Hacker Newsnew | past | comments | ask | show | jobs | submit | fsh's commentslogin

The original iPhone was genuinely terrible. A 2006 Nokia could surf the web on the go and tell you where you are. The iPhone could do neither, since Apple did not include a 3G modem or GPS. It also did not have any apps, and one of the key features highlighted by Steve Jobs was voicemail. The 3G one year later was the first truly usable iPhone.

> and one of the key features highlighted by Steve Jobs was voicemail

Visual voicemail was, and is still a fantastic feature, and phones without it existed for an embarrassingly long time afterwards. I don't remember the last year I had to dial a special number and type in my password, in order to get my voicemail read off to me one at a time in order, but... it was not a small number year.


Coal is much cheaper than gas.

Is it cheaper per MW of generated power? I thought that the main reason use of gas has increased so much (for power generation) over the past 20-30 years is that gas became cheaper.

It depends on the region. The US has shale gas which is genuinely cheaper than coal. Europe doesn't want to use coal for political reasons. China and India barely use gas since it can't compete with coal.

Gas allows to use combined cycle gas turbines (CCGT) which is more efficient and it makes gas cheaper for electricity generation.

That's still not enough to bridge the gap between coal and gas. See for example this article:

https://www.ffe.de/en/publications/merit-order-shifts-and-th...

The US is an anomaly due to shale gas:

https://energyathaas.wordpress.com/2026/05/11/how-much-has-s...


Gas prices vary across the world, mostly depending on if it's only LNG imports (high prices) or local sources (much cheaper, e. g. the US) or a mix (e. g. Europe but since 2022 it's heavy reliant on LNG which pushes prices up). Coal prices also vary. Guess we cannot compare gas and coal prices globally, only withing a specific region.

Not per-MWh in North America.

No, AMD wants to collect some rent from people running Vivado in CI/CD environments.

Vivado is an IDE for programming AMD FPGAs. One cannot use it without buying AMD hardware.

Hardware isn’t where the margins are, and probably is somewhat of a loss leader for small-batch users; for hobby users I would hazard a guess that they’re running at around -10% profit on small sales to try and drive subscription revenue multipliers, and for already-paying users this change is essentially irrelevant and will have zero downside impact on sub revenue. Terrible way to run a profitable business if you fuck up the hardware undercut, but if you can get away with it, subscriptions are certainly a valid answer to maintenance of the platform over time. I still think they didn’t go far enough to make a meaningful dent in conversions from free to paying, though.

(Note that mention of Steam Linux is not about the games aspect, but about the Valve’s seeming plans to become a competitive target for Linux support to the exclusion of other consumer-focused miscellaneity. But I tend to go on about this too often, and shouldn’t have invoked it here, apologies.)


This is not the case here as the software is required to use the hardware they’re selling at any quantity. The software is cost entirely for them, if you’re not buying the hardware you’re not using the software. Given they support Linux for the paid version, its development is already paid for. Absolutely say you won’t provide support for free tier users. Today’s free tier users are tomorrow’s purchasing managers. FPGA is not a big market, so you have to capture comparatively few people for each unit of market share. Good silicon without good software is just very expensive sand.

I doubt that. Dev boards are often not very high margin despite their costs, but absolutely the majority of their profit is from hardware sales, not software licenses. Small volume customers are a combination of a long tail and a loss leader for a marketing pipeline, and FPGAs are almost by definition something where you can't ignore that part of the market (because you only use an FPGA if you can't use anything else and you don't have the volume/margin to justify an ASIC, so it's a niche of niches).

These chips scale up into the price range of $100,000 per chip, I'm not kidding. You really want someone to pick your $100k chip instead of Intel's $100k chip. A single high-end deal offsets the entire tool sales over the entire lifetime of Vivado.

It might be excusable that they want to vet their customers receiving the tool chain for the high end chips to avoid leaking trade secrets to Intel, but that isn't excusable for the low end. Someone who starts with your $10 chips is likely to develop brand loyalty and if they need $100k chips later, they'll be more likely to pick your ones.


Lol, AMD FPGA's the same chip can cost you from $2000 to $10000 depending on which channel you buy it from, or what relationship you have with AMD. Don't be ridiculous. This is an extreme margin business. And especially small time buyers are completely skinned and trimmed of all fat here. That's why many hobbyists are using unsoldered/re-used chips from accelerators or discarded smart network cards, or other second-hand whatever, repackaged in china by smart hands.

> Hardware isn’t where the margins are

Baseless speculation

> probably is somewhat of a loss leader for small-batch users

Wrong. AMD/Xilinx doesn't sell devices directly to customers, they sell them to distributors in huge quantities. Those distributors then sell them to "small-batch users", and they're not involved with AMD/Xilinx free-tier software at all.

> they’re running at around -10% profit on small sales to try and drive subscription revenue multipliers

More baseless speculation


> More baseless speculation

Your elided quote removes the five words where I declared my views as speculation openly and in plain language. The complete sentence that you misquoted opens with that:

> I would hazard a guess

I’m perfectly content to be wrong at HN; it’s a forum where we all have opinions and people rarely restrict themselves to exclusively their own expert subjects, or else we’d all never learn anything! So I will be considering the arguments made here by others before engaging with this topic in the future.

Edit: https://news.ycombinator.com/item?id=48256417 from a few days ago apparently makes some of my points much more clearly, or at least with less hostile replies. I wish I’d found it sooner, but I didn’t realize this entire post was a dupe in time to go back through its comments in detail. Would have saved me commenting at all! Ah well.


The roundtrip time is measured and compensated. Even NTP does this. Knowing the distance is not necessary for time synchronization.


The users are paying plenty of money for AMD FPGAs.


This is terrible news for university users trying to professionalize their FPGA development with CI/CD. Which is probably the point of the change.


I don't get the point. The model has presumably been trained on all public GitHub code, so the evaluation is tainted anyway.


A couple of days ago there has been another thread about an experiment with many LLMs, where especially the Anthropic models were found to "cheat" in a large percentage of the coding tasks that had been benchmarked, by searching the Internet for appropriate code and inserting it in the program they had to write.

The conclusion of that study was that when benchmarking LLMs for coding ability, they should not have access to Internet, if you want to know their intrinsic abilities.

Moreover, this can be worrisome as a more direct copyright infringement than the one caused by training, because even if they find open source code on the Internet and they insert it in the generated files, it is pretty certain that it must have had a license that prohibits the removal of the copyright notice.


> A couple of days ago there has been another thread about an experiment with many LLMs, where especially the Anthropic models were found to "cheat" in a large percentage of the coding tasks that had been benchmarked, by searching the Internet for appropriate code and inserting it in the program they had to write.

Can you find the thread?


I have found it:

https://news.ycombinator.com/item?id=48045174

The study paper:

https://arxiv.org/abs/2605.03546

Look at Table 3, where the cheating rates of Claude Sonnet, Claude Opus and Gemini were between 20% and 36%, during the coding benchmarks.


swe bench pro has a public and private test set, where the private eval is from proprietary codebases only


At that point you can just pre-share a key and use AES.


Nope -- that gives neither public-key capabilities nor forward secrecy.


The point is there is no public key capability in BB84 that requires pre-sharing a symmetric key.

You absolutely do get forward secrecy with pre-shared keys. You just need to make the protocol derive the next key with a cryptographic hash function, and deliver the iteration count with the packet so the recipient knows which key is the correct one. This is called a SCMIP or hash ratchet, and it's used e.g. in Signal protocol.

(As implementation details, you'll also want to hash the hash ratchet counter with the key to prevent theoretical loops, and you'll probably want to encrypt the ratchet counter during delivery with static header key, or the very least authenticate it.)


Chernobyl was supposed to be an economically viable means of generating electricity. Comparing a tiny billion-dollar submarine reactor to a power plant simply doesn't make any sense.


The reactors on aircraft carriers have a similar thermal output to many commercial power reactors. The ones on submarines are around a third of that size, about the size of SMRs like NuScale VOYGR or the Xe-100 reactors proposed to be built at Long Mott in Texas.

Chernobyl was supposed to turn low enrichment uranium into plutonium for Soviet bombs. They made design choices that compromised safety to make plutonium production more efficient.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: