@demofox@aras Even before getting temporal: pre-sharpening before quantization, and using spatial error diffusion. And temporally, there is that interesting question of making it blue in time with fixed position, or blue in time with reprojected position (which is seriously hard, but would be fun to try on a GPU). Aras, it's inspirational to see blue 1-bit/pixel on that device, had been hoping others would have done that instead of fixed dither patterns :)
@aaaaaa_ Prey2 had an amazing graphics engine for 360 era thanks to Brian: idTech guts replaced, soft VT texture streaming, deferred, on-GPU HZB occlusion culling, atlased dynamic shadows, DOF, SDF baked shadows (if I remember right).
They had a great game designer at the time too. He used to dominate playing CoD with just melee. He deeply understood game mechanics. Years later he left this world early.
Wish world would have seen what that team could have built had things gone differently.
@aaaaaa_ ... Canceled Prey2 would have probably got a temporal AA too. Pre HumanHead I had prototypes going at home, and after leaving HH during the contractual turmoil, built out FXAA and then TXAA for NVIDA. Brian then started the UE4 TAA when he left HH for Epic. All that creative energy would have been placed into Prey2 instead had the project been supported to it's completion.
Sometimes things don't go as planned, but the inspiration you get working with great people lasts a lifetime :)
I am excited to finally share our recent paper "Filtering After Shading With Stochastic Texture Filtering" (with @mattpharr@marcosalvi and Marcos Fajardo), published at ACM I3D'24 / PACM CGIT, where we won the best paper award! 1/N
@BartWronski@BrianKaris@mattpharr@marcosalvi BTW, love the paper (quickly skimmed thus far). For screen-space: prior motivation to stochastic filter was to avoid gutters and/or multiple address translation cost in software virtual texturing. All my stratifed shading prototypes passed projected screen-space sub-pixel offset into the spatial/temporal reconstruction ...
@BartWronski@BrianKaris@mattpharr@marcosalvi ... for example it's possible to snap {x,y} sub-pixel offset of actual shading location to {i} index in a 16xMSAA-like (or higher) lookup table to send with low bit/pixel. While this is an interface data change from where UE4/UE5/DLSS2/FSR2/etc are today, I believe the benefits during reconstruction had been worth it at the time. But it's hard to say if that holds today in terms of !/$.
@BartWronski@BrianKaris@mattpharr@marcosalvi If one is going to go full CS atomic raster, might as well move to pixel stratified sampling of shading position (instead of frame viewport jitter) and adjust the TAAs accordingly. It is like a 2x (in area) improvement in neighborhood context (compared to regular grid frame jitter), so you can better infer reprojection.
@BartWronski@BrianKaris@mattpharr@marcosalvi Of course perhaps the step beyond that is to guarantee all important depths in a given tile have some representative shaded sample in the collection for the current frame, to avoid reprojection loss ... Can only correctly infer changes from the past if it is represented in what you know about the current frame.
@BartWronski@BrianKaris@mattpharr@marcosalvi Open question though of where to go: screen-space spatial/temporal reconstruction costs scale proportional to neighborhood needed to filter, so ultimately low shaded sample count mixed with huge contrast (HDR) forces bias, which perhaps could be amortized out towards convergence (after many frames), but it might be venturing off optimal !/$ solution space.
@aaaaaa_ Great read. It is interesting how far plasma got with temporal dithered PWM in comparison to the black crush (near zero step function) OLEDs seem to apply to hide the artifacts near zero.
Stuff gets so cheap it's not worth the time to rebuild it? Except this problem that US has transitioned {both at the individual level and many at the corporate level} into a country of accelerated trash production. So perhaps we should take some individual responsibility for keeping things running ... had this shop vac choaked (below) from mixed joint compound and plaster dust from my house rebuild, but post rebuild now runs like new. Engineering for customer repair is an important art!
Some industries have nitch restomod culture, like say this amazing 1960's Chevy II (Nova) seen at some random car show a week ago.
One large distancing factor from the modern the vintage era -> then things had been made to be simple, today things are made to be so complex they are nearly impossible to cost effectively service. Software and hardware together crossed over long ago. But it doesn't have to be this way.
Existance if a retromod culture means that some subset of the population finds value in the properties of the past. Yet more and more of today's goods have zero prospects of ever getting a retro culture following. For example CRTs are collected and restored, while LCDs are just landfill mass.
Something to ask yourself during engineering: does the product provide any intrinsic value that transcends just a simple increase in performance?
This trend in the PC industry of increasing 1984 is only accepted because of constant incompatibility breaks, and the prospects of that continuing will end as perf/$ tapers.
@yora Consumer CRT TV repair was a thing in the US, was certainly important in the arcades with near constant runtime, it's still a thing in say the Philippines where most of the population doesn't have the income to afford new.
But perhaps the point stands, high cost goods with long life get repair, as tech focus is cheap as possible without long life, it tends towards a landfilling culture. Amazon culture extends this, large % of goods broken from new factory.
Had bought a PVM with good electronics but with a metal case which was scrap due to rust damage (rust had pitted through the metal), sanded it down and painted with steel-it.com (black) anyway, good enough for and swapped to the parts PVM.
@sebaaltonen M4 is a 3nm process and the 14900K is on a 10nm process. In that context, perhaps 15% higher single core perf isn't all that impressive given the M4 has a few process generations of improvements to utilize.
Heard Orlando is going to get an H-Mart this 2024, but it is being built far on the West side. So the typical goto place for asian groceries is Lotte Market. H-Mart (at least the Raleigh one) has better bulk sushi. Often the Lotte stuff requires cooking. But at least fish eggs can be had in bulk, and if you actually go to the egg section they sell Balut (which likely no non-Filippino american could stomach)...
How does GPU memory actually work in an APU situation? There's a dedicated carve-out, but what specifically is the purpose of that? Can the GPU address all of the system memory in practice? Is the split between "device memory" and "host memory" in UMA devices on PC just there because otherwise titles / perhaps drivers / the OS would flip out about memory if you told the truth?
@dotstdy That and page allocation CPU side is quite expensive. For example on the AMD kernel driver on SteamDeck, if you try to allocate and usage really large regions of memory, it can even stall for a minute in kernel land, looks like a hang, but actually isn't.
Windows is getting increasingly hard to use with CRTs over the years, but at least with Win10 it's still possible to drive a BVM in 720p60 mode. Some docs in this thread.
This setup worked with both a BVM-A14F5U (pictured) and a BVM-D20F1U. The 14" has a custom SCART+VGA input card (using the VGA input below), and the 20" worked with a stock card's RGBS input.
Not figured out how to get a BVM/PVM to accept a raw VGA signal even with VGA input cards. This is true across the 2 different cards I have on two different CRTs. So I filter the signal through a Extron 201 Rxi (switch settings below). This also has the added benefit of allowing fine signal centering. The 201 outputs both HV and S for sync, so it works with VGA or RGBS input cards on BVMs.
Using an HDFury Nano external DAC (HDMI to VGA). I do make a custom 720p60 modeline in windows using the AMD driver matching the myth tv ATSC 'standard' mode. The defaults in Windows otherwise have porch settings that shift the video offscreen on the BVMs (too far off even for the 201 to recenter).
BTW, I'm still able to trick Win10 to drive 480p on both NV+AMD. I run through a scan converter to do 480p to 480i for my 15kHz CRTs. I've done this without an external EDID box. There is a trick selecting the mode using the 'List All Modes' interface that can sometimes workaround Win's default of upscaling in the GPU to the highest res the EDID allows. [below is a 720p config, but I do the same for 480p]
Little Big Econ Barr Street Trailhead (few minutes drive from my house) follows the Econlockhatchee River. Great place for a hike, part white sand trail, and wild gator viewing.
Once in a while you run across a Tortoise on the trails. This one didn't seem to mind being watched by a family of 4. It did walk off, but didn't turtle, and kept eating.
Have a pair of what I think are American Racer snakes living in my yard. Yard is either rock, or pine needles with bamboo, or ferns. In 2023 I accidently found their prior home under the pine straw. Not sure where they sleep now. I'm surprised they get enough to eat on 1/3 acre of my unusual (for FL) residential land, and that my 4 year olds didn't scare them away.
@bitinn Could be, but I don't usually see them out in the rocks like in the photos. Got lucky the light was right on a window so they couldn't see us watching from the inside of the house.
Desktop Vulkan question: is it reasonable to rely on resizable BAR being enabled these days? Just filling write-combined buffer memory directly, without all the code for sync/async copies from staging buffers, is super nice.
@dotstdy@sjb3d@dougbinks For AMD the answer is likely yes because APUs have shared memory, and pre-REBAR systems still have the Windows 256 MiB DEVICE_LOCAL+HOST_VISIBLE which might be enough for many games.
@dotstdy@sjb3d@dougbinks I remember NV didn't expose that 256 MiB chunk, but that seemed to have changed at some point maybe as early as 2017 (based off HW DB results).
Another place worth a visit in NC. Self guided tour, gets into the engine room, the mechanical computer room, machine shop, etc. Can see what was state of the art in engineering in the late 1930's.
If you enjoyed the visuals in the Dune 2 movie and want to try it out at a grand scale, at 230m tall, the Great Sand Dunes (CO) is a worthy stop. Just be careful, the nearby towns entire GDP comes from speeding tickets, and they will go as low as to ticket you for speeding during a lane crossing pass.
@castano Enjoyed the drive to the dunes (and valley with moving rocks) in Death Valley. But the CO Dunes definitely feel monsterous in comparison. April is a good time to go for CO Dunes, cool air, easier hike, no snow. There is a offroading section by CO Dunes too, but it requires airing down, and I didn't bring a pump this time.
Hiking in NC last week, came across a pair of trails next to each other, one with a "not a trail" sign, so we figured the 2nd must be a trail by elementary logic. Later, after a few unplanned river crossings, you know past the point of no return, it was likely we hit the local animal paths instead. Not much wildlife though, a few snakes (the kind that didn't attack, so likely not packing venom), a male dog tick, but no black bears.
Temp back from vacation, have a MakVision/Wei-YA M3129DS-LG new-old-stock arcade CRT sitting around ... was going to sell it, but after hooking it up to take a photo of it's 'poor linearity' (because it's the really short LG flat screen tube), realized it's actually not that bad, so will keep this one for a future cab.
Old curved CRTs are way better, but these end-of-stock Wei-YA's have new tubes ...
@castano Looks like a lot of fun. Week ago tried hiking the easy paths out of Bear Lake in Rocky Mountain, but with no gear (no cleats + flat shoes) and a pair of 4 year olds. Had to turn back at some point due to safety concerns. But felt easy in comparison to hiking the Great Sand Dunes later. Already miss the snow ... near acclimatization, we are already back in the flat lands.
I miss me some shader disassembly analysis and low level RDNA discussion from @NOTimothyLottes It has been over 10 days and the fans are getting restless 😆
Laser Phosphor Display (LPD) is the natural evolution for CRTs, however it will remain lost to history ... a low production / high cost video wall company owns the patent.
Getting a monochome phosphor screen seems possible (https://www.facebook.com/watch/?v=1917224951888550 lit with moving UV diode below). A combination of destructive interference film (had success coating a PVM) for anti-reflection, and extra screen tint mixed with high power laser could result in a high contrast dark black that isn't possible with CRTs.
For laptops that need external power bricks (big GPUs), would be amazing if someone like @frameworkcomputer would re-invent those with internalized power bricks. So you only need {laptop, cord} instead of {laptop, power brick, cord} when traveling.
Rental car review - 2024 VW Jetta. Really liked this car. Bad rental tires of course, but within those limits the handling was excellent. Engine is the base spec turbo 1.5L, and struggled a bit on steep hills, but it was fast enough to not be a problem. Adaptive Cruise Control is a first for me - slightly annoying in places, but once you get used to it, a handy tool. Android Auto, comfortable ride, huge storage space, average 33mpg over 800 miles. Would absolutely rent again.
@TomF Fun to rent the newer and automatic version of a vehicle you own. Tried renting Jeep JL recently. Takes roughly 3-4 seconds before stepping on the gas translates into forward motion (almost died pulling a left out into traffic). Doesn't seem to like to allow getting into M2 (auto gear 2) for engine braking on mountain downhill highways, prefers to fire the rotors instead. Couldn't figure out how to get transfer case into 4H. Ahh, so this is the modern Chrysler vehicle experience.
