Has it? CPUs ran much cooler back then. And gpus didn’t need brackets and special power cables and all that. Water cooling was harder. But also way less common or necessary.

CPUs now throttle to keep their temp in check. Back then, if the temp got too high they’d be permanently damaged like in the example above.

> Water cooling was harder. But also way less common or necessary.

Water cooling isn’t necessary today, either. The heat pipe heatsinks we have available are so good that you can cool every consumer CPU with a good air cooler. The automatic throttling means it won’t be a disaster if the temp gets high. Water cooling might get a few 100 more MHz during sustained all core workloads but it’s not necessary.

My Athlon 1100 was barely a third tdp at 60watts, yet was perilous to install, broke one (almost $1k, in 2001 money!), and then on the replacement had a 6500rpm screamer jet engine under my desk sitting on mobo that barely supported it and which could fail and kill my 1core cpu at any point :0

25 years ago I never had to wonder if my GPU or CPU cooler will fit the case, clear the RAM, or the HDD cage, if I have something to prop the GPU against so it doesn't sag, if the PSU power cables are fine or I need some adapter and if the adapter can cause a fire if not installed perfectly, if I can access the SSD slot after installing the GPU. That's before having the components in hand.

Installation brings new challenges, LGA sockets are fragile and generally unrepairable, the need for massive coolers leaves little room for maneuver in there, I need to replace the backplate and the general installation procedure is a lot more involved, many of the good thermal pastes are a pain to apply, and so on. The BIOS/UEFI is a jungle now but one you have to navigate to avoid issues introduced by manufacturers pushing setting to the limit in the default profile.

Nvme drives just plug in on mobo and take zero space. I don't have to navigate and bend and fit massive ribbon AND power cables from one end of case to another. 3-4 Nvme fit onto mobo basically invisibly. Want more - cheap easy riser card.

3-4 3.5" PATA drives with ribbon AND power were just fun times / exercises in masochism! And of course different types of ribbon cables, and pins on hard drive begging to be bent, and master / slave hardware settings, and some cables could burn things if plugged in the wrong way around and you had to pay attention that red line on cable aligned with tiny pin 1 on drive. And after a while or if you bend it too much that ribbon cable will give you fun intermittent errors.

Power supplies came with random assortment of built in cables you had to tuck whether you used them or nit. Now you pick and plug in cables you need.

Separate audio card and modem and network card, so definitely had to worry if I had enough Isa slots etc. Now mobo does it all. Maybe even a worthwhile integrated gpu - my 8845hs with 780m does ridiculous amount of gaming. Compared to a 2d card PLUS a 3d card and pass through cables between them - did we forget about those?? :-)

Absolutely did have to worry if my after market cpu cooler will fit back then. Perhaps with stock coolers ones we didn't but I think same is case today.

And I'm just not ready to accept the argument that thermal paste or BIOS were somehow better in 2001 without some solid evidence :-).

You don't have jumpers, you have confusing connectors (mPCIe/M.2/miniSATA differentiated by keying), you don't have a thick ribbon, you have a 3 slot graphics card blocking the SDD. Reaching the GPU PCIe retainer clip on most MoBos is a terrible experience with modern (massive) GPUs. Reconnecting the GPU power cable comes with renewed risk of meltdown. These aren't just esthetic issues (cables look bad, inflexible, hang around, etc.).

Yes, pins bent if you were really careless, pulled cables sideways, etc. And you'd straighten them and keep going, even with CPUs. They were more resilient. A bent LGA socket pin today probably kills your motherboard, and might take the CPU with it. Signal integrity and high currents don't leave room to (literally) wiggle.

> Absolutely did have to worry if my after market cpu cooler will fit back then

Weird, here [0] is a review of coolers of the time, I cannot remember a single time this was a concern, maybe short of a few fancier coolers in very special tiny cases. Today many cases put this on the label, gives away that it's a challenge for many.

> And I'm just not ready to accept the argument that thermal paste or BIOS were somehow better in 2001 without some solid evidence

I didn't say better, they were simpler/easier. The BIOS of 25 years ago had a handful of settings. A modern UEFI config and the amount of settings exposed to the users is crazy. Worse yet, today the defaults are no longer guaranteed aimed at safe long term operation, they're pushing for good bench results. They come set with high power limits, boost levels, voltages, etc. and this will bite the user later in longevity and stability.

As for the paste, reviews added a score for "ease of application", this sends a message. Liquid metals, many modern versions of classic pastes, PTM sheets, there are way more options that are a pain to apply or clean, and in exchange you get top performance. Because there are way more CPUs/GPUs that benefit or demand this.

[0] https://www.tomshardware.com/reviews/Can-t-Touch-This-A-Comp...

Well, it kinda depended on the CPU. And then sometimes the motherboard.

On the Intel side, P3s had a pin on Socket 370 to signal an immediate halt/shutdown. You could probably still damage the CPU, and I'm willing to bet a lot of the cheaper motherboards wouldn't respect it. But plenty did, since S370 needed other changes to support Coppermine anyway.

On the AMD Side, Athlon XPs and Morgan Durons did add a safety mechanism similar to the Pentium 3, but since the Thunderbird and Spitfire Durons did not, it was far far less supported.

And this wasnt the only mechanism. Before THERMTRIP# triggers there is also diode based temp sensor embedded in later CPUs (THERMDP, THERMDN) Mobo designer can use to trigger either STPCLK# directly

Pentium® III Processor Active Thermal Management Techniques http://notes-application.abcelectronique.com/027/27-46150.pd...

or southbridge THRM# pin which in turn actuates build-in ACPI interface with programmable Thermal Duty Cycle CPU throttling (THRM_DTY using STPCLK#) in 12.5% steps (6% steps in later revisions of ACPI interface).

82371AB PCI-TO-ISA / IDE XCELERATOR (PIIX4) datasheet https://theretroweb.com/chip/documentation/29056201-64f6fb6f...

AMD didnt have thermal sensor on CPU Die in early CPUs, they had an app note requiring Mobo makers integrate external sensor under socket. No one bothered, not even good brands like Siemens, and we got famous Toms Hardware article with video of AMD CPU burning into a crisp after taking off radiator on Siemens mobo.

Mobos and cases also seem better designed these days. Yes some components need more stuff than before but it all seems to plug in and slot and wrap and screw in nicely if you're patient and methodical. Heck even the cases themselves are somewhat less likely to slot your wrists :).

I feel ~2003 or so was the cusp when I started seeing less builds that were just maddeningly messy and sharp and hyper loud and crazy, and started getting more predictable and sane and workable.

All anecdotal and based on personal perception of course :)

These days I slap the CPU to the motherboard and a giant heatsink with two huge fans and call it a day.

Fair. I have a scar on the knuckle of my index finger from breaking off the cover on a 5 1/4” drive bay. I was trying to push it out and it suddenly gave. My finger slammed into sharp tin and cut deep.

I’m glad they don’t make them like that anymore.

That's because modern CPUs are thermally limited, so they are essentially designed to stay at their maximum temperature and regulate speed and voltage very quickly in order to achieve that.

I've not had as much experience with AMD CPUs, but have seen several times where the stock Intel cooler's mounting mechanism either wasn't fully latched from the beginning or worked itself loose over time, such that the heatsink was never in contact with the CPU, and the computer continued to operate normally, albeit much slower than it should --- sometimes for years --- with the CPU sitting at its maximum temperature and throttling all the time.

GPUs could be spooky hot. While the cards themselves didn't seem to care at all, the heat emanating from a 3dfx Voodoo3 card (I've had a 2000 and a 3500TV) could be enough to make other nearby cards stinky. Fan brackets and other hacks, from now long-defunct small companies like 3dfxcool, were pretty common.

We had PCI, then AGP, then PCI Express -- with overlaps. Woe be to those who errantly picked the older standard for their shiny new build; they would forever be stuck with their new hardware on the older bus, or with an even-deeper hole in their wallet.

There were other unfriendly things. RAM compatibility was weird; I remember a time around the turn of the century when double-sided RAM became inexpensive enough that was cheaper to buy 256MB of double-sided RAM that the PC may only use half of, than to buy 128MB of single-sided RAM.

Cables weren't always keyed, and when they were keyed it wasn't always in useful ways. Plugging a floppy drive in backwards was a common experience. A 40-pin IDE cable might have a key pin blocked off, while the hard drive had all 40 pins populated, and the two wouldn't fit together without modification.

Burning CDs was often a slow-moving disaster, involving feeding expensive blanks into an expensive drive installed in the machine every half-hour just to hope that things would work this time so you could finally listen to some of those MP3s in your Ford after you spent hours downloading them.

Unless you were foolish enough to buy an external drive, instead: As USB was attrociously-slow, those burners usually plugged into the printer port in what could be most-charitably described as an awful fucking mess.

And even if we weren't burning CDs, we had that pesky analog cable to deal with -- neither end of which was necessarily standardized -- so we could play Total Annihilation and listen to its (rather excellent) soundtrack at the same time.

We still had things like sound cards and modems and IRQs to deal with. NICs were often separate from motherboards. BIOS flashing felt like a dark and dangerous art. The fan headers often had different shapes, and usually only some of them were capable of ramping fan speeds.

At one point around that time, we rather quietly shifted the main power draw for new motherboards from the 5v rail to the 12v rail. Existing power supplies weren't always ready for this and it led some down expensive and unexpected paths.

I enjoyed working with PC hardware back then. But it was all pretty unfriendly, and I haven't even mentioned yet how much fun SCSI was to deal with.

Nowadays: We buy a fish tank and install a motherboard and power supply into it. The smart builder has already got the NVMe drive, CPU, and cooler fastened down into place in their purposeful mounts before this point. The fan headers all work. If we saved enough pennies, we add a GPU -- and it slots into the same x16 PCI Express socket that we've been using for over 20 years now. The NICs (often a plurality of them!) are built-in. There are no dip switches or configuration jumpers.

All that's left is to plug it in and install an OS. Easy.

I still remember, this causes the drive light to stay on all the time, but it didn't cause damage.

It died instantly.

Trying to run LLM models somehow makes 128 GiB of memory feel incredibly tight. I'm frequently getting OOMs when I'm running models that are pushing the limits of what this can fit, I need to leave more memory free for system memory than I was expecting. I was expecting to be able to run models of up to ~100 GiB quantized, leaving 28 GiB for system memory, but it turns out I need to leave more room for context and overhead. ~80 GiB quantized seems like a better max limit when trying not running on a headless system so I'm running a desktop environment, browser, IDE, compilers, etc in addition to the model.

And memory bandwidth limitations for running the models is real! 10B active parameters at 4-6 bit quants feels usable but slow, much more than that and it really starts to feel sluggish.

So this can fit models like Qwen3.5-122B-A10B but it's not the speediest and I had to use a smaller quant than expected. Qwen3-Coder-Next (80B/3B active) feels quite on speed, though not quite as smart. Still trying out models, Nemotron-3-Super-120B-A12B just came out, but looks like it'll be a bit slower than Qwen3.5 while not offering up any more performance, though I do really like that they have been transparent in releasing most of its training data.

There's also the Ryzen AI Max+ 395 that has 128GB unified in laptop form factor.

Only Apple has the unique dynamic allocation though.

I'm not sure what exactly you're referring to with "Only Apple has the unique dynamic allocation though." On Strix Halo you set the fixed VRAM size to 512 MB in the BIOS, and you set a few Linux kernel params that enable dynamic allocation to whatever limit you want (I'm using 110 GB max at the moment). LLMs can use up to that much when loaded, but it's shared fully dynamically with regular RAM and is instantly available for regular system use when you unload the LLM.

I configured/disabled RGB lighting in Windows before wiping and the settings carried over to Linux. On Arch, install & enable power-profiles-daemon and you can switch between quiet/balanced/performance fan & TDP profiles. It uses the same profiles & fan curves as the options in Asus's Windows software. KDE has native integration for this in the GUI in the battery menu. You don't need to install asus-linux or rog-control-center.

For local AI: set VRAM size to 512 MB in the BIOS, add these kernel params:

ttm.pages_limit=31457280 ttm.page_pool_size=31457280 amd_iommu=off

Pages are 4 KiB each, so 120 GiB = 120 x 1024^3 / 4096 = 31457280

To check that it worked: sudo dmesg | grep "amdgpu.*memory" will report two values. VRAM is what's set in BIOS (minimum static allocation). GTT is the maximum dynamic quota. The default is 48 GB of GTT. So if you're running small models you actually don't even need to do anything, it'll just work out of the box.

LM Studio worked out of the box with no setup, just download the appimage and run it. For Ollama you just `pacman -S ollama-rocm` and `systemctl enable --now ollama`, then it works. I recently got ComfyUI set up to run image gen & 3d gen models and that was also very easy, took <10 minutes.

I can't believe this machine is still going for $2,800 with 128 GB. It's an incredible value.

I've been a long-time Apple user (and long-time user of Linux for work + part-time for personal), but have been trying out Arch and hyprland on my decade+ old ThinkPad and have been surprised at how enjoyable the experience is. I'm thinking it might just be the tipping point for leaving Apple.

What do you mean? On Linux I can dynamically allocate memory between CPU and GPU. Just have to set a few kernel parameters to set the max allowable allocation to the GPU, and set the BIOS to the minimum amount of dedicated graphics memory.

Apple has none of this.

Setting the kernel params is a one-time initial setup thing. You have 128 GB of RAM, set it to 120 or whatever as the max VRAM. The LLM will use as much as it needs and the rest of the system will use as much it needs. Fully dynamic with real-time allocation of resources. Honestly I literally haven't even thought of it after setting those kernel args a while ago.

So: "options ttm.pages_limit=31457280 ttm.page_pool_size=31457280", reboot, and that's literally all you have to do.

Oh and even that is only needed because the AMD driver defaults it to something like 35-48 GB max VRAM allocation. It is fully dynamic out of the box, you're only configuring the max VRAM quota with those params. I'm not sure why they choice that number for the default.

On Windows, I think you're right, it's max 96 GiB to the GPU and it requires a reboot to change it.

Only Apple and AMD have APUs with relatively fast iGPU that becomes relevant in large local LLM(>7b) use cases.

Buses are mass transit. The real goal isn't serving poor people, but moving people with higher throughput than it's possible by cars individually (a single bus fits ~50 people). If you make bus lines slow and fail to attract significant numbers of passengers by forcing buses to serve every whatabout case, you're making them fail at their primary goal.

You can't make half-pregnant public transit. If you have a congested city, and just add nearly empty buses sitting in traffic and blocking lanes at every intersection, it will be strictly worse for everyone. OTOH if you can make buses an attractive option, then each bus can take 30+ cars off the road, leaving room for dedicated bus lanes, more buses, resulting in faster and more regular service.

I would agree with and extend your remarks that we also have problems where traffic patterns and geography don't match political boundaries and transit is traditionally locally run and locally budgeted.

So in the USA you end in scenarios where it takes 20 minutes to drive 20 miles but a bus would take four legs with three transfers across three separate city bus companies, figure at least three hours each way. And again, as per your "mass transit" you can't expect taxpayers in my city to provide a special bus run into my neighboring adjacent city much less the city next to that one.

This results in people being very happy indeed to pay the financial and environmental costs of car ownership to avoid sitting in a bus for six hours of daily commute.

There are also interesting social issues; if you're late its a personal failing, even if you take mass transit. I recall a friend at work getting fired because the bus was late too many times. Oh well, should have bought a car. The feeling of not being in control is further worse due to crime rates. No one will sneak up on my wife and stab her in the neck in her car, but it certainly happens on buses and no one cares if it happens depending on local race relations. None of the other passengers on the bus even cared, for racial reasons. Its pretty messed up here.

Its easy for the public in general to advise others to do inconvenient or career ending or life threatening activities, to "save the planet" or whatever, but I wouldn't do it, and I'd certainly never let my wife or kids do it, so we own cars and avoid public transit at all costs. Not taking that advice as been pretty nice so far.

The authors get mixed up equating count of marked stops with dwell time. Running leapfrogging vehicles , or numerous other strategies, reduces dwell time because one boards passengers and the other disembarks at any given stop or vice versa.

In fact, I’d argue bus fare gates, steps, 1-door loading and traffic signal/stop interactions are far more significant than stop count.

How exactly does that help? If you’re suggesting every bus go to alternate stops leapfrogging each one in the middle then that will cause a lot of confusion especially for tourist heavy cities.

So if we had a bus stop every 800 feet, on average half the stops would be empty and passed by. If that high level of use is causing too much congestion and slow down at stops, if we had two buses running out of phase, pax arrive at the same rate, so we'd pick up a pax every 3000+ feet driven. So if we had bus stops every 500 feet to keep people happy, on average the bus would drive right by about 5 out of 6 empty stops, which seems reasonable and would not result in unusual delays or congestion. Also the bus would pass by every half hour not every hour, which would probably increase ridership a lot.

So if the only labor expense were the $23/hr driver, and we pay 10 drivers on 10 routes, to drive twelve times, thats $23/hr * 10 routes * 12 hours if everything except driver labor were free that means we spend $2760 per day to transport 1556 people, or about $1.77 per trip (assuming diesel is free, buses never wear out, etc). If we doubled the number of bus that would be $5520 of driver labor to move 1556 people per day or $3.55 cost per pax trip. On one hand the actual annual total "OE per UPT" counting weekends and maint and office people and dispatchers etc, according to the annual report is $13.94, so an extra $1.77 would seem cheap, but the bus does not run for free and the total expense of doubling the runs might cost as much as an extra $14 per pax trip.

The costs don't really matter, if the taxpayers want it as a luxury bragging feature of the city. Everyone wants everyone else to use it even though no one would be caught dead actually using it. My point being that adult fare is $2 but adults don't ride its mostly elderly and disabled at the $1 fare, so a profit (loss) ratio of (28 - 1)/28 with two buses per route isn't much worse than (14 - 1)/14 with one bus per route.

Maybe another way to look at the analysis is in my city if the stops are more than 1600 feet apart there will be multiple people per stop and that would "slow things down" whereas a small fraction like 400 feet would mean the bus mostly just speeds by.

No one can seem to explain why we can't have infinite bus stops. How about every stop sign is a bus stop? The bus has to stop anyway. Artificial scarcity to drive down ridership, I suppose.

Then you end up with buses piling up at stops, which slows everything down even more.

As an aside, I was thinking about something similar to this tool for a while now after seeing this post (https://news.ycombinator.com/item?id=46723990) where someone was using Claude to troubleshoot a PCAP. It made me think that it would be nice just to have a nice collection of tshark one-liners to quickly weed out any weird stuff right off the bat. I would assume that it would be a lot more performant than using a LLM and more scalable if you have large PCAP files.