blingthinger

Sure and drug dealers are good to keep around town because even though "way too many of them" sell drugs, some also sell chocolate?

"Way too many"? That's horrific justification for trying to declare something stable enough to not cause the bug report statistics to slope back up. Sounds like you need to do more evangelizing in all the other flight sim communities; teach the plebs how to make proper filter combos that won't CTD. Apparently you're the only person who knows how and you're obviously deficient in your teaching activities needed to regain unfettered reshade access.

GL a thing of the past? Have you told Zink that yet? I'm sure it'd like to start planning now.

There's one more... If the token suspecting linux user has Wine installed and replaces line 426 with: os.execute('WINEDEBUG=-all wine ' .. ah_helper_exe .. '&') it works on that OS too. I believe macOS has wine too? No clue how to start it there though. Of course that's irrelevant in the event you were already planning on cross compiling AHHelper.exe for the other OSes...

There's only 1 thing that will fix the pausing for everyone: async web access. Either have to background the process in powerhell or somehow bundle an external fancy library with FWL or switch to a language that has threads.

Yep. I was actually trying this just a few days ago with the latest GE Proton release. Also can't install the service or Acceleration packs (english version required blah blah). The general web search fixes didn't fix. And don't forget to turn on zswap. Probably at 30-50%. It's a swap space compression method that's very very light weight. You'll most likely benefit from using it on these systems.

Windows defaults to many gigabytes for its pagefile and it can dynamically resize. This trips up many linux users because it's not a visible option in windows. If you want to be safe, start with 16gb.

3-core cpu. This is an interesting case study for how good the die fab processes have evolved to, given that we don't see odd number core counts anymore. I would think 11 would run on system 2. Possibly 12. How much swap did you set up? They might have been getting killed for lack of virtual memory space. If so, increasing swap size would probably resolve it. I'd guess that you'll also want to set up zswap for both of those systems. And yes, low settings. System 1 certainly wants 10. Side note: FSX does run in WINE. Mostly.

IXEG/Tom needs to finish rewriting the XP documentation for LR. (read as: make XP's expansive API more accessible to future devs). My wager's on a 737 rewrite happening in XP13...

There's no elite-izing in any of those comments. More like (fairly accurate) age-ism. Or anti-corpo(rate) altruism. Simzilla's UI has never won any awards. Drenching it in the neon signage of what are effectively ads doesn't exactly help retain the demographic looking for simplicity. Nothing elitist or superior-ist about it. At some point you just want to go stare out into the fields and trees instead of squint through the blinding pizzazz of modern society.

Sounds like VRS (eye-tracked if the hmd has that bit) could be an alternative to quad views. And here's hoping that linux VR gets some attention.

Default. I've never bothered to try installing opensam.

Looks like I actually did lose 10 fps (~10%, X3D cpu). I didn't notice it because my monitor is 60hz and I limit the max fps. I forgot that the limiter was turned on and actually set at my monitor refresh rate after the recent multi threading updates. At that airport it's consuming a bit less than 1 additional gig of vram. Worth those costs I'd say.

Frontal? That looks more like it was the precursor to the rear-bird-strike jokes placed upon early Citations.

I ask because I bet it is filled up and you simply don't notice. The sim has gotten really good at processing vram scaling tasks with its multithreading and I've never been able to discern a performance hit when it's scaling textures. LR keeps optimizing what textures get moved around and when and it's better with each release. I do think AEP uses more vram if you have spare available, but I don't think most users will notice much. What they will notice is that it looks better than default.

What is your texture slider set to? And what % of your vram gets used when you run the sim? I've noticed the opposite if I leave texture slider maxed out. I've seen it gobble up 3 gb of vram at default KPDX.

As a linux nerd, everything in that post is entertaining.

And that's their private opinion. Like I said...it better get there in the aspects they're bragging about. That it's so much more difficult to get there than it used to be, is telling. He's used both models now and out of the 2 groups, he's the engineering PhD. I'll go with that "private opinion" over a few web devs any day. What's funny is you trying to bring "systems depth" into a neural network discussion. Such fight-picker, much surprise.

Oh wow I'm sure that's really big for you to be able say... it uhhh, it better be able to do that. FSX/P3d could do that and it's at the core of asobo's model. What's amusing is to hear Toliss describe interacting with the fs2020/4 dev ecosystem. He took one look at the FM and said "uhhh nope" and wrote his own. And that's just a drop in that bucket. All of his dev work was done in XP because it's so flexible and intuitive. The only part of the XP model he was overriding was the turbine engine. He trusted the default aerodynamics enough to just let it be as is. Not so with asobo's concoction. But this conversation isn't about what's potentially "missing". Nobody declared that it's missing anything. It's if one could create a generalized NN model to be able to simulate any airframe in any flight condition with any level of accuracy.

Just need train it with enough data and any blob can be whatever your heart desires. 2d. 3d. red trees. blue trees. It doesn't matter. Replace them with perfectly manicured shrub art of cats. I'm really curious what details the devs have to provide to train the NN. It rings a lot like dlss 1.0-style requirements. There were a couple scenes with windows/glass. Hogwarts and Starfield. Not a lot of attention paid there though. I agree that one will be interesting to watch out for. Though again...it's all in the training set. The reflections of off-screen geometry were in place. I struggle to see transparency being a problem. Just give it enough data and you'll have Pitt and Cruise dancing on a rooftop. Meteoblue tried. I'm sure it would be more successful now with current methods. Generalized flight model is where I draw the line for a consumer sim using NN simply because of the randomness and non-linearity (worse than short term wx forecast) and amount of quality data needed to build it. I'll eat my hat if it ever shows up in XP. XP's already really computationally efficient. I struggle to see a NN buying in especially when the GPU is already blowing out smoke on the graphics calcs. Ground physics? That's just a glorified mass-damper model. No need for NN there. Eh... maybe tire friction/flex models would benefit with a small NN but again...need lots of either high quality numerical simulation results or experimental data. Not easy to come by.

Me? No. Only if someone else wants it to. We've had plenty of off-thread-topic discussions in the past that didn't get locked because they remained science/engineering-focused. CFD-based weather prediction, NN discussions, etc. Climate, weather, and cgi are each rougly equidistant as far as algorithm differences go. Under the covers they're all neural networks. Huge matrices of weights waiting for a GPU to multiply them together. "AI" is a silly wallstreet buzzword. The initial convolutions will be different, but ultimately they're the same concept: aggregate massive quantities of data and spit out the most likely outcome. In the case of weather (climate didn't even need NN models...well they kind of did an aggregate prediction with lots of approaches, so still a statistical conclusion) you have tons and tons of historical sensor data and CFD predictions that are all in a perfect archived state for data mining and subsequent model training. Much like the process I described earlier with automating camera view and object motion to train a RT NN. There's still lots of non-linear behavior though, so it won't be as precise as dlss5 is right now (RT is just a bunch of straight lines and reflections). But compared with what we had before, yeah, it's a significant improvement. Jcomm and feline are career weather experts, hence my ask of him.

They are making a blooper here by not addressing the Resident Evil character. Her geometry is different. All the background remains the same. I'm thinking they did rework her underlying 3d. There's another Starfield character in the same situation. But otherwise everyone else is identical. JH isn't wrong, nor is the engineer. The YT'er is wrong. If you factor in the training data which does include underlying 3d, they're both correct. As for the sideburns? Most hair is shader-based. Not 3d geometry based. It's easy to see where the model would be left to its own measures there. What's more is that these "revealing" YT vids only ever mention the same 2-3 examples. If every frame was equally different, there'd be a story here. "Angry" YT slop otherwise.

Do you folks use Deepmind at all? Or is it being looked at? https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models

Oh did they also release full path traced samples of those scenes too? React? No, I'm talking simply distinguish. The higher res and bigger screens do become more obvious. VR does like a good 90 for the vast majority of the population to not get sick. 30 fps was is for 800x600 CRTs.

Oh heavens no. They'd build the app as they normally would and then run an automated script that moves the camera and characters/objects in thousands (if not 10s or 100s of thousands) of different lighting scenarios and positions, dumping the needed data along the way. Then you train the NN based on that monster data set. In this case the training would focus on generating the resulting lighting distribution for a given frame. Shadows, specular highlights, reflections from off-screen geometry, etc. That latter item was noted by one of the few neutral/positive reviews of dlss5, which is either too costly for full path tracing or simply impossible with pure screen space methods. You could hypothetically reduce that data set size to some degree with what they're calling PINNs: physics informed neural networks, which might include the energy-conserving refraction, reflection, and light propagation equations in the error-calculation steps during training. That could actually significantly reduce the data set. Maybe still a more experimental method though. What I'm describing yields essentially a per-frame lighting bake. Except that it's dynamic in its ability to respond to scene changes. And with the amount of data you would generate to train the NN, it's no longer a 'hope'. It's never needing to extrapolate.