Not really. AMD's APUs, especially mobile are cut down versions and not similar to desktop chips. Meanwhile, Intel basically lowers TDP and that's mobile chip. nVidia used to rebrands their mobile chips, use different dies and in other ways cripple GPUs and thus they weren't similar to desktop parts.
APUs are not GPUs. Nor are Intel CPUs GPUs. Also, what you're describing is (a slight misrepresentation of) how the chip industry has always operated: any chip that can be used for multiple purposes is used for those purposes as long as it makes sense economically. AFAIK, Nvidia has never used "different dies" for mobile chips (outside of a few rare edge cases). Chips are binned during production and different bins are used for different purposes.
Imagine they just integrated adapter part into board and you can't literally see MXM card.
What you are describing is a PCIe dGPU AIC. Literally nothing else than that.
BTW MXM is dead, but they still use some kind of interconnect for mobile chips.
Yes, it's called PCIe. MXM is dead because essentially all dGPU-equipped laptops have the GPU integrated directly into the motherboard.
It's that converted for desktop usage, but you clearly see how the GPU itself is mobile chip due to various bizarre limitations.
Okay, the problem here is that you're mixing up two quite different understandings of "GPU" - the one denoting the die itself, and the one denoting a full implementation including RAM, VRMs, and other ancillary circuitry. It is entirely true that this GPU - the die, and its physical characteristics - is primarily designed for mobile use. I've gone into this at quite some length in both this thread and others. What
isn't made for mobile use is its desktop implementations. And, crucially, just because the die is primarily designed for mobile use doesn't make it any kind of special case - as I said above, chips are binned and used for different purposes. That's how the chip industry works.
What people are discussing in terms of implementation, which you keep derailing with this odd "it's like an MXM adapter" nonsense, is
the design choices made by AMD when this die was designed, and the tradeoffs inherent to this. When I and others say it was designed for mobile, that means the it has a featureset balanced against a specific use case, and has zero overprovisioning for other use cases (mainly desktop, but also potentially others). The board implementation is essentially irrelevant to this, and it doesn't relate in any way to MXM adapters, as - as you point out - MXM is dead, and isn't relevant to how mobile GPUs are implemented today.
So: what we have here are completely normal, entirely regular desktop AIC GPU boards with nothing particularly unique about them, designed around a die that seems to have its design goals set for a highly specific mobile use case with some inherent major drawbacks due to this.
Power consumption is mainly decided by amount of cores and their voltage/clock speed.
Jesus, dude, seriously? Do I need to spoon feed you
everything? Yes, that is the main determinant for power consumption. But if your goal is
as low power consumption as possible, then you also start cutting other things. Such as memory bandwidth and PCIe, which both consume power - and in a 25W mobile GPU, even a watt or two saved makes a difference. These are of course
also cost-cutting measures at the same time. As I've pointed out plenty of times: it's clear that this die is purpose-built for
cheap, low power laptops with entry-level gaming performance.
I think that you are overestimating difficulties. You can literally saw-off those pins and card will work just fine. There were 4X cards in the past and they didn't cost any extra compared to 16X or 1X cards. Motherboards even had 4X slots.
I never said it was difficult. I said it takes time and work to remove them from the design; time and work that costs money while providing no material benefits, and savings so small that they don't matter in the end. Thus the simplest and cheapest solution is not to bother doing so.
I read everything, either your English failed or you said 6400 and 6500XT have decoding. What else would your question "How many more options do you need?" imply here, in case of non recording capable cards?
They were literally saying that there are so many
other options that having
one option without them shouldn't matter. That was pretty easy to understand IMO.
Nothing much, but explains why 6400 and 6500 XT are so bizarrely limited.
But you're misusing the term "harvested" here. "Harvested" implies they are chips that failed/were binned too low for some use case. That does not seem to be the case here - the 6500 XT seems to be a fully enabled die, but clearly one that is binned for high clocks rather than low power. The 6400 seems to be a middle-of-the-road bin, with a few CUs fused off. Neither appear "harvested", as there's no major cuts made (4 CUs isn't a lot, and everything else is intact). They're just different versions of the same chip, all equally valid implementations, and none bear the "we'd rather use these somehow than scrap them" characteristic for harvested chips.
That's not exactly what I meant. It's just about artificially creating demand and selling low end poo-poo for huge premium. It's not making money, it's straight up daylight robbery. AMD pulled the same crap when they launched Ryzen 5600X and 5800X with huge premiums, when Intel sold 20% slower parts for literally half the price. And they dared to claim that 5600X was some value wonder. Only to release 5600, 5500, when people realized that Intel has some good shiz too. They also intentionally didn't sell any sensible APUs, only 5600G or 5700G, also for nearly twice what they were actually worth, but fanboys didn't question that and instead bought as much as AMD managed to make. Had they released 5400G (hypothetical 4C8T rDNA 2 APU), it would have outsold 5600G/5700G by times, but why do this, if they can artificially limit their line up and convince buyers to buy way too overpriced stuff instead? That's exactly why I call this toxic capitalism, because goods can be available, but companies don't make them, due to lower, but still reasonable margins. If you look at their financial reports, they made an absolute killing during shortage and pandemic, so that basically confirms that they had huge mark-ups. That also explains why RX 6400 and 6500 XT lack features, lack performance and are overpriced, crappy products. 6500 XT is literally that poo, that RX 570 is equivalent of it, but RX 570 was made ages ago, wasn't gimped and cost way less. Even with inflation included, there's no way that 6400 and 6500 XT must cost as much as they do. Their mark-up is as high as 40%, if not more.
While I mostly agree with you in principle (though I'd go a lot further than you do here on some points), I think this is not the most suitable case for this critique. Navi 24 is more of a strangely unbalanced design than it is a cash grab - a cash grab would try to lure people in somehow, while this just behaves strangely instead. A cash grab needs to present itself as good value or enticing, which this doesn't do - and AMD doesn't have the GPU market mindshare to really change that. Its high entry price (on the desktop) takes this from bad to worse, but that's universal across the GPU market right now, and in no way unique to these two GPU models. And yes, AMD, just like other major corporations, are in it for profits first and foremost - as a publicly traded US corporation they are legally obligated to do so. This system is an absolute travesty in so many ways, but it plays out in much more nefarious ways than these unbalanced GPU designs.
There are also valid reasons for
some cost increases - compared to even two years ago, many raw materials (copper, aluminium, many others) now cost 2-3x what they used to. International shipping has also increased massively, which also affects MSRPs. Neither of those are sufficient to explain the inflated MSRPs of these or other GPUs on the market currently, but they go
some way towards explaining them. Another explanation is AIO partners resisting scraped-down marigns - many GPU AIO partners have reportedly had essentially zero margin on their products, with MSRPs previously being so low they have struggled to break even on their own designs. Again this doesn't necessarily justify increased MSRPs -after all, the GPU makers themselves have very high margins overall, in the 40% range.
come for the ""utilisation of ray tracing games to proceed"" !!
after all ??