> It's very hard to beat the encoding performance of libjpegturbo, even with an asic.
Pretty much any platform with a hardware jpeg encoder beats libjpegturbo. So probably all smart phones, large scale CPUs, even tiny few $ microcontrollers like ESP32's have hardware encode/decode that outperforms libjpegturbo.
Not linearized as in samples do not represent linear color space? I think that's untrue, as all sensor work I've done (a decent amount across many projects) are linear: double light = double photon count = 1 bit of sensor space. There are technically often some bends from physical and engineering effects (photon interaction with substrate has some variability, temperature affects things a bit).
Or is there some other meaning of "not linear" you mean?
I would not say $1-$2k/mo extra income counts as a huge amount richer. It’s like 58th percentile by family income. I don’t really think it makes sense to call solidly middle class income rich in any sense…
In the sense they were not opposed to all machines as a concept. They were definitely against machines they thought reduced their labor value:
"They wrecked specific types of machinery that posed a threat to the particular industrial interests in each region. In the Midlands, these were the "wide" knitting frames used to make cheap and inferior lace articles.[22] In the North West, weavers sought to eliminate the steam-powered looms threatening wages in the cotton trade. In Yorkshire, workers opposed the use of shearing frames and gig mills to finish woollen cloth.[23]"
Yes, that's what I wrote. "They were opposed to bad labor practices, with machine-breaking as way of applying pressure on employers."
Otherwise it would be like claiming that an industrial union is opposed to people working in factories by pointing to a strike action and concluding they members just want to stand around outside the factory instead of work.
Recall that my comment is in response to "it is economically infeasible for me to be a luddite".
If that means something like "infeasible for me to oppose bad labor practices" then collective action is the typical response to bad labor practices.
If that means something like "infeasible for me to destroy the employer's use of AI" then I think that's mistaking what Luddites did as being the Luddite goal.
You wrote "The Luddites were not opposed to machines." What you wrote makes it sound as if there was no problems with machines, only with labor practices. That's simply misleading to untrue in this chain.
You can say Luddites were not only opposed to machines they thought cost them (which was the main reason they started the trouble they did at that point), but they also (insert labor practices).
Read their writings or contemporary reports. The wiki page has ample references to good material.
"Luddites were not opposed to the use of machines per se (many were skilled operators in the textile industry); they attacked manufacturers who were trying to circumvent standard labour practices of the time".
How is that meaningfully different than what I wrote?
I think your maybe counting just physical sales? 15 years ago was the start of the shift to digital sales, which have greatly grown. Just steam (which is the bulk of digital PC sales):
Steam units sold: 402M (2019) → 704M (2024), up 21% in 2024 alone.
Steam peak concurrent users: 3.52M in January 2011 → 42,042,778 on January 11, 2026. Roughly twelvefold, directly measured from Valve's API.
Valve's own disclosed number, which isn't an estimate by anyone: between August 2024 and August 2025, Steam averaged 109,000 people making their first ever purchase every day. That's about 40 million brand-new paying customers in one year.
Games released on Steam: ~8,100 in 2019 → 20,282 in 2025.
Some of the best selling games did come out roughly 15 years ago, but that's because they've been selling consistently ever since. Like Minecraft, the Witcher, etc.
Games released is not indicative of revenues: 80% of steam games make < $5k, and median steam game makes at most a few hundred $. So that doesn't really tell much about sales.
Recent data seems available in reports from Newzoo: their The PC & Console Gaming Report 2025 states there was 1.2% total PC growth 2021-2024 in nominal dollars (40.5B to $40-42B). In real dollars that is around -15%, so enough shrinkage that many industries would really feel the pain. In smaller sections, their 2014, 2017, 2024 reports show PC share of global games spend fell 17%. They (and SuperData) data show premium PC fell a few percent (nominal, more in real $). Playtime per player fell 26% (Newzoo 2024, 2025).
Also, around 50-60% of revenue, which is nice, is not new game sales, but microtransactions, and these are extremely weighted to the largest few games. Of the remaining, around 10-15% is subscription fees, also not new sales.
Using Steam as a proxy for the whole industry hides losses if more title % moved to Steam, since then the numbers you posted are not comparing like to like.
Some Newzoo data is easily found, use AI to find copies of their closed reports floating around the internet.
As someone making mobile apps used by billions, demands from mobile users certainly grow and want this class of hardware capability. We spend significant effort trying to figure out how to deliver features asked for by our users for high end phones and cutting edge hardware, while not alienating the wider user base without these capabilities. Competitors do exactly the same.
Pretty much all cutting edge graphics papers now use some form of NN. To ignore the techniques and gains at the cutting edge would leave any hardware mfr behind. It’s not nonsense or marketing speak. It’s genuine capability improving technology.
To see a representative stream of papers, go to arxiv and look through the recent graphics papers.
Datacenters don’t run on desktop class GPUs. They’re well past that. It’s why NVidia moved most production of desktop GPUs to data center class hardware.
That’s completely backwards by orders of magnitude. Modern CPUs outrun L1 cache speed by a lot, L1 cache speed is on order of 1000-5000 GB/s, a GB/s is real bits used, is 8x of Gbps, and network speeds include significant overhead bits for frame packing and other data unused for the actual payloads.
Pretty much any platform with a hardware jpeg encoder beats libjpegturbo. So probably all smart phones, large scale CPUs, even tiny few $ microcontrollers like ESP32's have hardware encode/decode that outperforms libjpegturbo.
reply