The US did it did they had to pay a lot for that ?If the US one opens couldn't AMD also produce there if they pay like they do with TSMC?
Sure. I would guess that TSMC USA will cost more than TSMC in Taiwan, American labor, etc. TSMC is basically the reason Biden said they would protect Taiwan with military actions if China stepped in, semiconductors are now that important, it's crazy.
Are you really trying to make the case that AMD isn't actively attempting to increase market share? That is ludicrous.
All you have done is demonstrate that Nvidia is indeed the market leader. That AMD has to be more competitive in price/perf. That it takes multiple GPU generations to make a dent.
They have better high margin businesses than trying to make a GPU price war against a company that has much deeper pockets than they do. They are competing on two fronts with Intel (1.2x AMD's market cap) and Nvidia (3.5x AMD's market cap). Intel seems to be very aggressive on pricing which will likely force AMD to slash prices of the 7000 series. Why would AMD start a price war with Nvidia? Even lower margins? Why did they not slash RDNA 2 prices way deeper? They were in a very good position, especially with Nvidia being on a weaker node at Samsung.
It's not that they would not want to increase market share, but that they are not on the same level playing fields. Nvidia is not sleeping at the wheel like Intel was for years. Even with Intel literally handing the market on a silver platter, AMD managed to go from 17% to 32% over 5 generations of the popular Ryzen. Making a dent is in GPU market is being optimistic.
Don't forget that there was a crypto mining boom affecting cards getting into consumer hands. Also worth noting is the fact that AMD didn't release the full GPU stack for a few years, not to mention Nvidia went with Samsung 8nm and potentially didn't face the same shortage issues. Nvidia hard countered upon the launch of 6000 series with a full lineup of Super cards. All the above are factors at play.
Crypto mining affected everyone, not sure what the point is? Are you trying to say that there's a ton of 6000 series cards out there, but they were sent for Crypto? That same crypto that favored fast VRAM such as 3000 series? Nvidia used Samsung with worse yields and countered with super, yes, so what? What's the angle here, that AMD fumbled their chance for RDNA 2 to dominate the market?
You should listen to Lisa Su discuss the desire for increased GPU market share in one of her stockholder meeting/financial analyst days. You have a point regarding the difficulty in attaining fab time for chips but you have pushed that idea too far.
Maybe AMD can go with Samsung to unbottleneck, seems like it was an advantage to Nvidia..
Yeah they won't have a huge supply.
Not because they can't get it - they absolutely can and they have. They'd just rather allocate more of that capacity to Server CPUs.
Also what do you mean "Law breaking specs".
What laws are being broken exactly?
Peoples are expecting a monster MCM with a die roughly equal to a 4090 in area, but non of the negatives. MCM is automatically gonna pull more heat compared to monolithic. MCM is going to add latency. More chiplets, the more crossbars, the more the data has to make a jump at a node, it's the basics of NUMA topology. "B..b.. but Ryzen?" you say, CPU tasks not sensitive inter-GPM bandwidth and local data to latency like GPUs are. AMD's MI200s and Nvidia's (2) H100 chipsets were MCM and were made for tasks with low latency requirements such as scientific computing. NVlink's 900GB/s and MI200s infinity fabric's 100GB/s per links with 8 links providing 800GB/s, are still no match for the whooping 2.5TB/s Apple made for the M1 Ultra. That 2 chipset MCM basically had double CPU performances, while GPU had a +50% increase on their own freaking API! Because don't forget, this segmentation of tasks that are ultra sensitive to fast local packets of data such as FSR/RT/ML will have to be entirely invisible from the API's point of view and since we're on PC, it's on AMD's shoulders to make drivers for that.
What else is rumored.. oh, let's add 4GHz into the mix, 100 less watts, match or surpass a 4090 in rasterization, expensive communication crossbars that are outside of lithography but still manage -$600 over competition. Pulling performances out of a quantum parallel universe basically.
AMD is simply magic
Or maybe
something will have to give. You can't stretch the limits in performances and have better of everything, something that tech youtubers don't seem to understand.
Xbox and PS5 silicon comes from AMD's allowance, under their whole "Semi-Custom Solutions" arm, and that'd be only N7 for now. I assume they only need to meet contractual obligations there.
And what an obligation. These consoles are juggernauts of manufacturing, not even the full market share of the Ampere/RDNA 2 would come close, and they'll continue to have huge supply requirements for years to come. Add in some pro/slim models on top of it, already signed and reserved.
The node has little to no impact nowadays in allocation unless you want to compete against Apple. The problem is shortage of silicon wafers. The lithography part is not really the problem. TSMC's 300mm wafer is competing with customers that will process it with a range of 2 microns to 5nm. They still have to send some wafers so that everyone get some fucking wifi with a screen on their fridge/toaster/espresso machine/washer/cars.
I'm almost certain AMD are also the second biggest customer for TSMC's N5, behind Apple. "Bidding wars" for contracts happened in the past for current nodes. Yes, trying to take the whole capacity would be expensive and unlikely - as would building a new fab in collaboration with TSMC. They are definitely split a lot of different ways with the large capacity they have and realistically they're going to shift whatever they can in the highest profit segments.
Exactly.
Not having a huge supply != doing a paper launch. Also AMD have reportedly already cut back on 5nm reservations for their CPUs, and Navi 33 will be on 6nm, which saves more 5nm wafers.
AMD snarked at Nvidia's supply problems by saying they wouldn't have a paper launch, just to have the 3090 wipe the floor with the whole RDNA 2 lineup. What's AMD's definition of a paper launch then as they said Nvidia had one? I would be curious.
As i said above, 5nm, 6nm, 7nm, doesn't matter. Silicon wafer manufacturers can't provide fast enough for the demand. Some of TSMC's biggest suppliers are booked up to 2026. The lithography is not the biggest bottleneck here unless you want to push Apple's out of their premium node.
Hope the laptop doesn't melt.
, at least the other 2 put some budget on their projects when they sell a remaster for 70 dollars
