jdewerd

Loads of games from the era roundtripped their textures through lossy S3/DXT compression and then stored them as uncompressed RGB or RGBA.

I know this because I wrote a Unreal Engine texture repacking tool with a "DXT detection" feature so that I wouldn't be responsible for losing DXT compression on a texture which had already paid the price, only to find that this situation was already hyperabundant in the ecosystem.

Many Unreal Engine games of the day could have their size robotically halved just by re-enabling DXT compression in any case where this would cause zero pixel difference. This was at a time before Steam, when game downloads routinely took a day, so I was very excited about this discovery. Unfortunately, the first few developers I emailed all reacted with hostility to an unsolicited tip from what I'm sure they saw as a hacker, so I lost interest in pushing and it went nowhere. Ah well.

How long until it comes with a DRM AI and then my anti-DRM AI will have to fight it in a virtual arena (with neon lights and killer soundtrack, of course)?

ISM is thriving, the only tragedy is that carriers haven't figured out how to charge rents on it and that's a tragedy for them, it's a spectacular success for everyone using it for free.

Carriers don't need 6GHz for backhaul. They have fiber and cable and (other) microwave. Not to mention the ability to shape their own links with antennas and beam forming and do a good job of it rather than a "default job." What they don't have -- and shouldn't be given under any circumstances -- is the excuse to build a moat in the bustling public park.

Completely independent of bandwidth, higher frequencies also fall off faster. That's bad if you are trying to cover max space but good if you are trying to avoid noisy neighbors.

I never said anything about free or government-run WiFi, just about auctioning off the spectrum. Companies that build out the infrastructure should be able to charge for access, but they shouldn't be able to prevent others from competing by paying the government for exclusivity. That's a scam.

The most efficient way to extract money from people is to sell off the spectrum to the highest bidding rent seeker, I agree.

As for most efficient use of the resource, well, consulting my spectrum analyzer, ISM bands are winning by a mile and we should want more of them.

Is that because defense doesn't like them or is it because (non-wartime) defense moves on geological timescales and these are "new"?

Yes, but the more we mix sieve rejection into candidate selection the more we complicate the rule of thumb. "Reject even numbers as prime candidates" is probably OK to leave as an exercise for the reader, as is the equivalent "round every candidate to odd" optimization. The point about random vs sequential is well taken, though, and it doesn't complicate the rule of thumb, so I changed it.

Yes, excellent point! I originally omitted this detail for simplicity, but on reflection I don't think it actually achieved much in the way of simplifying the rule so I changed it to reflect reality. Thanks for pointing that out.

EDIT: the rush of people offering up sieve optimizations is pushing me back towards formulating the rule of thumb on a consecutive block of numbers, since it makes it very clear that these are not included, rather than implicitly or explicitly including some subset of them (implicit is bad because opacity, explicit is bad because complexity).

> The prime-counting function approximation tells us there are Li(x) primes less than x, which works out[5] to one prime every 354 odd integers of 1024 bits.

Rule of thumb: Want a 1024-bit prime? Try 1024 1024-bit candidates and you'll probably find one. Want a 4096-bit prime? Try 4096 4096-bit candidates and you'll probably find one.

The approximate spacing of primes around p is ln(p), so ln(2^1024) = 1024*ln(2), and ln(2)=0.693 so if you are willing to absorb 0.693 into your rule of thumb as a safety margin you get the delightfully simple rule of thumb above. Of course, you'll still want to use a sieve to quickly reject numbers divisible by 2, 3, 5, 7, etc, and this easily rejects 90% of numbers, and then do a Fermat primality test on the remainders (which if you squint is sort of like "try RSA, see if it works"), and then do Miller-Rabin test to really smash down the probability that your candidate isn't prime. The probabilities can be made absurdly small, but it still feels a bit scandalous that the whole thing is probabilistic.

EDIT: updated rule of thumb to reflect random candidate choice rather than sequential candidate choice.

Oh, he only busted the Great Depression, won WWII, built half of the infrastructure that we keep kicking the expiration date on, and negotiated 80% of the beneficial fine print in your employment contract. Don't you think he could have done a bit more?

My list would be: 1. FDR, 2. Carter, 3. Teddy. Carter because he sacrificed his career to fix inflation (Republican attempts to rewrite history notwithstanding), and Teddy because he wasn't merely an excellent man with excellent politics, but also because whenever present-day Republicans try to claim the man without claiming his politics I can turn it into a teachable moment, and putting him on a list with the other two is the perfect bait.

I love cheap and reliable TP-Link routers as much as the next guy, but it's definitely also a security issue. The CCP almost certainly has a backdoor. Maybe a respectable one in the form of an undisclosed bug or the ability to lean on an update provider, but the point stands: it's absolutely a security issue and denying this is cope.

Routers are going to be a bit more expensive and a bit less reliable for a while. We'll live.

I thought Shor's algorithm could attack ECC too and the lattice crypto with the sci-fi crystal names (Kyber and Dilithium) was the response?

If I go to https://www.google.com using Chrome and Inspect > Security, I see it is using X25519Kyber768Draft00 for key exchange. X25519 is definitely ECC and and Kyber is being used for key encapsulation (per a quick google). I don't know to what extent it can be used independently vs it's new so they are layering it up until it has earned the right to stand on its own.

Cool, I hadn't run into it before so thanks for introducing me!

I was going to include the digits for comparison, but yes, on second thought 6002 digits is probably too much for polite inclusion in a HN post.

The 50 000 000th prime is 982451653, but fun fact: you may have already memorized a prime much larger than this without even realizing you have done so!

2^255-19

This is where Curve 25519 and the associated cryptography (ed25519, x25519) gets its name from. Written out, 2^255-19=57896044618658097711785492504343953926634992332820282019728792003956564819949.

> There's already a panoply of CUDA alternatives

Is there?

10 years ago, I burned about 6 months of project time slogging through AMD / OpenCL bugs before realizing that I was being an absolute idiot and that the green tax was far cheaper than the time I was wasting. If you asked AMD, they would tell you that OpenCL was ready for new applications and support was right around the corner for old applications. This was incorrect on both counts. Disastrously so, if you trusted them. I learned not to trust them. Over the years, they kept making the same false promises and failing to deliver, year after year, generation after generation of grad students and HPC experts, filling the industry with once-burned-twice-shy received wisdom.

When NVDA pumped and AMD didn't, presumably AMD could no longer deny the inadequacy of their offerings and launched an effort to fix their shit. Eventually I am sure it will bear fruit. But is their shit actually fixed? Keeping in mind that they have proven time and time and time and time again that they cannot be trusted to answer this question themselves?

80% margins won't last forever, but the trust deficit that needs to be crossed first shouldn't be understated.

Haha, yep. Under construction!

Jargon and shared context are barriers for newbies. In a new field they simply don't exist yet. The avenues for accidentally excluding people (or intentionally but I like to be charitable) don't exist yet.

> Could have done without the first paragraph. :)

Sure, let's turn down the temperature.

> youre being flippantly dismissive to the point of shading instead of illuminating

Let's turn down the temperature.

> I wouldn't be excited to have either Australia ... or Chile ... as my sources

China is even less excited than you are to have Australia and Chile as their sources. Eliminating a small pain from the USA and a big pain from China gives a net benefit to China, so it's weird to see it advertised as a net benefit for the USA.

That said, raw material availability isn't the limiting factor here. We probably shouldn't even be discussing it.

> requires discussion about new battery chemistry supply chains to only discuss lithium-ion (??)

That's the alternative Na-ion has to beat. We could build lithium refining and manufacturing capacity in the US sphere of influence. Evaluations of any new technology should compare it to the best available alternatives, yes?

Lithium Iron Phosphate (LFP) has been ramping for a decade, with "no cobalt" as a selling point. It's not huge in the US, but it is in China.

https://www.isi.fraunhofer.de/en/blog/themen/batterie-update...

Why did you spend 200 words arguing for the premise of my second point? Usually one spends time arguing against your opponent but ok, sure, I'm glad we both prefer refining and manufacturing in friendly countries.

> [the commencement of mass production of Na batteries in the USA] refutes the core premise of your statement

How so? This plant is one tiny step on a very, very long road. I'm glad to see it happen, but extrapolating the outcome of a race from the first few steps would be incredibly foolish. China can build Na-ion too, so the question becomes whether the difference in chemistry creates an advantage for one party or another.

> unlike Li batteries, the materials are trivially sourced domestically

So China can cut out the only part of the supply chain that leans towards the US sphere of influence, while the US gets a discount on shipping? This isn't the own you think it is.

Yeah but do you have a gas station at home so it's "always full"? That's really nice ;)

Is charge time important because you can't install a L2 charger at home / apartment to keep it "always charged" or because your usage pattern is too heavy duty for a 30 minute break every 200-300mi?

If it's just the former, the slow steady march of EV mindshare might solve your needs before the "L4" super-fast-charging battery. I am starting to see L2 chargers pop up in apartment parking lots, for example. IMO, "always charged" is significantly more convenient than short stops at a station, so it would still be desirable in a world where "L4" batteries and stations were common.

"Mostly from Australia and Chile" seems like the opposite of baggage, that sounds like about the best you could hope for in a global commodity, so many of which come from unstable regions, conflict regions, or outright adversaries.

Yes, I know that China does most of the refining/assembly, but that has little to do with the chemistry. "Building new Na-ion capacity outside China" is probably even harder than "building new li-ion capacity outside China."

> Without a direct connection to Jensen, there is a 50+ week lead time on Infiniband

Oof, that's new to me, sounds like you're between a rock and a hard place. Here's hoping you get the wrinkles ironed out and make it to the other side, the ecosystem will be certainly be better for it.

Cheers.

> It is rife with firmware bugs, limitations, and hardware failures.

Is it? I've been out of the space for a while but I have only seen firmware bugs/limitations/failures rise to strategic importance with AMD installs, which is why the space has such a strong preference for NVidia despite spec sheet cost disadvantage.

> Our first offering is the AMD MI300x instead of Nvidia or Intel products.

Oh no. I'm sorry.

The other day, the Synchrony chatbot was able to remove a fee that they had previously agreed to remove (delays on their end created a late fee on my end). I was shook.

But yeah, 99% of the time the bots are as useless as IVRs. "Please listen carefully as our menu options have changed. For quality assurance, your call may be monitored or recorded."

Rebecca Tinucci started showing up above Elon in listicle popularity contests. Investors loved her, the industry loved her, the media loved her. She wasn't yet a challenger for CEO, but she could have become one if investors soured enough on Elon.

Ford bought a newspaper to complain about the Jews, and just wait until you hear about the VW guy...

One type of foundational infrastructure got a few years ahead of its applications? Oh, the horror. The horror!