Article author here:

Here's the DVD:

https://www.extremetech.com/wp...

Here's the upscale:

https://www.extremetech.com/wp...

Differences (Focus on the model, because the disrupter beam isn't frame-perfect -- there's a one-frame offset by mistake that I need to go back and fix).

Sharper text across the hull.
Cleaner hull, period. All of the shapes are sharper. You can make out the fine detail in the greebling at the rear of the model in the upscale, where these blocks tend the blur together in the DVD.

If you want to see the difference (and this one *is* frame-perfect) against the previous upscale, look at the disrupter beam itself:

https://www.extremetech.com/wp...

There's far less distortion and many fewer artifacts in the energy beam hitting the shields in "CurrentDefiant" than in "Defiant3", and the CurrentDefiant model should look much sharper to you than Defiant2.

Article author here:

I cannot show you the DVD credits at genuine quality because YouTube's compression algorithm makes such terrible hash of them, they are *vastly* worse than anything you see off the disc. That makes it really frustrating to do comparisons, as you can imagine.

https://youtu.be/OqG_A72Q5fM?t...

The upscaled credits are probably the single-best place to see what the show can look like compared to what you'd see on the DVD. While I realize I'm asking you take my word for it, I didn't pour 20-40 hours of work per week for 9 months into this project to scrape out a tiny, near-unnoticeable quality gain. (Whether you like the quality of the upscale is a different question, of course, but the *difference* is noticeable).

The version of the credits above has a much sharper station and deals with the heavy aliasing that crawls across the entire image when you play back the DVD. The PAL versions are much better than the NTSC versions in this regard.

The S1 - S3 credits are a different story. They're in far worse shape and I can't do much with them beyond a bit of sharpening. The S4-S6 credits, however, clean up beautifully.

The problem with AI upscaling is that it's uneven. Some stuff gets a huge amount of extra detail, some stuff doesn't. It's jarring, you are admiring the great skin detail one second and the next someone's hair is smudged and back to SD.

This is much less of an issue than it used to be and different models are tuned differently as far as what kind of an effect they create. Some, like Theia, are user-adjustable. Like you, I hoped the studio would create its own version. I've given up on that and decided to do it myself. It won't compare to what ViacomCBS could do, but it certainly beats the nothing they've done.

(Also, running the DVDs through the processing methods I outline will improve their output quality, even if you do nothing to upscale them thereafter).

You said: " Will it ever be possible to design and manufacture your own CPU, GPU, ASIC or RAM chip right in your own home?"

The answer to this question is that it's already *possible* to build these components in your own home. The problem is that the manufacturing techniques readily available to consumers for building and wiring hardware together do not lend themselves to the rigors of modern semiconductor manufacturing.

But can you build *something?* Hell yes you can. Check this thing out:

https://www.extremetech.com/ex...

It's a 16-bit CPU with 256 bytes of memory and every single component is implemented in human-scale components.

No advance in 3D printing is going to allow you to manufacture, say, a Core i7 in your house because you lack all of the industrial manufacturing and processing tools necessary for creating the wafer that such a chip requires. But there have absolutely been explorations of using 3D printing to create circuits that can cheaply and easily be applied to all manner of surfaces, including clothing. The final product of these efforts wouldn't be the sort of silicon you'd play a game on, so it might not meet your definition of being a CPU, RAM, ASIC, etc -- but these are definitely subjects of existing research in manufacturing.

Software-as-a-service is only a value if it provides me with access to the same (or better) software for less money than I would have paid otherwise. This is generally just about impossible to do. In my case, I buy one copy of Office and use it for 8-12 years. I'm not remotely interested in Office 365, which offers the unparalleled option of... paying Microsoft far more money over that same 8-12 year period.

Other people, with different needs, feel differently. If you need 3-5 Office licenses and being up on the latest and greatest is important to you, then maybe O365 is a good deal. If you really want cloud storage, maybe O365 is a good deal. I do not want cloud storage and do not need multiple licenses. I therefore consider it a remarkably poor deal indeed. It would need to cost on the order of $10 / year for me to be interested, and MS isn't cutting the price *that* low.

Most companies have engineered these deals to pad their own bottom lines first, and any actual concern for the customer is... decidedly secondary.

Both, depending on the situation and the context of the market. Companies in highly competitive spaces racing to beat each other to the next big breakthrough will put technology into production as soon as it is ready. Companies with the luxury of waiting to maximize profits and R&D will wait as long as they can.

There is a third option here, however: Technology can be both long-planned *and* spontaneous. It is entirely possible to have a very difficult problem you need to solve that you invest a great deal of R&D into. After a decade of work, you find the solution you've been searching for. In this case, a long, steady investment yielded the result that allows for quick deployment thereafter.

In other cases, the long, slow, grinding work simply yields a usable product. Both OLED and EUV technology are examples of projects that were trumpeted 10-15 years before being ready for commercial deployment. It took OLED TVs over a decade to get to market. Intel started talking about Extreme Ultraviolet Lithography back in 2002. They thought it'd be ready, IIRC, by 2004 - 2005.

EUV will actually be ready for manufacturing in 2020, 15-16 years after the research on it began.

Since AMD launched Ryzen in 2017, Intel has added Hyper-Threading to some of its low-end processors, added +2 CPU cores to its Core i3 products, added +2 CPU cores to its Core i5 products, and added up to four CPU cores on its Core i7 CPUs. Prior to the launch of Ryzen and Threadripper, an 8-core Broadwell HEDT CPU cost over $1K, while a 10-core chip was ~$1700.

Today, Intel sells 8-core CPUs for $500 (instead of $1000) and 10-core chips are down from ~$1700 to ~$1000. And AMD's Threadripper offers far higher core counts and higher overall performance in many workloads, especially at the per-core level. A $750 - $850 2950X will outperform a Core i9-7900X 10-core CPU for significantly less money.

CPU performance per dollar has increased significantly over the past two years, thanks to increased competition from AMD. In GPUs, we've seen less movement on the whole, but that should hopefully change starting this summer.