Read more.Source says that AMD could dualize some lines, as TSMC is so busy with Apple etc.
Read more.Source says that AMD could dualize some lines, as TSMC is so busy with Apple etc.
Samsung cannot meet what nvidia need for supply, so how are they going to manage with amd as well
What makes you think that? Nvidia not being able to meet demand does not mean Samsung are at max capacity for the node. Large ICs like GPUs have lead times extending into several months, and then you have to factor PCB manufacturing, assembly and transport on top of that.
Chip development architectures cannot be 'ported' easily; 8nm Samsung and 7nm TSMC are radically different (INFACT Samsung is not better) and the costs involved to a new FAB design are very high, its better to wait for TSMC to catch up (they are building new facilities). LETS NOT FORGET AMD has GlobalFoudries who can use TSMC licence (speculation), maybe its GF who want to use 8nm Samsung licence- they did that with 14nmLPP.
I presume they're talking about 5nm, since that's what apple etc. are on.
Only part of the issue for Nvidia is Samsung, they've had issues with GDDR6x supply, even vanilla GDDR6 supply is woeful, they've had issues with their cooler supplies for their reference cards as the cooler is quite specific and hard to manufacture, then there are pcb supply issues, etc...
So you add all things up and you have parts arriving at different times, you had covid19 so a lot of delay there as well, so by the time Nvidia receives all parts its probably at least 1 month delayed. So then you have the time for assembly, for testing, for shipping, etc...
AMD's woes mostly come from limited TSMC 7nm production, but they are also having some issues with GDDR6 memory being supply limited.
You're all missing the boat. I've been experimenting with those phone sanitiser boxes and potatoes.
EUV becomes PUV. And it's going to be awesome.
Actually.... I wonder what happens if I take a potato and expose it to lots of UV....
Of course. It will be a case of weighing up redesigning products for another node against potential losses for restricted supply and considering offset of likely lower per-wafer costs (if volume is high enough to offset initial costs). That's assuming 8nm, but as kalniel says it could be more a case of planning for future products, particularly as the source mentions pushing against Apple for capacity (which could imply 5nm). And this wouldn't be an overnight switch anyway, even a backport of 7nm to 8nm for a complex product like CPU/GPU would take many months, by which time we might be seeing early ramp of 5nm anyway.
I wouldn't be surprised if the consoles were designed to be fab at more than one company. IIRC,this was the case with previous consoles. If this is indeed the case,AMD might already have done tests for CPU/GPU production at other companies. I remember CD from Semi-Accurate many years ago wrote an article about AMD Jaguar,and how AMD had designed it to be more easily ported to different fabs and process nodes. I suspect it might have been one reason why the console APUs used AMD Jaguar(and also its smaller size in MM2 compared to the bigger cores).
Yeah the Jaguar core was used across GloFo and TSMC. I think even some of their PC processors based on Jaguar were made across foundries depending on which SKUs you bought.
I forgot to add to my previous post. Regarding GloFo being a licensee or adding capacity for AMD; I don't think it's the most likely scenario. Part of the reason they abandoned their own 7nm was ostensibly down to floor space - they were already close to utilising what they had available and a shift to 7nm would have meant more new build (so more capital and waiting years) or converting existing floor space, reducing existing capacity in the interim and losing money.
Amusing to read this after those rumours that Ampere launching on Samsung's 8nm was only a stopgap and the plan was to move to TSMC in 2021.
There are currently 1 users browsing this thread. (0 members and 1 guests)