details

Pathal @PathalEJ

8 years ago

@HamletEJ I *think* it's just legacy/congruency issues. The WC3 engine had some limitations with proc chances and it just carried through to DotA and DotA2. It's not something I've dug into a whole lot, and I can't be sure that isn't outdated info, or that the data I read wasn't flawed.

1 0 0 0 0

View Details

Pathal @PathalEJ

8 years ago

Not at SIGGRAPH and the news coming out on RTX+RT Cores are a little vague. However RT Cores seem like a good solution to the SIMT problem for path tracing solutions. Doesn't fix the whole problem but those are some impressive results regardless.

0 0 0 0 0

View Details

Pathal @PathalEJ

8 years ago

@PlainSimpleBen Well I don't think it had much of a choice. It was a classic 5core lineup problem where every bot needed farm and priority. If it knew how to gank (which clearly it didn't understand the concept of moving to gank, not just to push) maybe it could have salvaged that though.

0 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@PlainSimpleBen @followNoxville AFAIK, If there's a rock-paper-scissor tie, then "Total Groupstage wins - losses (game difference)". If that is even, then another round of BO3s.

1 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@Enzo90910 @PlainSimpleBen This is a good paper on dealing with the issue. They don't get too low-level with their latency claims though. research.nvidia.com/sites/default/…

0 0 1 0 0

View Details

Pathal @PathalEJ

9 years ago

@Enzo90910 @PlainSimpleBen Issue is you constantly pull small bits of data from a massive data set. SIMT is the bigger issue, but you use enough registers in accurate ray tracing that the GPU has to rely off shared and texture memory quite a bit.

2 0 1 0 0

View Details

Pathal @PathalEJ

9 years ago

@PlainSimpleBen With the potential growth of ray tracing within a decade, do you think the industry should migrate towards a new or additional GPU design/architecture? Doing research in GPU design it just seems horribly inefficient for global illumination needs.

1 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@PlainSimpleBen Honestly, even if not motivated by gaming/ray tracing, this could also be useful for AI and other innovations over the last decade that have been offloaded to the GPU. Can't say I'm an expert in modern machine learning, but surely there's some problem set that doesn't fit SIMT.

1 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@PlainSimpleBen If CPU cores weren't so physically large, they would frankly be much better at ray tracing than GPUs. Feels like there should be some sort of middle ground with DDR VRAM and small cores (to get maybe 1-2000 cores on a chip) without relying on the SM/Warp philosophy.

2 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@maelfus @Fierydemise Yes! Just not wow.

0 0 0 0 0

View Details

Pathal @PathalEJ

9 years ago

@Fierydemise Yeah it was a corsair or a fractal design case. I love the modular design of most Fractal Design cases, but the 200R I swapped to was dirt cheap.