Saturday, January 23rd 2021
AMD is Allegedly Preparing Navi 31 GPU with Dual 80 CU Chiplet Design
AMD is about to enter the world of chiplets with its upcoming GPUs, just like it has been doing so with the Zen generation of processors. Having launched a Radeon RX 6000 series lineup based on Navi 21 and Navi 22, the company is seemingly not stopping there. To remain competitive, it needs to be in the constant process of innovation and development, which is reportedly true once again. According to the current rumors, AMD is working on an RDNA 3 GPU design based on chiplets. The chiplet design is supposed to feature two 80 Compute Unit (CU) dies, just like the ones found inside the Radeon RX 6900 XT graphics card.
Having two 80 CU dies would bring the total core number to exactly 10240 cores (two times 5120 cores on Navi 21 die). Combined with the RDNA 3 architecture, which brings better perf-per-watt compared to the last generation uArch, Navi 31 GPU is going to be a compute monster. It isn't exactly clear whatever we are supposed to get this graphics card, however, it may be coming at the end of this year or the beginning of the following year 2022.
Source:
VideoCardz
Having two 80 CU dies would bring the total core number to exactly 10240 cores (two times 5120 cores on Navi 21 die). Combined with the RDNA 3 architecture, which brings better perf-per-watt compared to the last generation uArch, Navi 31 GPU is going to be a compute monster. It isn't exactly clear whatever we are supposed to get this graphics card, however, it may be coming at the end of this year or the beginning of the following year 2022.
141 Comments on AMD is Allegedly Preparing Navi 31 GPU with Dual 80 CU Chiplet Design
All I see is 2300 Euro worth of cards over at eBay. Don't know what to do and how to get one without getting robbed.
I'm reading online all sorts of stuff and it seems to me (albeit I ain't no expert), that it must be due to a combination of things really.
1) Miners indeed.
2) Scalpers also.
3) Unexpected demand much higher than usual.
4) Lack of foundry diversity.
Just talking out of my ass here, but it seems to me that only having tsmc producing stuff, is a pretty frightening thought. I mean, even Intel considers outsourcing to tsmc? Are we for real? :confused:
Someone outside of the business, would assume that it's an ideal opportunity to jump in the wagon and make some money. Lots of money, in fact. However and now things get really interesting and pretty strange for an outsider, that seems to be out of the question? I mean, is it so inherently complex to start designing a new graphics card architecture as well as a fab to manufacture it? It must be cause nobody seems to be able to do it. Even the lads with the deep pockets are not interested, it seems.
Also, I've been reading some stuff about the start ups designing AI chips for NN training etc. They are quite a few around and again, all they can afford is the DESIGN? I mean, as far as construction, they send orders to tsmc also? What the hell is going on in here? :(
Anyway, I'm really really disappointed I cannot get a couple of 3080s for the 2 PCs we have at home, let alone just 1. Sorry about the rant, lads. :oops:
In this context, small improvement is highly unlikely to have visible impact.
I.e. if we call bump by 5% to be "small improvement" supply would need to be at 95% demand for situation to change from "missing" to "available".
This also addresses "things could have been better... if they'd started one month earlier" => highly unlikely.
Also, those percentage examples are meaningless. Neither demand nor supply can be expressed without taking time into account, so unless you're speaking of, say, X%/month, those numbers mean nothing. Of course shipping and distribution makes this all the more complex as there's a 1-3 month delay between production and sales. Demand also always tapers off over time (more people are interested in buying while the product is new, and over time the market is saturated), meaning that boosting production will help get closer to being ahead of the curve there. So yes, any improvement helps.
It's also pretty telling that all of a sudden you've abandoned all of your claims of be saying that "they" (presumably specific people, or at least specific groups of people) are "literally idiots". Hm. Might it be that you had absolutely zero basis for saying this? Moving the goal posts, personal attacks, straw man arguments and accusing people disagreeing with you of acting poorly are all classic signs of bad-faith arguing, so I'd recommend you take a minute to try to identify what aggregated you so much about this and why. Because it doesn't seem to be relevant to this thread.
I'm telling you, that to stop "out of stock" situation supply needs to beat demand.
If you are able to satisfy only 40% of orders, even DOUBLING number of units available, i.e getting to 80% of demand, won't get you out of "out of stock" situation.
Designing chips is overall relatively cheap - they're all built from the base silicon components (the specifics of which are provided by the fab you're working with for your chip), so you "just" need appropriate design software and a team with the skillset necessary to understand how to make transistors do whatever magic it is you want them to do. That's a millions-of-dollars type of cost scale, unless you're designing mature, highly optimized products in a very competitive market, in which tuning and tweaking massively increases costs. Manufacturing chips is incredibly expensive and complex. Developing cutting-edge fabrication nodes is something very, very few companies are capable of. Intel's persistent failure to deliver new nodes on time for the past 7 years and GloFo dropping out entirely of the cutting-edge node game are both clear signs of this. We're talking development costs in the billions of dollars, as well as similar costs for every single fab built.
Another crucial issue that causes some of this that you didn't mention is that there's just one company producing lithography machines for these nodes: ASML. Everyone buys the machines they use to make chips from them - TSMC, Samsung, Intel, everyone. And they have relatively limited production capacities - on the order of ~100 machines a year, which when considering that each fab needs more than one of these for any kind of volume production, is not a lot. There are other companies making lithography machines for legacy nodes (IIRC both Zeiss and Nikon have been in this market, though I don't know if they still are), but given that the manufacturing and development of this equipment is in and of itself incredibly expensive and complex, things tend to consolidate in unregulated capitalist systems. What we're seeing in this regard is just an expression of a decades-long process.
I don't see this improving without public/government intervention, as the pressures on companies to deliver profits to investors in most parts of the world are far too harsh for such a venture to be feasible. The world needs to start recognizing that silicon fabrication is a crucial aspect of global infrastructure, and that it needs to be broadly supported if we are to avoid potentially very dangerous shortages. And happily a lot of governments (at least the US and EU) are recognizing this, but these things are slooooooow to get off the ground. I mean, first you need billions in funding, then you need land (with highly specific requirements) and 100% stable power infrastructure and a qualified (and extremely specialized) workforce, then you need to build a fab, which takes a long time, then you need to get your lithography machines, which takes several years .... yeah, this is going to take some time to fix. But for any fix to be sustainable, it needs to be kept at least partially out of the control of kleptocrat investors and financiers - unless the knowledge requires to build, maintain and develop this is at least partly publicly owned and available, we'll just keep falling into this situation. Again: please stop pretending that demand and supply aren't affected by time. Speaking of these things as if they are static or in "snapshots" like that is meaningless. Demand is dynamic and finite, just as supply is. As more is produced, you get closer to meeting demand.
There are easily 50 million people worldwide interested in a new console. That obviously doesn't mean even remotely close to all of them have the money to buy one now - that number (which is a random example, obviously) is total possible demand for the product. We know console sales are highly cyclical, with massive sales spikes around new releases and smaller spikes around refreshes, with everything else following a pretty typical decelerating decline until some semblance of a steady-state level of demand is hit. So there might be 15 million willing and able to buy a console in the first 6 months, but another 10 million over the next year, and another 5 the year after that, and so on. To meet demand, you need to chip away at that initial mass, as that's where the bulk of sales are found and where issues arise. And the longer you take to meet that initial demand, the longer you'll have shortages, angry customers, etc., as the volumes are that much higher. In this example scenario, if you're only able to meet 50% of demand for the first 6 months, that pushes another 7.5 million buyers into the next year, nearly doubling demand for that period. So the longer you're unable to meet demand, the more harmful it is, as you just keep pushing customers away, and keep piling up unmet demand. That's why you do everything in your power - hire more temporary staff, order more parts, rent more facilities, even build new factories! - to meet demand as early as possible, as those are a) sales just waiting to happen, meaning revenue, and b) potential angry customers that will hurt your reputation long-term if you fail to meet demand. Compared to that, whatever initial costs are needed are typically quite small, and can be made up for in various ways (if you've built a factory, you can always sell it or use it for something else, after all).
And again, I've never argued that this would magically fix things, I said it would improve things. If you meet a higher proportion of demand at any given time, that means you've eliminated more of the total pool of demand. That means there's less future demand to deal with in order to catch up. Thus, any increase in production volumes helps. Is that really so hard to grasp?
Healthy and rich bit is not only "not hard to grasp" but nothing someone is arguing with you about.
Also, you know that you've been arguing that nothing would have been better if they had made some more units, right? Putting your oversimplifying metaphors aside, delivering more consoles is good in and of itself. Isn't that an improvement, then?
Indeed I forgot to mention ASML - the company behind all of tech nobody ever heard of. I recall some videos I've watched on youtube and I was amazed with the size of those tools and therefore their complexity. It was like watching a Sci-Fi bio thriller or something.
You' re quite right that maybe it's the time for governments to chime in and save the industry, because clearly this ain't going anywhere. It will just get worse given that more and more industries get digital. Now we hear about cars, tomorrow it will be refrigerators with cameras and sensors and ethernet cards and all sorts. Like having agents in a program watching when we run out milk and fruit and then send an order directly at Tescos or something.
Far-fetched maybe, but this scenario may unfold sooner than later in smart cities and the like.
Cheers
We have seen acute shortages of PS3, PS4, XBoxSomething, Nintendo Switch that lasted months and months after release.
I still don't understand why you're taking on some imagined burden of defending the poor, put-upon amorphous mass of the console production and distribution chain, as if it's somehow being mistreated by me stating the simple fact that from everything we currently know, it sounds reasonable that more could have been done earlier on to alleviate supply issues. There is nothing personal in saying that - it's a perfectly normal thing. How is anything supposed to improve if we're not allowed to point out things that haven't gone as they should?
- pandemic; a virus is a direct countermeasure of the planet against overpopulation. Its an ecosystem and we're trying to break it, it does fight back - and we will probably win, like we did in the past, making the problem and the threat even greater. The only way forward here is escalation, if we keep growing as a species. There are no square miles added to the earth surface and something's gonna give. More people & animals per square mile = more disease, its a statistical truth we'll never overcome.
- demand; we want and require chips in everything these days. Mechanical is being replaced with digital processing, in every place all over the world. Automotive is going full mental now, essentially turning cars into massive computers - Internet of Things is here.
- wealth; China is now at the point where it has feature and technological parity with the Western world. They have in-house development instead of copying, and they have a third of the world's population now gearing up to do as we have done for decades. Another major demand factor with a trickle effect to other Asian countries, while more developed Asian countries show no signs of decrease of demand that was already there - they have the same IoT development going on.
- shortages; chip production is stalling not because of fab capacity, but supplies to make them. The whole supply chain needs to satisfy the demand - not just lithography.