You know there's a bump coming when a new generation of entertainment machines comes out (also applies to anything that doesn't update every year like apple ipad mini may cause a bump but it's not nearly as competitive as gaming)
2017 and 2018 were actually good years for electronic component sales, 2019 was bad (especially in the Americas), and 2020 is kind of a mess. My source for this information is an industry report (which I can't share).
sidenote - "Lean and JIT style management of inventories" is one of the operational causes of those supply chain inefficiencies https://en.wikipedia.org/wiki/Bullwhip_effect#Operational_ca... . It is a classic result from the dynamic systems that improving the feedback in a given system may lead to destabilization of the system. Somehow it is lost on the proponents of Lean/JIT/etc.
It is interesting how it has similarity with software development - several years ago i worked on a project that was being done under total Lean/Agile/Scrum regime. Scrum forces significant lead times into the upstream/downstream components chain(graph) like in a supply chain. As a result the whole upstream/downstream components graph relationship on that project got extremely slow (a sprint wait to get new upstream component version, try to use it in this sprint and fail because of bugs, file the bugs, the bugs get fixed the next sprint at best, while other components trying to move forward in their development ) and ultimately completely broke down after about 2 years to the point where the product couldn't even be built as not just functionality (the functionality got really lost months before that), the interfaces got completely out-of-wack without any chance of reconciliation in the setting of that process, and the product was quietly canceled soon as a result.
Willfully ignorant of. I remember when JIT was seen as this revelation due to UPS and other shippers stream lining their processes. So this window opened up and folks were like, "we can collapse our supply chains to UPS/Fedex" ...
This is directly analogous to the buffer bloat problem. Smaller inventories and more responsive suppliers means better communication of intent and supply in the supply chain and therefore greater efficiency and prices. Why are advocates of big redundant inventories willfully ignorant of this?
If we optimize for efficiency, we get supply disruptions every 5 years or so. If we optimize for redundancy, we suffer from higher prices all the time, and delayed availability of all new tech. I don’t know which is worse.
Depends on the item. If it's medical supplies or food, "supply disruptions" can kill people. If it's iPhones or Christmas novelties, it's not a crisis.
Folks keep inventories because the cost of not having X at time Y is not priced accordingly at the supplier. Digikey doesn't care or know if the 0.10$ part is going in a 10M one-off or a 4$ toy.
JIT works for fungible commodities, you suddenly realize how much the cost/volume/downside ratio works when your always-on JIT capacity goes to zero.
I don't want my water treatment plant to go down because a bean counter decided that spare parts are now stocked entirely by McMaster-Carr.
Higher risk strategies are necessary in some industries super competitive industries. You play it save and your competitor takes the risk and they put you out of business before the rare event happens. If you both take the risk, then when the rare event happens you both are screwed equally and aren't out of business.
Combine that with the evolution of an industry (safe players dying out) and it's another dynamical systems thing. Industries get riskier and more coupled and more efficiently tuned to normal circumstances then blow up.
Lean is quite a broad practice that goes way beyond inventory management. Also, the Lean software development methodology is very different from the Lean process management one.
Anyway, yes, all those things are tools that should be used with an understanding of their positive and negative impacts (even Scrum has some positives...), not something undiscerning managers impose despite the realities of the field.
And if you go read Deming, the guy that created most of what goes into Lean (production, not software development) and JIT, he has some very interesting opinions on undiscerning management.
Samsung should be good enough for anything but the super cutting edge (TSMC "5nm"). The Nvidia 3000 series GPUs for example are definitely the most powerful GPUs ever built and Samsung fabbed those.
Given that hopper is supposedly going to be on 5nm, and seeing the M1 performance, I actually cancelled my early 3090 order. I'm going to go with something cheaper and leave my big purchase until next year.
Not really. As the article points out, this is about chips coming out on older (8" being "older" is funny) fab technologies.
There are three major prblems:
1) Nobody holds inventory.
This is the primary root of all problems. Because everybody is on just-in-time inventory, nobody can absorb even the slightest shocks. Maximally efficient is minimally robust.
2) Semiconductor industry consolidation
The semi companies have been on a competition buying spree over the last couple of years. There's no need worry about a competitor holding inventory and winning a contract from you if you've bought them all out.
3) The "minimum volume" of a chip run through a fab is enormous.
An 8 inch wafer is about 32,000 mm^2. If your chip is 2mm per side (pretty big for "popcorn" type chips), a single 8 inch wafer produces 8,000 of them. If it's 1mm per side, you've got 32,000 of them.
You probably won't produce less than 10 wafers at a time.
So, that means your inventory goes from dead 0 to roughly a quarter million in a single whack with an 18 week lead time (and I'm just counting wafer starts--I'm not even talking about testing and packaging resources getting choked). And most projects never reach a quarter million in volume--and the ones that do generally blow WAY past it in a hurry.
It's been a pretty rough year for anyone trying to buy any sort of PC part this year (CPUs, GPUs even PSUs). Had to go stand in line to get one of the 3060 Tis like it was an Apple launch or something.
On the flip side, RAM prices are good and SSD prices are amazing. But yeah, CPUs and GPUs are a gongshow. A lot of people are blaming scalpers, but I think it’s just demand is much greater than supply.
I'm honestly amazed at the PC component market for the past decade. The Thailand flooding that messed up HDD production, and crypto mining jacking up the price of GPUs, are just two examples
Is this actually happening? A bunch of people buying something for cheap and then immediately selling it for what the market can bear isn't cornering the market. For that you need a cartel (or a group of people acting in unison) buying the goods up, then hoarding it to drive up the price.
Scalpers are also proving that companies can charge more. People laughed at the price of the AirPods Max, but they're selling very consistently and quickly on ebay for 750+
I think this is especially clear in nvidia GPUs. It would have been unlikely for nvidia to try doubling GPU prices for fear of the backlash and the risk of losing out if that wasn't countered by enough people willing to buy at the higher price.
But then the bitcoin gpu shortage happened and people were paying double for mid tier GTX 10xx cards. The shortage went away and prices went back to normal, but then the 20xx cards launched with a huge price bump, likely helped by the shortage scalpers proving how much people would pay.
That's likely solely because of supply-side constraints. Trying ordering AirPods Max on Apple's website right now, and you'll be informed of a 3-4 month ETA.
I don't believe Apple would want to charge $750/ea if they had ample supply.
Companies have other concerns than the immediate revenues though.
Consumers are very sensitive to price fluctuations, it might makes sense short term to raise the price when demand outstrips supply, but consumers will remember the companies that "gouged" them.
Apple maybe would be able to get away with it, but for big companies having stable price is pretty important.
I do think logically, if supply is ample (greater than demand) than the return on scalper investment may drop below the point of being worthwhile. Transaction costs and the risk of no longer being the preferred direct retailer could result in scalping losing you money. As long as demand readily exceeds supply, having the in-demand item gives you the power to collect a greater price on that item.
If it's evenly matched, scalping would likely be a minor, potential problem, but not to the scale we're seeing on all of these gaming items.
Scalping disappears if supply matched demand (with perfect geographic efficiency). Buying from a scalper is strictly riskier than buying from a retailer (might not be able to return faulty equipment, etc). There might be exceptions for rare / limited edition items because then scalpers can potentially control the whole market and extract monopolistic profits (assuming the demand curve allows it).
Scalpers only exist when the true market price of something doesn't match the list price (Demand >> Supply). Graphics cards worldwide, microphones, at-home lifting equipment, USD in Venezuela... it's basically Uber's surge pricing but for products :)
They've been pretty much flat for the past 18 months. I've been wanting a 2TB SSD since 2018 and I've been watching the price sit stagnant at ~$250 canadian since june 2019 (dropped from ~$600 in the year before that).
It's just a want, not a need, so I haven't looked any further (just keep the occasional eye on the amazon price). May become a need (or at least, a financially-justifiable want) a few months after the google photos storage limit starts
There are some, but also SanDisk, Dell (rebranded Toshiba), etc...
I've had pretty good luck with everything so far, knock on wood. The 1.92TB drives I mentioned are SanDisk CloudSpeed Eco II, and they have been pretty good. What's really impressed me are the Dell/Toshiba write-optimized 1.6TB drives I picked up. Pricey, but excellent performance.
I have a few I think 240GB Intel drives I picked up as well. They initially had some of the problems you're referring to, which for me was not showing up after a warm reboot, but they haven't done that for a while, so I'm thinking it may have been fixed by an OS update.
I'm humbly trying to upgrade from an i5-4670k to an i7-4790k (both of which are 7.5 years old) and it's a steal if you can find one for $150 + tax used on ebay. Madness.
I just got home from waiting for a 3080, got there over an hour before opening, and they still ran out before I could get one. This is something like >3 mo after release.
I had Microcenter build me a new PC a couple of months ago, and supply was only a problem for the very newest top-of-the line stuff. Everything else was okay. Since I didn't really need bleeding-edge, I had no issue.
I tried to order a custom desktop with a rtx2070 this November from a dozen sellers. Everyone said they had problem with their own suppliers and I should expect to wait until mid January or buy at something like 800 euros for that card. Insane (I just took a xbox series s instead).
That's it. No other RX 5000, no RTX 2000, no RTX 3000, no RX 6000.
So that's two of them don't even have gaming GPUs in stock, as an entire product category, and the others you get some last gen low end cards and one retailer with a 5700XT. It's a little more noticeable here.
They list 5600XT, card slower than Nvidia 4 year old 1070, at 350 Euro. 220 Euro for a 5500, slower than 4 year old 1060, slower than $160 MSRP 1650s. But wait, there are no 1650s in stock, and no 1060, and no 1070. There is 1660 Ti however, $280 MSRP, listed at 410 Euro .... thats almost twice the msrp for a 2 year old product.
The only products they havent run out of are listed at 2x the markup.
I disagree, the latest consoles have a historically weak launch lineup and don't run the full steam library, mods and all. On a compute per dollar comparison, sure they are a fantastic deal, but in terms of gameplay experiences per dollar they don't do nearly as well. They are like locked down budget computers, and it's almost as hard to get them at msrp as gpus too.
I'd say I agree it's a terrible time to build a high-end PC system. But I'm also saying that based on idle searches for components. I bought most of my PC parts in April 2019, and my (mid-range $300) GPU in spring 2020, with no issues with availability.
But it seems like motherboards are 25-100% more expensive now, and we are still waiting on $200-300 graphic cards from this generation.
I do think, though, that comparing a relatively single-purpose item like a gaming console to a general purpose desktop computer is also going to have pros and cons on both sides.
I'd much rather be shopping for mid-range (or last generation but still quite performant) computer parts than trying to get one of the consoles right now.
Here in Europe I'd have to spend about 450-500 on 2060, which is a cheapest previous gen GPU. I reckon that gives you a decent 1080p experience, not more.
A PC is used for much more and has much more expansion capabilities. Not to mention my PC is better than any console even though the GPU costs as much as a retail PS5. If I was on the console upgrade cycle I could wait until PS6 to upgrade my computer. That's a portion of the PS4 cycle and all of the PS5 cycle for ~$1200 (no need to upgrade storage and other accessories for a new build).
People on the high end could even wait 2 console cycles to refresh.
Everyone decided to wait to upgrade till the new gpu's were announced from what I can tell. I upgraded my mobo & cpu in october and the same cpu (ryzen 5 3600x) is about £50 more expensive and there doesn't look to me that many units available.
I just built a new office PC for my lady as a holiday gift. Most components were pretty fairly priced but I paid way more than expected for a Ryzen CPU with the integrated Radeon graphics near the price point I wanted to keep the build at. It's still a great build compared to what I'm replacing but that CPU markup was a shock, and I'm happy to not have to add a GPU to the mix.
This seems like a huge failure to correctly price products if there are actual shortages and people standing in line (except maybe on launch day, that might be good publicity). The gaming market seems very elastic on the demand side, few people actually _need_ a new rig urgently, so an increase in price could effectively limit the initial demand and get them to people who need/want them most while increasing margins for manufacturers or sellers.
As it is now private middlemen take that profit, presumably by standing in said lines wasting a lot of time. And buyers now have to trust some rando on ebay to actually ship a 2k$ graphics card. That is not an optimal outcome.
Companies take a big brand hit if they increase prices to match their low output. Adidas, too, can sell Yeezys for $1000 a pop, but chooses not to. It also helps keep the hype for the products up.
In Japan, a notorious and No1 import agency called "ASK" does this operation for video cards. They sets higher prices at launch (3060Ti/3070 is 30%-40% expensive than US' MSRP) and finally decreases the prices. As a result, now we can easily buy any 3060Ti/3070, and 3080 if you don't choose brand, but expensive. What's interesting here is that sold price in Japan is sometimes far cheaper than scalper's price in US.
Microsoft and Sony should have offered their new consoles using this method. Start at a very high price like $10,000, decrease the price until people start buying, resume decreasing the price when demand is satisfied, and repeat as needed.
About 4 months too late on this. Apparently I will receive my 3070 at the end of January. Also took 2 months to find a Zen 3 CPU. Bought a 5800x simply because it was the only thing I could find.
I picked a poor time to decide to upgrade after putting it off a year.. Great product cycle from a feature standpoint, terrible for stock. Don't believe I'll get either a 5xxx or 6xxx before the new year. Bought a 3950x just to have something in the meantime.
Sounds like you are doing okay, that 3950x is pretty great and will re-sell for a decent amount on the second hand market. Good luck with your upgrades!
I was lucky that Microcenter has a unofficial discord channel. I followed that and was ready to go pick up parts anytime. I was able to snatch a 5900x two weeks ago and then 3070 yesterday.
I have tried online but cannot beat the bots. I would have to write my own.
Chip fabricators are planning 2 years of record spending on capital equipment and fabrication facilities[0]. Note that when a new 'fab' is built, the spend is on the order of $10 Billion. Also note that 85% of the world's capital spend on chip fabs is in Korea, China, Taiwan and Japan. Hence, 85% of the world chips are fabricated in that region.
Notably, the bulk of the equipment that is purchased is made by US firms (about 1/2), with 1/4 from Europe and 1/4 from Japan.
> note that 85% of the world's capital spend on chip fabs is in Korea, China, Taiwan and Japan. Hence, 85% of the world chips are fabricated in that region.
I don't follow the logic. Why would the first 85% imply the second 85%?
That's true, it really doesn't immediately mean that. Over time, should the regional spend patterns persist, it would in fact mean that in dollar terms. Edit: Perhaps should have said 'will be'.
It's difficult to nail down the breakout of production regionally, because, for example, the US has 85% of the global semiconductor sales, but only 44% of that 85% are fabricated in the US. Here is the 2019 breakdown from Semiconductor International Association[0].
In unit terms, it could be quite different since there are so many low-value chips made on legacy equipment that has been depreciated.
> because, for example, the US has 85% of the global semiconductor sales
This is incorrect. I looked at the PDF that you had linked to[0], and it says, on page 14 (of the PDF) that the total sales in the Americas is 22% of global sales. They say "Americas", so that includes Canada, Mexico, Central, and South America. The Asia Pacific region is where 60.3% of all semiconductors are sold to. And China is the largest consumer of semiconductors, at 33.8%.
I believe the 85% refers to global sales by firms headquartered in the US. That includes Intel & TI, but also Fabless, like nVidia, Broadcom, AMD, Qualcomm.
Assuming that fab equipment has a relatively flat price in terms of production capability, it will be true over time. These are the places that purchase 85% the equipment, the equipment produces chips, so these places produce 85% of chips. Of course, some of the equipment will yield higher or lower chip outputs than others, and some of it will be shipped/resold elsewhere, but with large geographic and time scales, the output percentage will tend toward the capital spent percentage.
If you mean the SoC, the latest model is fabricated on a 28nm process. This is a relatively ancient fabrication technique (about a decade old) which is widely available from several contract manufacturers.
The ROI calculations for a nuclear power plant are very different though, in terms of the dollars per gigawatt-hour that will be sold to the grid over many years, for the total expected life span of the power plant.
Usually because they're built with some extreme form of government subsidy, or the fully loaded cost of a nuclear power plant is not taken into account in the $/GWh, such as eventual storage and disposal of nuclear waste at a government-run, government-funded facility.
No, it doesn't include accident remediation or dismantling that plant. So start by finding a radioactive storage facility (oops!), then triple the cost.
Chernobyl and Fukushima will cost about $1 trillion each ... for just the first 100 years.
We'll, it will take a decade, but this is one of the few European projects I believe might actually work. It's a well defined goal (fab chips) and the winners of the contracts is more or less predetermined (AMSL, etc). That's much less fluffy than "we're going to build a system of standards so that Europe can build a federated cloud".
>More fundamentally, however, there has been under-investment in 8-inch chip manufacturing plants owned mostly by Asian firms,
8-inch or more commonly known as 200mm Wafers even in the US, are NOT under-invested. The total Wafer capacity of 200mm has been steadily increasing for a few years. ( I mean Reuters / Bloomberg reported this at the time, although fairly late in the cycle ).
You cant have a perfect Supply and Demand cycles, there will never be one. And especially during uncertainties.
I'm surprised that industry hasn't diversified the geographical locations of important manufacturing like this especially after the great hard drive shortage of 2011 after flooding crippled most hard drive manufacturing.
I think the efficiencies of manufacturing in one country still outweighs the benefits of diversification. I'm not an expert on these matters but it just seems obvious. You'll have to move around materials and there are trade/geopolitical issues at play there. Japan and Korea briefly had a dispute in 2019 and Japan stopped 3 chemicals from going to Korea. This affected their semi production.
That's my point is that it's so concentrated it takes very little to completely disrupt the entire industry. The question to me is are the increased costs of diversification worth the risk of temporary (or maybe permanent) disruptions to manufacturing.
Then again, when I studied political science and economics there was a common belief/theory that this sort of increased interdependency between countries (globalization) would decrease the likelihood of war and increase opportunities for democracy. So maybe it's a good thing.
In my opinion, under ideal circumstances will you have interdependency between countries with everyone sharing a part in the process. What will end up happening is that countries will split into two ideological ‘nations’ like in Axis vs Allies.
Obviously not much has changed because SEA is where the cheap labor is, and it's not too far from China since many rare-earth materials are from there. I wonder if in the coming decades, as SEA's middle class continues to grow (as China's has now), we'll see companies migrate to Africa, the last frontier of relatively cheap labor. We know China has already begun doing that.
Not going to happen. Consumers demand a lower price today and would buy from a competitor if the price is 3% lower. They won't pay extra for the resilience of the supply chain.
Also, the choice of geographies is not that wide. You need a place with lots of reasonably skilled but relatively low-paid people.
I got a flat this summer. Brought my bike in to my local shop to have them fix it. They said "I hope you have a spare because we don't have tires, and it may be two months before we get any". I did have my spare that I had brought in with me. And I felt good about giving the shop some service works because I noticed right away that they had almost no bikes in stock.
On the other hand this year I felt joy of restoring some 13+ years old laptop computers for school as donation. Amazing how cheap new 120GB SSD plus used 4GB RAM upgrade let them run modern OS, the Internet and edu software nicely. There is abundance of second-hand parts on ebay ready for delivery.
Don't know what the OP was doing, but schools operate with different rules for licensing. Some scenarios that aren't possible for individuals or enterprise are possible for schools, depending on the agreements in place.
Mostly Windows 10 - supports even 32-bit CoreDuo, all edu software works, staff knows it. In EU you can buy second-hand software licenses legally. Some schools participate in volume licensing schemes with MS too.
But I also install Ubuntu on some specialized "workstations", e.g. pendrive/Internet terminal laptop connected to beefy old laser copier/printer where CUPS PCL works with zero drivers. Other use case: jukebox laptop for music classes :)
Vast numbers of perfectly servicable computer hardware is thrown out all the time. We would be well served to use this time to get used to reusing and extending the life of the hardware we already have. And, as software engineers, we need to reign in our reckless disregard for performance concerns to ensure that hardware remains useful, longer.
Performance may be improved purely in software (I don’t think this is true) but to improve power use you definitely want to keep throwing out old hardware. We’re not at the end of process shrinks yet and old stuff always uses more power.
I would imagine the extra power use probably isn't enough to outweigh the environmental costs of manufacturing new hardware, at least for your average consumer device. For servers, maybe the higher power usage is sufficiently high for it to be beneficial?
Yes, I meant at datacenter scale. In your own house the power bill/the environment won't really be helped, but newer tech does have better battery life, and that's always nice.
I mean "capable of running useful software". There are vanishingly few use-cases which actually demand the resources of newer computers - 99% of software with high system requirements is due to programmers who have grown fat on the teat of their high-end workstations.
This seems to be an argument for not buying new computers.
I don’t entirely disagree with you. I would say that a lot of compute resources are used to make user interfaces more impressive, but not necessarily better than what could be done if we designed interfaces around lower resources.
However that isn’t what we’ve done.
Older computers often can’t run software that is useful today even if the software they could run when they were new still does the job. Things like file formats and web standards have moved on and are not backwards compatible, and a lot of what people use computers for is communication, so that matters.
Despite all the defenses of such by progammers who claim hardware always improving and code cosmetics/organizational trends obviate software performance concerns, I can attest to the reality that I composed electronic music with completely software instruments and effects on 512mb ram on earlier versions of certain unnamed DAW's with all the "latest" plugins at the time. My last Macbook pro 2011 with 4gb ram on Snow Leopard was sufficient, but once I updated it to Yosemite and went from Cubase 5 to Cubase 9, from Ableton 5.5 to Ableton 10, that it went from "performing just fine" to basically running the fan the entire time and glitching at basic playback of sessions I was previously working on no problem. (same plugins).
The Steinberg Cubase one being the biggest noticable turd, and I blame their acquisition by Yamaha, as the customer-service also went down the tubes, as I waited months to get a reply on a replacement USB dongle (also a sucky thing for a laptop with 2 USB ports, but I digress)
Software bloat kills. It's not all deliberate, but if our values don't change from "who cares" to "we care", we can't expect any change.
Is uncaring mega-corp structure the cause? (programmer wants to go home, project is badly managed, bugs outnumber features, let's start a new project, etc?)
I'm curious what folks think is the cause of poor-quality buggy bloated "modern" software that is demonstrably worse than older software.
It's not an improvement in technology, it's simply that our modern world cares more about superficial animated menu openings than about performance.
Mind you, not ALL software follows this trend. Reaper still kicks A, but that's because Mr. Frankl has a vision, integrity, and still maintains control lol
anyway, about chips and processes, it's not about getting tinier and tinier, it's about an intelligent design.
All the efficiency improvements in the process are wasted if the goal is to make a mega-poo-renderer that renders 0.1 turds per KW of power expended.
Controversial thought of the day:
I will reckon that a simple interrupt driven event-loop on a microcontroller running all HID input is a million times more efficient than the convoluted way a context switching multitasking operating system handles the same under full load with full memory saturation.
Not too surprised. The semiconductor industry have been dominating and buying up all the OEM equipment and hardware. Just get a quote from THK and be appalled at the lead time. The semi industry is extremely cyclical and fairly predictable. But this 5 years have really changed this. It is interesting that the downcycle for the industry never came this time AND it's booming greater than ever. Unfortunately it takes almost 5 years to ramp up the equipment to meet this new demand.
The fire in Japan is kind of a big deal. They turn out to have been the only supplier for that part and the result is that a ton of downstream manufacturers are affected. Your GCM team may have thought they had second sources, but when they're all dependent on a part from this one factory you see them all pull back their supply forecasts at the same time. It's been a trial.
> U.S. sanctions-hit Chinese tech giant Huawei Technologies, a fire at a chip plant in Japan, coronavirus lockdowns in Southeast Asia, and a strike in France.
It’s always a strike in France. Unionists dream state.
146 comments
[ 5.5 ms ] story [ 210 ms ] threadSome people say it was the best one they saw in their entire careers since dotcom bubble.
PS5 and Xbox
You know there's a bump coming when a new generation of entertainment machines comes out (also applies to anything that doesn't update every year like apple ipad mini may cause a bump but it's not nearly as competitive as gaming)
Rolling supply chain disruptions were predicted to happen early this year when China shut down much of its economy due to the coronavirus.
It is interesting how it has similarity with software development - several years ago i worked on a project that was being done under total Lean/Agile/Scrum regime. Scrum forces significant lead times into the upstream/downstream components chain(graph) like in a supply chain. As a result the whole upstream/downstream components graph relationship on that project got extremely slow (a sprint wait to get new upstream component version, try to use it in this sprint and fail because of bugs, file the bugs, the bugs get fixed the next sprint at best, while other components trying to move forward in their development ) and ultimately completely broke down after about 2 years to the point where the product couldn't even be built as not just functionality (the functionality got really lost months before that), the interfaces got completely out-of-wack without any chance of reconciliation in the setting of that process, and the product was quietly canceled soon as a result.
yeah.
If we optimize for efficiency, we get supply disruptions every 5 years or so. If we optimize for redundancy, we suffer from higher prices all the time, and delayed availability of all new tech. I don’t know which is worse.
JIT works for fungible commodities, you suddenly realize how much the cost/volume/downside ratio works when your always-on JIT capacity goes to zero.
I don't want my water treatment plant to go down because a bean counter decided that spare parts are now stocked entirely by McMaster-Carr.
Combine that with the evolution of an industry (safe players dying out) and it's another dynamical systems thing. Industries get riskier and more coupled and more efficiently tuned to normal circumstances then blow up.
Anyway, yes, all those things are tools that should be used with an understanding of their positive and negative impacts (even Scrum has some positives...), not something undiscerning managers impose despite the realities of the field.
And if you go read Deming, the guy that created most of what goes into Lean (production, not software development) and JIT, he has some very interesting opinions on undiscerning management.
Yup, and has been unable to meet nV's demand-- such that this has been accused of being a "paper launch".
There are three major prblems:
1) Nobody holds inventory.
This is the primary root of all problems. Because everybody is on just-in-time inventory, nobody can absorb even the slightest shocks. Maximally efficient is minimally robust.
2) Semiconductor industry consolidation
The semi companies have been on a competition buying spree over the last couple of years. There's no need worry about a competitor holding inventory and winning a contract from you if you've bought them all out.
3) The "minimum volume" of a chip run through a fab is enormous.
An 8 inch wafer is about 32,000 mm^2. If your chip is 2mm per side (pretty big for "popcorn" type chips), a single 8 inch wafer produces 8,000 of them. If it's 1mm per side, you've got 32,000 of them.
You probably won't produce less than 10 wafers at a time.
So, that means your inventory goes from dead 0 to roughly a quarter million in a single whack with an 18 week lead time (and I'm just counting wafer starts--I'm not even talking about testing and packaging resources getting choked). And most projects never reach a quarter million in volume--and the ones that do generally blow WAY past it in a hurry.
Good luck managing that inventory flow.
I always like to tell my friends and family the same thing by saying, "Efficiency is not the same as efficacy"
It saves money to have everything co-located.
It also amplifies the magnitude of physical-world disruptions.
Scalpers just appropriate the profit which the producers could make if they could afford it. Often they cannot because they declare a price at launch.
But then the bitcoin gpu shortage happened and people were paying double for mid tier GTX 10xx cards. The shortage went away and prices went back to normal, but then the 20xx cards launched with a huge price bump, likely helped by the shortage scalpers proving how much people would pay.
I don't believe Apple would want to charge $750/ea if they had ample supply.
Consumers are very sensitive to price fluctuations, it might makes sense short term to raise the price when demand outstrips supply, but consumers will remember the companies that "gouged" them.
Apple maybe would be able to get away with it, but for big companies having stable price is pretty important.
I don't think so... EA still sells a lot of games
If it's evenly matched, scalping would likely be a minor, potential problem, but not to the scale we're seeing on all of these gaming items.
Scalpers only exist when the true market price of something doesn't match the list price (Demand >> Supply). Graphics cards worldwide, microphones, at-home lifting equipment, USD in Venezuela... it's basically Uber's surge pricing but for products :)
They've been pretty much flat for the past 18 months. I've been wanting a 2TB SSD since 2018 and I've been watching the price sit stagnant at ~$250 canadian since june 2019 (dropped from ~$600 in the year before that).
I've had pretty good luck with everything so far, knock on wood. The 1.92TB drives I mentioned are SanDisk CloudSpeed Eco II, and they have been pretty good. What's really impressed me are the Dell/Toshiba write-optimized 1.6TB drives I picked up. Pricey, but excellent performance.
I have a few I think 240GB Intel drives I picked up as well. They initially had some of the problems you're referring to, which for me was not showing up after a warm reboot, but they haven't done that for a while, so I'm thinking it may have been fixed by an OS update.
OverclockersUK: 5700XT, 1050 Ti, 1030
Komplett.ie: GT 1030 and some random Quadro
computeruniverse.net: 1650 Ti, GT 710, RTX 3090 (for $2400)
scan.co.uk: Teslas, Quadros, GT 1030, GT 710, GT 210
That's it. No other RX 5000, no RTX 2000, no RTX 3000, no RX 6000.
So that's two of them don't even have gaming GPUs in stock, as an entire product category, and the others you get some last gen low end cards and one retailer with a 5700XT. It's a little more noticeable here.
Half of the ones you have listed are UK retailers.
I haven't checked the others but computeruniverse.net has various RX 5000 series GPUs in stock. So does mindfactory.de at the moment.
The new RX 6000 serious GPUs have been selling out quickly but they were just released this/last month and new stock does keep popping up.
The only products they havent run out of are listed at 2x the markup.
I'd say I agree it's a terrible time to build a high-end PC system. But I'm also saying that based on idle searches for components. I bought most of my PC parts in April 2019, and my (mid-range $300) GPU in spring 2020, with no issues with availability.
But it seems like motherboards are 25-100% more expensive now, and we are still waiting on $200-300 graphic cards from this generation.
I do think, though, that comparing a relatively single-purpose item like a gaming console to a general purpose desktop computer is also going to have pros and cons on both sides.
I'd much rather be shopping for mid-range (or last generation but still quite performant) computer parts than trying to get one of the consoles right now.
People on the high end could even wait 2 console cycles to refresh.
As it is now private middlemen take that profit, presumably by standing in said lines wasting a lot of time. And buyers now have to trust some rando on ebay to actually ship a 2k$ graphics card. That is not an optimal outcome.
Microsoft and Sony should have offered their new consoles using this method. Start at a very high price like $10,000, decrease the price until people start buying, resume decreasing the price when demand is satisfied, and repeat as needed.
Apple can’t afford the optics of launching a product at one price and then halving it a few weeks later.
There’d be angry op-EDs in Forbes, and the HN threads would be longer than the queues.
I have tried online but cannot beat the bots. I would have to write my own.
Notably, the bulk of the equipment that is purchased is made by US firms (about 1/2), with 1/4 from Europe and 1/4 from Japan.
[0] www.semi.org/en/news-media-press/semi-press-releases/semiconductor-equipment-2020-year-end-forecast
I don't follow the logic. Why would the first 85% imply the second 85%?
It's difficult to nail down the breakout of production regionally, because, for example, the US has 85% of the global semiconductor sales, but only 44% of that 85% are fabricated in the US. Here is the 2019 breakdown from Semiconductor International Association[0].
In unit terms, it could be quite different since there are so many low-value chips made on legacy equipment that has been depreciated.
[0] https://www.semiconductors.org/wp-content/uploads/2019/05/20...
This is incorrect. I looked at the PDF that you had linked to[0], and it says, on page 14 (of the PDF) that the total sales in the Americas is 22% of global sales. They say "Americas", so that includes Canada, Mexico, Central, and South America. The Asia Pacific region is where 60.3% of all semiconductors are sold to. And China is the largest consumer of semiconductors, at 33.8%.
[0] https://www.semiconductors.org/wp-content/uploads/2019/05/20... (under Section 2: Global Market, on page 15 of the PDF / page 11 of the numbering scheme they use)
I can assume that the price per wafer is relatively flat, but neither the size, nor the complexity, nor the dollar value linearly follow from it.
The parts, of course, come from various places.
Usually because they're built with some extreme form of government subsidy, or the fully loaded cost of a nuclear power plant is not taken into account in the $/GWh, such as eventual storage and disposal of nuclear waste at a government-run, government-funded facility.
Chernobyl and Fukushima will cost about $1 trillion each ... for just the first 100 years.
that is going to change very soon as the european union is going to make their own supply chain in fabrication of chips. for security reasons.
Hasn't anyone learned anything when it comes to central planning in the past 100 years?
Page says "Something went wrong."
8-inch or more commonly known as 200mm Wafers even in the US, are NOT under-invested. The total Wafer capacity of 200mm has been steadily increasing for a few years. ( I mean Reuters / Bloomberg reported this at the time, although fairly late in the cycle ).
You cant have a perfect Supply and Demand cycles, there will never be one. And especially during uncertainties.
Then again, when I studied political science and economics there was a common belief/theory that this sort of increased interdependency between countries (globalization) would decrease the likelihood of war and increase opportunities for democracy. So maybe it's a good thing.
Also, the choice of geographies is not that wide. You need a place with lots of reasonably skilled but relatively low-paid people.
But I also install Ubuntu on some specialized "workstations", e.g. pendrive/Internet terminal laptop connected to beefy old laser copier/printer where CUPS PCL works with zero drivers. Other use case: jukebox laptop for music classes :)
Generally, I take it to mean still runs old sofware, but not compatible with current software.
I don’t entirely disagree with you. I would say that a lot of compute resources are used to make user interfaces more impressive, but not necessarily better than what could be done if we designed interfaces around lower resources.
However that isn’t what we’ve done.
Older computers often can’t run software that is useful today even if the software they could run when they were new still does the job. Things like file formats and web standards have moved on and are not backwards compatible, and a lot of what people use computers for is communication, so that matters.
That's exactly what this is an argument for. It's also an argument for programmers to accomodate this.
Despite all the defenses of such by progammers who claim hardware always improving and code cosmetics/organizational trends obviate software performance concerns, I can attest to the reality that I composed electronic music with completely software instruments and effects on 512mb ram on earlier versions of certain unnamed DAW's with all the "latest" plugins at the time. My last Macbook pro 2011 with 4gb ram on Snow Leopard was sufficient, but once I updated it to Yosemite and went from Cubase 5 to Cubase 9, from Ableton 5.5 to Ableton 10, that it went from "performing just fine" to basically running the fan the entire time and glitching at basic playback of sessions I was previously working on no problem. (same plugins). The Steinberg Cubase one being the biggest noticable turd, and I blame their acquisition by Yamaha, as the customer-service also went down the tubes, as I waited months to get a reply on a replacement USB dongle (also a sucky thing for a laptop with 2 USB ports, but I digress)
Software bloat kills. It's not all deliberate, but if our values don't change from "who cares" to "we care", we can't expect any change. Is uncaring mega-corp structure the cause? (programmer wants to go home, project is badly managed, bugs outnumber features, let's start a new project, etc?) I'm curious what folks think is the cause of poor-quality buggy bloated "modern" software that is demonstrably worse than older software.
It's not an improvement in technology, it's simply that our modern world cares more about superficial animated menu openings than about performance.
Mind you, not ALL software follows this trend. Reaper still kicks A, but that's because Mr. Frankl has a vision, integrity, and still maintains control lol
anyway, about chips and processes, it's not about getting tinier and tinier, it's about an intelligent design. All the efficiency improvements in the process are wasted if the goal is to make a mega-poo-renderer that renders 0.1 turds per KW of power expended.
Controversial thought of the day: I will reckon that a simple interrupt driven event-loop on a microcontroller running all HID input is a million times more efficient than the convoluted way a context switching multitasking operating system handles the same under full load with full memory saturation.
It’s always a strike in France. Unionists dream state.