AMD is Allegedly Preparing Navi 31 GPU with Dual 80 CU Chiplet Design

[medi01]

Console APUs have been in mass production since before summer 2020; stocks have had a lot more time to build up than GPUs. Them still being out of stock speaks of gross underestimations of demand there too

No.
Out of stock situation means that console has not failed (yet), merely that.
Shortage itself is a mere reflection of supply chains not being capable of "burst" deliveries.
You can calculate whatever demand figure you want, it doesn't magically increase your production capacity.

This discussion is particularly ironic in the light of EU's vaccine woes...

As for speaking of AMD vs. Nvidia, I still can't find any noticeable stock there either, so...

I see in stock overpriced (not even "accepted" 20% on top of MSRP, thank you, 2080Ti, but +60% and more) GPUs from both manufacturer, with only 3080 being rather rare (but still real) guest.

Sales are also 60% vs 40% green vs red, so, AMD has grabbed some of the market back, despite production shortages.

 

[Valantar]

No.
Out of stock situation means that console has not failed (yet), merely that.
Shortage itself is a mere reflection of supply chains not being capable of "burst" deliveries.
You can calculate whatever demand figure you want, it doesn't magically increase your production capacity.

I never said they had failed, I said they had grossly underestimated demand. I never said they could have increased production volumes up until now either, only that they had been mass producing consoles for several months before launch and should have had a better estimate of necessary stocks. Of course, Covid has likely increased game console demand noticeably, which would have been very difficult for manufacturers to plan for, but they've nonetheless clearly and significantly underestimated demand. When the consoles are still out of stock globally several months after release, they fucked up. Plain and simple. Calling it a paper launch might be a tad harsh on my end, but it's essentially that in the end - no availability, and no actual games to play on them beyond forwards-compatible previous-gen titles.

I see in stock overpriced (not even "accepted" 20% on top of MSRP, thank you, 2080Ti, but +60% and more) GPUs from both manufacturer, with only 3080 being rather rare (but still real) guest.

Sales are also 60% vs 40% green vs red, so, AMD has grabbed some of the market back, despite production shortages.

10-20% premiums for high end OC models with fancy coolers is somewhat acceptable. 60% is a travesty, and really shouldn't affect any account of whether a product is in stock or not. Whether that is from a major retailer or a reseller, those are scalper prices, and can't be taken into account alongside normal sales at normal prices. Any product can be kept in stock given a sufficient increase in the sales price, after all. As for sales distributions, this is too small a sample size to really say. I'm hoping AMD grabs back some market share (we despreately need a more competitive GPU market), but it's going to take a long, long time for them to reach 40% global marketshare.

 

[ratirt]

I like the idea of dual GPU but in a chiplet form not 2 monolithic chips paired together. I think chiplets, is not such a bad idea to introduce in the GPU market especially for AMD. The company has huge experience with chiplets with Ryzen and thus far it does have the performance and efficiency. Both have been improved by every new release. It is promising and I'd like to see more of this solution and how the company will introduce it in GPUs. I'm only hoping this not to be another SLI/Crossfire like solution though.

 

[MikeMurphy]

Never mind, even if you could afford it and actually buy it, there's no power supply on this planet that can feed this beast.

Power consumptions comes way down when run at the golden ratio along the efficiency curve. I could see this as being 350-400w.

 

[AusWolf]

Power consumptions comes way down when run at the golden ratio along the efficiency curve. I could see this as being 350-400w.

That is true, though I don't want to see 350-400 W being the new "way down".

 

[Valantar]

Power consumptions comes way down when run at the golden ratio along the efficiency curve. I could see this as being 350-400w.

350-400W is the most that can feasibly be cooled in a consumer PCIe product, so aiming for that at stock clocks is... iffy. Slow and wide is alswus faster than narrow and fast, so they don't have much to lose by clocking these really, really low. Also, there'd be lots of OC headroom for whoever wants it!

 

[InVasMani]

350-400W is the most that can feasibly be cooled in a consumer PCIe product, so aiming for that at stock clocks is... iffy. Slow and wide is alswus faster than narrow and fast, so they don't have much to lose by clocking these really, really low. Also, there'd be lots of OC headroom for whoever wants it!

I feel with MCM GPU's 3-slot coolers will be upper mid range 4-slot high end 2-slot lower mid range and 1-slot low end or SFF. Which chips are used and why will depend on circumstances.

 

[pantherx12]

So in Australia I've seen the 6900XT being advertised for $2K so I would expect this ridiculous card to be over $4K as we'll no doubt see AMD also increase prices again with next gen CPUs and GPUs.

I'd rather they increased bus width as it's patently obvious that the reason Navi 21 falls behind at 4K is the 256 bit bus width. It was always a stupid decision for the high end cards despite what infinity cache achieves. Same will happen with 6700XT being only 192 bit.

If they have IO as a separate chip then it will be easier to have higher band width cache presumably.

This could also bring down prices as they could do as they do with zen now and have the IO on a cheaper larger process node and just have the compute parts on 7 or 5nm etc.

 

[medi01]

I said they had grossly underestimated demand.

This statement is based purely on fact of supply not being demand.
And it turns blind eye to already voiced "but you can't just ramp up production at will".
And no, "but you can produce it in advance" does not fly, production starts only a few months before sales, as chips that go inside consoles are rather new.

 

[Valantar]

This statement is based purely on fact of supply not being demand.
And it turns blind eye to already voiced "but you can't just ramp up production at will".
And no, "but you can produce it in advance" does not fly, production starts only a few months before sales, as chips that go inside consoles are rather new.

Mass production of console SoCs start at least 6 months before launch, precisely due to the long and complex supply chain for these devices as well as the need to build up volumes for launch. That obviously doesn't mean their makers have much opportunity to increase volumes after that start, given the typically ~6-month lead time on foundry orders and the lack of sales data before launch. But it's still undeniable that their initial sales predictions have very clearly been woefully low compared to actual demand. That might be due to Covid, it might be due to a heap of other reasons, but that still doesn't change the fact that they severely underestimated demand, and have completely failed to supply enough consoles for them to be widely available. But as I said above, there aren't any current-gen titles either (well, a handful at best), so it's not much of a loss even if the backwards compatibility experience is excellent. And hopefully by now they've long since increased their orders with TSMC - but it might be that they too are held back by other current shortages such as in chip packaging, silicon wafers or even DRAM.

 

[r.h.p]

The Voodoo that you do so welllll!

"paper launch q1 2023!"

voodoo 4 i think , was excited i was .Swapped to Ge force from memory

 

[Vayra86]

I never said they had failed, I said they had grossly underestimated demand. I never said they could have increased production volumes up until now either, only that they had been mass producing consoles for several months before launch and should have had a better estimate of necessary stocks. Of course, Covid has likely increased game console demand noticeably, which would have been very difficult for manufacturers to plan for, but they've nonetheless clearly and significantly underestimated demand. When the consoles are still out of stock globally several months after release, they fucked up. Plain and simple. Calling it a paper launch might be a tad harsh on my end, but it's essentially that in the end - no availability, and no actual games to play on them beyond forwards-compatible previous-gen titles.

10-20% premiums for high end OC models with fancy coolers is somewhat acceptable. 60% is a travesty, and really shouldn't affect any account of whether a product is in stock or not. Whether that is from a major retailer or a reseller, those are scalper prices, and can't be taken into account alongside normal sales at normal prices. Any product can be kept in stock given a sufficient increase in the sales price, after all. As for sales distributions, this is too small a sample size to really say. I'm hoping AMD grabs back some market share (we despreately need a more competitive GPU market), but it's going to take a long, long time for them to reach 40% global marketshare.

The shortage is global. I'm not sure if fucking up is the right word for it. When you haven't got capacity, its just not there. This is not directly pandemic related either, the whole supply chain is under stress because demand has exceeded supply for years now. This is why prices have been surging for several components like NAND, RAM, etc and keeps doing so.

At the same time, these chips are produced for the cutting edge of consumer hardware, so you can't just relegate that to older nodes and be done with it - that directly hurts the selling point of your new product, if it doesn't kill it altogether.

I'm still puzzled by the whole entitled attitude wrt buying luxury products like these. Still whining about scalper prices. This is how free markets work, if it took you until today to figure that out... well... welcome to Earth. It's been like this for over 2000 years, ever since we discovered the concept of trade. Scarce = expensive. Simple.

 

[kapone32]

The shortage is global. I'm not sure if fucking up is the right word for it. When you haven't got capacity, its just not there. This is not directly pandemic related either, the whole supply chain is under stress because demand has exceeded supply for years now. This is why prices have been surging for several components like NAND, RAM, etc and keeps doing so.

At the same time, these chips are produced for the cutting edge of consumer hardware, so you can't just relegate that to older nodes and be done with it - that directly hurts the selling point of your new product, if it doesn't kill it altogether.

I'm still puzzled by the whole entitled attitude wrt buying luxury products like these. Still whining about scalper prices. This is how free markets work, if it took you until today to figure that out... well... welcome to Earth. It's been like this for over 2000 years, ever since we discovered the concept of trade.

I was talking to a friend at my local PC shop. They have lots of 3070 and a few 3080 from last week's stock drop. They had 30 6000 series cards listed at the same time and they sold in less than 2 hours with pick up only.

 

[Valantar]

The shortage is global. I'm not sure if fucking up is the right word for it. When you haven't got capacity, its just not there. This is not directly pandemic related either, the whole supply chain is under stress because demand has exceeded supply for years now. This is why prices have been surging for several components like NAND, RAM, etc and keeps doing so.

At the same time, these chips are produced for the cutting edge of consumer hardware, so you can't just relegate that to older nodes and be done with it - that directly hurts the selling point of your new product, if it doesn't kill it altogether.

I'm still puzzled by the whole entitled attitude wrt buying luxury products like these. Still whining about scalper prices. This is how free markets work, if it took you until today to figure that out... well... welcome to Earth. It's been like this for over 2000 years, ever since we discovered the concept of trade. Scarce = expensive. Simple.

I don't disagree with any of that, especially the expectation of widely available parts - it's more that I'm surprised at the seeming lack of foresight of these companies. Though there are of course relevant questions to be asked regarding their actual influence in that regard, seeing how they are all fighting over the same limited supply.

I don't quite remember where, but I raised the exact question of why nobody seems to have seen this coming despite a situation like this needing several years to take place in some other thread. It's frankly rather baffling, and makes me wonder if it's simply down to consolidation (ever fewer actors competing in every part of the value chain) or if the problem is more fundamental than that. I mean, we all know how the memory industry actively works to maintain just the right amount of scarcity to keep prices high without triggering a crisis, so I'm suspecting similar ideologies to be at play here as well in some form or other. I.e. nobody wants to compete in the high-end fab business just to break even (or perhaps more precisely, investors wouldn't allow them to, would tank their stock prices and drive them into bankruptcy for not being profitable enough), but someone somewhere must have seen where this was going.

A decreasing rate of node improvements has no doubt contributed by squeezing ever more products onto an ever smaller collection of nodes, as has ever-increasing chip demands for an ever expanding range of products (light bulbs 10 years ago didn't have much silicon in them, nor did toasters or coffee makers), but the assumption from that if going by simple supply/demand logic would be a continuous expansion of the silicon fab industry. Of course building a fab is a 5+-year, multi-billion-dollar bet, so there can't be many companies willing and able to make those bets, but it still seems really weird that so few seem to have seen this as a major opportunity. I guess the increased digitalization of ... well, everything is likely the biggest contributor here (the amount of advanced computation taking place in contemporary cars is downright staggering), but it also seems to me that the loss of GloFo in the high-end fab space was a bigger blow than anyone seemed to notice, and it's starting to look like Intel's previous 14nm shortage was more of a precursor to a broader problem than anyone thought. I mean, since 2017 we've gone from 3 major high-end fab actors to 1, with Samsung being the perennial outsider looking in but somehow seeming to be number two now? It's all rather weird.

It's maybe even more baffling again that there are currently major, long-term shortages for ... silicon wafers and chip packaging substrates. I mean, really? I get that wafers require very pure silicon and highly specialized equipment, but it's still just monocrystalline melted sand spun into a cylinder(-ish) and cut into discs, and packaging substrates are ultrafine fiberglass and copper sandwiches. Are these things that difficult to scale up? Or are the forces behind these industries just not interested in investing into expansion due to the commodity nature of the products?

So while I am in some sense frustrated that there's a sudden and systemic shortage of what seemed to be widely available consumer products pretty much yesterday, I'm more baffled by just how unprepared the industry seems to be for the direction it's growing in, especially given just how the very same industry constantly promotes how computerization of anything and everything will somehow save the world. It just looks like they "suddenly" got their wish (well, over 10+ years), yet had never really imagined it actually happening, and certainly hadn't actually considered what that scenario coming true would actually mean.

 

[Vayra86]

I don't disagree with any of that, especially the expectation of widely available parts - it's more that I'm surprised at the seeming lack of foresight of these companies. Though there are of course relevant questions to be asked regarding their actual influence in that regard, seeing how they are all fighting over the same limited supply.

I don't quite remember where, but I raised the exact question of why nobody seems to have seen this coming despite a situation like this needing several years to take place in some other thread. It's frankly rather baffling, and makes me wonder if it's simply down to consolidation (ever fewer actors competing in every part of the value chain) or if the problem is more fundamental than that. I mean, we all know how the memory industry actively works to maintain just the right amount of scarcity to keep prices high without triggering a crisis, so I'm suspecting similar ideologies to be at play here as well in some form or other. I.e. nobody wants to compete in the high-end fab business just to break even (or perhaps more precisely, investors wouldn't allow them to, would tank their stock prices and drive them into bankruptcy for not being profitable enough), but someone somewhere must have seen where this was going.

A decreasing rate of node improvements has no doubt contributed by squeezing ever more products onto an ever smaller collection of nodes, as has ever-increasing chip demands for an ever expanding range of products (light bulbs 10 years ago didn't have much silicon in them, nor did toasters or coffee makers), but the assumption from that if going by simple supply/demand logic would be a continuous expansion of the silicon fab industry. Of course building a fab is a 5+-year, multi-billion-dollar bet, so there can't be many companies willing and able to make those bets, but it still seems really weird that so few seem to have seen this as a major opportunity. I guess the increased digitalization of ... well, everything is likely the biggest contributor here (the amount of advanced computation taking place in contemporary cars is downright staggering), but it also seems to me that the loss of GloFo in the high-end fab space was a bigger blow than anyone seemed to notice, and it's starting to look like Intel's previous 14nm shortage was more of a precursor to a broader problem than anyone thought. I mean, since 2017 we've gone from 3 major high-end fab actors to 1, with Samsung being the perennial outsider looking in but somehow seeming to be number two now? It's all rather weird.

It's maybe even more baffling again that there are currently major, long-term shortages for ... silicon wafers and chip packaging substrates. I mean, really? I get that wafers require very pure silicon and highly specialized equipment, but it's still just monocrystalline melted sand spun into a cylinder(-ish) and cut into discs, and packaging substrates are ultrafine fiberglass and copper sandwiches. Are these things that difficult to scale up? Or are the forces behind these industries just not interested in investing into expansion due to the commodity nature of the products?

So while I am in some sense frustrated that there's a sudden and systemic shortage of what seemed to be widely available consumer products pretty much yesterday, I'm more baffled by just how unprepared the industry seems to be for the direction it's growing in, especially given just how the very same industry constantly promotes how computerization of anything and everything will somehow save the world. It just looks like they "suddenly" got their wish (well, over 10+ years), yet had never really imagined it actually happening, and certainly hadn't actually considered what that scenario coming true would actually mean.

Simple answer: managers have zero knowledge or care about details. Their reality held up in normal times but when the shit hits the fan, it all falls apart. I no see it at work, I see it in news and in siciety at large. We managed away responsibility.

We live in a paper reality and the pandemic shows us where the problems really are. The same thing occurs in politics especially in first world countries. Netherlands, the UK, the US.. we are fuckihg failures with no direction whatsoever. No perspective and no vision. The stuff that happens over here lately you coukdnt think it up. And its all down to that same distanf mentality: managers trying to manage nothing but their own PR.

Thats that top 3% for ya.

 

[Valantar]

Simple answer: managers have zero knowledge or care about details. Their reality held up in normal times but when the shit hits the fan, it all falls apart. I no see it at work, I see it in news and in siciety at large. We managed away responsibility.

We live in a paper reality and the pandemic shows us where the problems really are. The same thing occurs in politics especially in first world countries. Netherlands, the UK, the US.. we are fuckihg failures with no direction whatsoever. No perspective and no vision. The stuff that happens over here lately you coukdnt think it up. And its all down to that same distanf mentality: managers trying to manage nothing but their own PR.

Thats that top 3% for ya.

That's true in a lot of ways - IMO this can mostly be attributed to the growth of neoliberalism since the Reagan era, with its combination of deregulation of business, export of production to wherever the cheapest (and least organized, least protected) labor can be found, and introduction of an ideology obsessed with measuring "productivity" and "profitability" in ever-smaller and more arbitrary metrics while ignoring the big picture whenever possible. As long as the numbers look right (which they do, as they've defined what to count by what looks best when counted and can be optimized for counting), nobody is to blame when the house of cards collapses. While I'm rather scared of the isolationist/xenophobic undertones in a lot of it, I'm glad various governing bodies around the world are finally starting to see the necessity of maintaining a global distribution of high-end industries. Who knew it would be a bad idea to let the industries producing the things we fundamentally need for everything to be concentrated in a few regions, in constant competition, go through heaps of buyouts and mergers, and become ever more consolidated? Yeah, someone ought to have seen that one coming. At this point, who knows if it's too late already and we're heading into some major slump? Guess we'll see in a year or two.

 

[medi01]

But it's still undeniable that their initial sales predictions have very clearly been woefully low compared to actual demand.

Dude, you are freaking me out.
There has not been A SINGLE MAJOR CONSOLE LAUNCH when product was not sold out for many weeks.
Because there is BURST demand in the beginning that nobody can match production capacity wise.
Because you cannot have BURST production and neither could you afford producing stuff YEARS before launch.

"Oh, you suddenly need 10 times normal supply of X? Well, am I supposed to build additional production capacity just to satisfy your needs for the next 3 onths???"

How hard is that to comprehend?

In such situation, something being sold out is NOT and can NOT be a sufficient indicator of someone mis-calculating sales.

There is exactly ZERO evidence either Sony or Microsoft underestimated demand for PS5 and XSex respectively.

That might be due to Covid

Someone sane expected demand for AT HOME ACTIVITY goods to DROP due to covid?
/facepalm

PS
And on purchasing and supply chains, including logistics.
I have, in fact, been working with IT projects in the are.
Stop trash talking about those guys, the do their job pretty damn well, in facts, it's shocking how good they are at something that complex.

 

[Valantar]

Dude, you are freaking me out.
There has not been A SINGLE MAJOR CONSOLE LAUNCH when product was not sold out for many weeks.
Because there is BURST demand in the beginning that nobody can match production capacity wise.
Because you cannot have BURST production and neither could you afford producing stuff YEARS before launch.

"Oh, you suddenly need 10 times normal supply of X? Well, am I supposed to build additional production capacity just to satisfy your needs for the next 3 onths???"

How hard is that to comprehend?

In such situation, something being sold out is NOT and can NOT be a sufficient indicator of someone mis-calculating sales.

There is exactly ZERO evidence either Sony or Microsoft underestimated demand for PS5 and XSex respectively.


Someone sane expected demand for AT HOME ACTIVITY goods to DROP due to covid?
/facepalm

PS
And on purchasing and supply chains, including logistics.
I have, in fact, been working with IT projects in the are.
Stop trash talking about those guys, the do their job pretty damn well, in facts, it's shocking how good they are at something that complex.

First off: being "sold out for many weeks" does not equate to zero availability for several months, as we're seeing now. A restock blip every two weeks that sells out in a few hours is still zero availability. Previous consoles have been hard to get after launch, but not at this level, at least in recent history.

Secondly, the burst of initial demand is exactly why stocks are built up beforehand. You're not saying anything that hasn't already been addressed at length. If absolute peak production is X consoles/month, expected normal demand is, say, 1/2 X (to ensure sufficient production overhead to account for sales fluctuations), and launch month demand is expected to be 5X, then you try to produce consoles at maximum capacity for 5 months before launch to ensure sufficient supply. That's common practice, I'm just saying they missed their estimates. Of course there are tons of complicating factors like the added difficulty of shipping and distributing huge quantities just before launch (compared to the steady flow of regular sales), but given the sustained lack of stock, that clearly wasn't the issue. It is of course entirely possible that demand has been so high that no possible amount of pre-launch production could reasonably have ensured availability, but if that's the case, the companies involved should (and likely would) have addressed it far more clearly than they have.

Third, of course there is evidence of underestimated demand. There being no stock is evidence of that. Nobody plans to be sold out for months and months - consoles are a low-margin business reliant on after-sales, so every sale that doesn't happen hurts their bottom line. You can't buy games for a console you don't have, so they clearly plan to have the product available at all times. Nobody plans for stock shortages unless they're in the fashion industry. Again, there is a possible second explanation: component shortages beyond the control of the console maker. But again, if that was the case, why haven't they addressed that publicly? It's no skin off their backs to say "sorry, we're working as fast as we can, but we're struggling to get a hold of [component A] and production volumes are thus lower than our targets."

Fourth: please reread. I very clearly didn't say anyone expected at-home activity to drop due to covid, I said that the influence of covid on demand might be one of the reasons why their estimates were off, as it's an unknown factor that obviously makes estimating demand much, much more difficult. (Especially as it works both ways: on the one hand there's increased demand for at-home activities, while on the other hand a lot of people are struggling financially. Balancing the two without any real data to go by - which doesn't exist - is essentially just guesswork.) I mean, at this point you're just misreading me on purpose if that's what you got out of that sentence. Please take a breath and swallow your indignation, as it's causing you to not actually read what I'm saying.

Fifth: "trash talking"? Seriously? I mean, get a grip, man. This is just ridiculous. Saying that a vaguely defined group of people spanning dozens of companies and hundreds if not thousands of people missed some estimates and thus seriously messed up a product launch constitutes trash talking to you? Calm down, please. There's nothing personal in this, neither directed at you or at anyone involved in this process. Not even anything saying they're bad at their jobs - I've never said anything to that effect. Estimates are always estimates, they are never a sure thing. This just happens to be a much bigger mess-up than what we normally see. Stop making this personal.

 

[Sunny and 75]

Another reason to skip current gen! RX 7000 series and RTX 40 series is the place to be. And maybe Xe-HPG if it proves to be a competitive Intel product. We'll see.

Chiplets on GPU! Curious about the result of such implementation and if RYZEN is anything to go by, I'd say Nvidia Lovelace just got a good run for their money.

there aren't any current-gen titles

Sony themselves said the real PS5 games won't be available before 2022.

Sony boss: no 'generation-defining' PlayStation 5 games until 2022

Sony Interactive Entertainment CEO Jim Ryan says that 'generation defining' games will arrive on the PlayStation 5 in 2022.

www.tweaktown.com

 

[watzupken]

Actually I will be really interested to see how hot these chiplet GPUs run. Considering the Ryzen 7 5800X easily hits 90 degs with the 142W thermal limit and a 360 AIO cooler, these high end GPUs are not going to be just sipping power if the current gen GPUs are a precursor of what to expect in future GPUs. I think it will be a challenge to try and keep 2x 100+ watts of chiplets cool since each chiplet should be quite small in size.

Console APUs have been in mass production since before summer 2020; stocks have had a lot more time to build up than GPUs. Them still being out of stock speaks of gross underestimations of demand there too. At least there are no miners using consoles ...

As for speaking of AMD vs. Nvidia, I still can't find any noticeable stock there either, so...

Also, what they are encouraged/discouraged to produce is meaningless - if a GPU hits shelves in Europe/the US today, that means its silicon started its journey through the fab 3-4 months ago at the very least. If Nvidia made adjustments to which die each wafer made is immediately after launch, we'd be seeing those adjustments about now, if not a bit into the future. And there's no way they made any such adjustments that early. So your quasi-conspiracy theory has pretty wobbly legs to stand on at best.

Actually I think it is a combination of supply and demand that is causing havok.

Supply - If you look at the timeline, COVID lockdowns hit really hard when the products are supposed to start or started mass production. So if AMD for example have an obligation to fulfill XXX number of XBOX and PS5 SOCs, it is likely they will fall behind in the production. So when things started to improve towards the end of the year, they are still busy trying to fulfill those orders. As a result, there is a knock on impact to producing Ryzen 5000 and RX 6900/6800. And don't forget, the lockdowns did not happened concurrently globally, they took place at different times, and with different durations. So I am sure there will be parts/ components shortages.

Demand - Having launched multiple times in the past, both AMD and Nvidia are very seasoned when estimating demand. But I feel the exception this year is that AMD became very competitive against Nvidia's Ampere. In the past few generations, it has always been a cake walk for Nvidia and they can afford to price their products at a premium. As a result, Nvidia generously handed gamers a GA102 chip @ USD 699, and the GA104 @ US 499. I feel Nvidia would have priced the RTX 3080 easily at USD 999 without competition, or give you a RTX 3080 based on the smaller GA104 chip. So it came as a surprise to us all, and sells like hot cakes. Of course the GA102 being a big and complex chip, was never meant to be mass produced in large numbers to begin with. As AMD joined the party late, they know they can't compete in terms of features, so accordingly have to price their RDNA2 flagships lower to compete. So its not that AMD or Nvidia did not expect demand to be high, its because competition forced them to compete on price vs performance, hence causing the imbalance between supply and demand. At least this is my opinion.

 

[medi01]

First off: being "sold out for many weeks" does not equate to zero availability for several months, as we're seeing now

"Sony has sold X million units" somehow equates to "zero availability".

Secondly, the burst of initial demand is exactly why stocks are built up beforehand.

Except that "beforehand' is never long enough to accommodate for the burst, as designs are finalized not long before launch.

Third, of course there is evidence of underestimated demand. There being no stock is evidence of that.

So, repeating moot argument as third point somehow makes it true. got it.
Once again, for particularly bright people: there are reasons OTHER THAN underestimating demand that lead to product shortages. Such as, you know, MANUFACTURING. AMD would love to bump production of its stuff by about 50% more waffers, except, nopie, TSMC is at full capacity, you know.
BUT CAN'T I BUILD IT IN ADVANCE? Yes, you can, what we got now was built in advance, as early as it was technically possible.

 

[evolucion8]

GPU market now is terrible. I had a Radeon VII Anniversary Edition which I paid $699 for it in April 2019 directly from AMD, and sold it three weeks ago for $1300 in Ebay, like I just put it on bid and see how far would it go. I end up using that same money and add $100 more and purchased an RX 6900XT from Facebook marketplace, brand new, unopened. Very happy with the purchase but just insane how it goes regarding pricing of new and old hardware.

 

[Valantar]

"Sony has sold X million units" somehow equates to "zero availability".


Except that "beforehand' is never long enough to accommodate for the burst, as designs are finalized not long before launch.


So, repeating moot argument as third point somehow makes it true. got it.
Once again, for particularly bright people: there are reasons OTHER THAN underestimating demand that lead to product shortages. Such as, you know, MANUFACTURING. AMD would love to bump production of its stuff by about 50% more waffers, except, nopie, TSMC is at full capacity, you know.
BUT CAN'T I BUILD IT IN ADVANCE? Yes, you can, what we got now was built in advance, as early as it was technically possible.

And 'round and 'round we go. Did you notice that nothing you said there actually contradicts my reasoning? I've never said that factors beyond poor planning haven't contributed significantly to this. Nor have I said that they haven't sold a significant amount of units. It's obvious that these products are subject to the same limitations as everything else. What I did say was that it's clear that they underestimated demand - which they did, otherwise they would have moved up preproduction to increase stocks - which by itself amounts to poor planning, regardless of contributing circumstances. I never said that this is the sole reason for the shortages, not anything to that effect. Though there's one thing worth pointing out: final console specs are decided quite late, but the silicon design for the SoC is finalized at least a year before launch, though often more like 18 months. Then it goes to tape-out, test production, then mass production - as I said, at least half a year before launch. There is typically wiggle room there to get things going faster if necessary.

And again: total sales numbers don't matter much if there's a sustained shortage - the shortage itself tells us that there is insufficient supply, which is a supply-side issue. You're presenting it as if this is only down to higher than normal demand, which is simply not true. Demand is indeed higher than normal, but supply is also significantly constrained, and its reasonable to say that console makers should have been aware of these challenges and done their best to account for them. It's also entirely possible that they have done so, but given the distinct lack of clear public statements about this that doesn't seem like the most reasonable assumption.

Actually I will be really interested to see how hot these chiplet GPUs run. Considering the Ryzen 7 5800X easily hits 90 degs with the 142W thermal limit and a 360 AIO cooler, these high end GPUs are not going to be just sipping power if the current gen GPUs are a precursor of what to expect in future GPUs. I think it will be a challenge to try and keep 2x 100+ watts of chiplets cool since each chiplet should be quite small in size.


Actually I think it is a combination of supply and demand that is causing havok.

Supply - If you look at the timeline, COVID lockdowns hit really hard when the products are supposed to start or started mass production. So if AMD for example have an obligation to fulfill XXX number of XBOX and PS5 SOCs, it is likely they will fall behind in the production. So when things started to improve towards the end of the year, they are still busy trying to fulfill those orders. As a result, there is a knock on impact to producing Ryzen 5000 and RX 6900/6800. And don't forget, the lockdowns did not happened concurrently globally, they took place at different times, and with different durations. So I am sure there will be parts/ components shortages.

Demand - Having launched multiple times in the past, both AMD and Nvidia are very seasoned when estimating demand. But I feel the exception this year is that AMD became very competitive against Nvidia's Ampere. In the past few generations, it has always been a cake walk for Nvidia and they can afford to price their products at a premium. As a result, Nvidia generously handed gamers a GA102 chip @ USD 699, and the GA104 @ US 499. I feel Nvidia would have priced the RTX 3080 easily at USD 999 without competition, or give you a RTX 3080 based on the smaller GA104 chip. So it came as a surprise to us all, and sells like hot cakes. Of course the GA102 being a big and complex chip, was never meant to be mass produced in large numbers to begin with. As AMD joined the party late, they know they can't compete in terms of features, so accordingly have to price their RDNA2 flagships lower to compete. So its not that AMD or Nvidia did not expect demand to be high, its because competition forced them to compete on price vs performance, hence causing the imbalance between supply and demand. At least this is my opinion.

I think you're being far too generous with Nvidia here. Remember, Turing prices were unprecedented bad value over previous generations. $699 for the 3080 isn't generous as much as it is a return to a semblance of normalcy, though the lack of $300 and below products still tells of skewed market with a harmful focus on high ASPs and profits. Hopefully the Ampere and RDNA2 stacks will fill out soon to rectify this, but until then we still have a really bad GPU market regardless of supply. Of course production costs are rising (10GB of GDDR6X costs more now than 4GB of GDDR5 did for the 980, after all, and a huge die on 8nm costs a lot more than a huge die on 28nm), but that still doesn't explain the huge cost increases we've seen in recent years.

But other than that, yeah, it's clear that there's an unprecedented confluence of supply-side issues and higher than normal demand. It also seems like the supply-side issues run down the production chain quite a bit, which is quite worrying. A demand spike is one thing, but it shouldn't be enough to trigger what we've seen over the past few months.

 

[medi01]

What I did say was that it's clear that they underestimated demand - which they did, otherwise they would have moved up preproduction to increase stocks

Oh boy.
You were specifically told about "no, you can't go into preproduction as early as you will at least twice in this thread.

And 'round and 'round we go.

Ironic.

Though there's one thing worth pointing out: final console specs are decided quite late, but the silicon design for the SoC is finalized at least a year before launch, though often more like 18 months. Then it goes to tape-out, test production, then mass production - as I said, at least half a year before launch. There is typically wiggle room there to get things going faster if necessary.

There is exactly ZERO indication of console APUs not being produced half a year before launch, nor is there any evidence of that possibly having any visible effect on availability TODAY.

It's also entirely possible that they have done so, but given the distinct lack of clear public statements about this that doesn't seem like the most reasonable assumption.

In other words, instead of admitting the obvious (made up accusations of "miscalculation of demand") let's call that weird theory "most reasonable assumption", shall we...

 

[InVasMani]

Perhaps by the year 2035 we'll have a GTX 1080 level of performance from a new GPU for $200's with the rate of innovation that's happening. The Radeon 580 is still on a island of it's own about 5 years after it launched in terms of value for dollar at the $200 price point it's tragic.