This is what it will take to get PC prices back to normal...
Video Statistics and Information
Channel: JayzTwoCents
Views: 986,838
Rating: undefined out of 5
Keywords: pc prices, gpu prices, video card prices, cpu prices, processor prices, why are gpus so expensive, why are video cards so expensive, why are computers so expensive, why are pcs so expensive, why are gaming pcs so expensive, when will crypto currency end, video cards, video card, gpu, nvidia, amd, intel, how to build a pc in 2021
Id: BEQMfF-DKr8
Channel Id: undefined
Length: 16min 10sec (970 seconds)
Published: Sun Mar 21 2021
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.
I would rather take advice from random people on Reddit than Jay.
capacity and demand were always pretty close, as capacity is bloody expensive to add⁵, and capacity sitting unused isn't just not making money, it's actively costing money as it requires a massive amount of water, power, and other resources just to maintain a clean environment. like texas is finding out, once the clean is not maintained, even for a few hours, it takes months to get everything back up and running, get the cleanrooms clean, verify all the tools work and are calibrated, verify your room is actually clean... and then do that for your water and the rest of your supplies.
so we have added a bit of a spike in total demand... but our total capacity has been reduced. between the global plague, an actual shortage of 200mm & 300mm blank wafers¹, and the usa banning chips from several chinese companies², our global capacity shrank, what, like 4-8%? which is pretty huge, considering our demand spiked and set off a cascade of effects like dominos.
keep in mind, when all your shit is taped out and ready to go, you have mask sets, you have your tooling and production lines set to go, and your testing, binning and packaging set to go... once the order is placed it'll take something like 3-6 months before you get chips ready to populate pcbs.
so you end up with a traffic jam slinky spring effect. what happened 4-6 months ago is being felt now.
anyway you look at it, it's a lot more complicated than jay's video, and even if a cryptobubble cryptopops, it'll take months before the effects reverberate through the industry, and not all the effects will be in phase. for example, if we have the cryptocalypse, gpus will get cheap, quick, because the used market with flood with all the ”entrepreneurs³” trying to not default and break their maxed out credit, they won't be contesting for new cards, and people buying new used cards won't be buying new new cards. therefore, all the 3rd party vendors are going to have to drop their insane prices because not much differentiates gpus these days besides bling⁴. however, this isn't going to unscrew playstation and xbox, nor unclog the automotive pipeline, nor anything else... they'll have to wait for the 3-6 month slinky like effect to hit them.
for you usa'ers out there, a good example of this is what y'all went through with toilet paper last year, and how it took time for the shelves to fill up again, and how it happened in waves.
anyhoo, that's krista's 10¢ :)
1: i forget what it was that happened in 2018-9 that kicked the wafer shortage off, but it wouldn't have been too bad if there wasn't a spike in demand due to plague.
2: and the resulting zomgwtfbbq of everyone trying to find out how to shuffle their production around.
3: we need a new takes on ”the emperor's new clothes” for a modern era. maybe ”the entrepreneurs new hobby”, or ”the n-peror's new nft” or some such.
4: let's face it: the days of buying a higher end gpu with better cooling and getting a better overclock and better fps are over. probably about the same with cpus. intel, nvidia, and amd are all binning harder and using decades of overclocker's experience to make their auto turbos run really damn close to optimal for the specific copy of the chip, not just the model.
good cooling is still important for noise, how long you can burst/turbo/whatchamacallit (to an extent), and even a bit of how well your burst/turbo/whatever works, but after a decent cooling rig, everything is going to be within a couple few percent. the days of 150% (or even 20%) overclocks are over :(
5: as well as taking a very long time and a lot of planning. plus, the people like asml and zeiss optical that make the tools for these new fabs are themselves usually running very close to capacity. right now, there's more demand then they can handle, and it's been that way since a bit before all this plague stuff. heck, old fabs are running at capacity, and because the older processes are affordable, well understood, and very mature, 200mm fabs that make mixed mode chips or small batches or sensors... or a lot of things, are booming and that segment is growing, too!. adding capacity is part of out slinky spring model of supply, capacity, and demand, too, but it's a long wave. think a couple of years, minimum, to get a new, not state-of-the-art, fab from idea to paper.
--= warning: sleep deprived digression follows. enjoy (or not) at your own risk.
asml is the company making euv and high na euv equipment. these are perhaps rhe most astoundingly advanced, complex, precise, and accurate machines made by humans... ever. if you want to be amazed, look up ”how does an euv laser light source work”. each euv machine costs around $100 megabucks, and takes months to make. asml can only make a couple dozen per year, if that. they make some modules at their clean factories, and ship the modules to the customer, along with the engineers, technicians, electricians, plumbers, and scientist that it requires to finish making it. although they don't ship those guys, they fly them, of course! maybe even first class.
then these ninjas go into the customer's iso class 1 cleanroom (< 10 particles per cubic meter, < 200nm particles), which is mind-boggling itself, and assemble the $100 megabuck machine on site and calibrate it. hell, getting the parts it in to the cleanroom with a crane and keeping the room clean is... amazing.
fabs are among the most intense science and technology humans can do. they're at the edge, and pushing it. adding capacity at new nodes is... well, there's 2.5 companies doing it right now, out of the entire world.
so because of all of that, and because certain types of chips don't benefit from cutting edge lithography, and because they work and are paid off, older processes that aren't on the edge of possibility are still popular and even gaining in popularity! the price of used chip lithography equipment and 200mm wafer equipment (state-of-the-art is 300mm wafers, possibly 400mm in 2 generations) is on par with what it originally sold for. the price of old chip making tools, like old gpus, is spiking, too. it's to the point where companies are designing new generations of ”old” process machines.
the ”old” processes, 65nm-130nm (iirc, could be wrong, but i'm close) are well understood, have great yields, and cheap to make chips on, and can make small batches of chips easier and faster. the resulting chips are usually more robust because the gates are larger, and can handle more power easier. they're not as fast, and in a lot of cases not as efficient, but even ”old” processes still being used are being refined for desired characteristics, such as low gate leakage, or very precise and repeatable current control. it's these older processes that are leading to chips popular in the iot space and popular with the maker movement. it's these older processes and fabs that use them that are getting small runs of chips on a sub 100nm process into the tens-of-kilobucks price range, and allowing loads of fabless chip design companies to make new and exciting things. like triad semiconductor and valve's vr lighthouse 6dof tracking that uses triad's light-to-digital mixed signal ics.
TL;DW: Another cryptocurrency bubble burst would be a major downward driver, because as it is right now, much of the demand today is not from gamers. But it might be a really long while with just how much backup there is in other industries like automotive and K12 education also wanting a slice of TSMC wafers, and there are just simply not enough wafers to go around.
Its never going to happen. Everyone has already shown AMD and Nvidia that they'll pay outrageous prices for graphics cards.
Every time miners go wild we always ask these same questions and everyone always says it'll get better after miners are finished but instead AMD and Nvidia raise their MSRP every generation.
The only question is whats the highest they'll get to.