If there is any gift we could give to the next generation, it would be a way to undo fixed tech gluttony, and enable both the simple and maximal things to co-exist in economy of scales that are priced so consumers can choose.

I personally prefer my prior 60's C10 truck which had just a few things to replace to fix at any time {carb, distributor, plugs, plug wires, fuel pump, etc}. Even engine work was approachable by a kid in a garage. Just as good for transportation as anything today.

Cars with 100+ devices connected on a network bus, with OEMs that are moving to encrypting in-car network traffic. Door handles with like 4 ICs. The minimum cost/complexity floor simply increases as the consumer is boiled. No one makes a vintage simple car at scale.

In the CRT era, economies of scale existed for the parts to build simple things. So one could build a relatively simple device like say the original B&W CRTs. As tech ratchets, the parts for simple things are no longer made in any scale that is cost effective . .

Seems like today, finding a way to bring back a good display (like CRTs had been) that isn't at crazy high pixel rates, and is quite cost effective, is the inflection point where one can start to bring forward quality alternatives to today's personal computers.

Vintage Arcades for the longest time standardized on 224p 60Hz CRTs while the rest of the world rez'ed up. But that kept pixel art alive and practical for a long time. Tech tech was approachable . .

How would say a new Commadore or say fantasy console arch work in an era where all graphics scaled from lower res looks horrible unless you invest in $0.75K CRT scalar? Pushing 4k at a good FPS like 60Hz is effectively impossible for a cost effective FPGA based arch . .

When all the new HW being attached to a computer is only max spec, it's a complete lockout of any new startups, only monopoly players have the scale of money for bleeding edge . .

And it compounds, storage, GPU, CPU, etc, all have to ratchet up in cost now, because !/$ in process tech has tapered off . .

The problem as they start pushing the boundaries of everything is the shortcuts they employ. Only one 27" QD-OLED at 360Hz supports a DSC (compression) disable. Some day they will require compression on everything, and probably chroma decimation, locked in reduced quality . .

Things like displays, each year the minimum spec required to drive a display at native ratchets up. Ratchets as in once they finish the 4k transition, 1440p and smaller just won't exist. At some point 4k won't exist. Interfaces of lower cost or burden won't be manufactured . .

__ Tech Spec Gluttony Age __.It is what we are handing off to our children. Engineering culture focused on peak marketing graph numbers. Crazy hi min-spec display and camera res. They will carry burden of dealing with unconstrained tech obesity, with no weight loss plan .

That new 1440p WOLED has a built-in 720p at 720 Hz mode. Who is ready for rending the entire frame in 1 millisecond?.

If anyone is reading between the lines, this pair of talks is basically the blue-print to getting spatial+temporal scaling that works down to low end devices without needing ML TAA and Frame Gen crutches, and it's all public domain or MIT [free for easy integration].

So the Shader BFI talk is getting rescheduled to time with something, and I'm moving the Display + CRT Spatial Scaling talk to release first. Mostly done with slides now, it's loaded with cool stuffs.

RT @dan5sch: Here's an update on my small-voxel renderer, inspired by the look of games like Doom and Quake. The work I shared last year fo….

Haha Radio Shack . Fry's . once upon a time kids built radios, now they just doom scroll their phones. If we want skills grown in the US, they have to tinker with real things . .

Fun watch, just for the part where they pull the GPU chip and build a double VRAM card: - Lets be honest, the US is massively lacking in that boutique electronic mall culture that is all over Asia.

Note all these are trick questions, because the micro bench information doesn't specify memory access patterns (tiling) and how compressible the data is. So there is no trusting those results. Rather it is a highly open ended question: name all the possible factors.

Follow-up interview questions on write speed: why for all granularity of stores is image stores faster at larger working sets? Also why is small granularity and small working set stores faster for buffers than images?