No they didn't. If they wanted to throw more silicon at Epyc they easily could have. Each chiplet is only ~80 mm2 and there is room left on the interposer, they easily could have made each CCD bigger but they chose not to.
Current generation of interconnect has been a challenge to route through the substrate with 8 chiplets. I don't know if it can be extended in the current form any further, since Genoa will not only require extra 4 to 8 chiplets, the bandwidth should ideally double again.
AMD will have to switch to completely new technology to go forward.
Do you have any calculations for this?
Cutress has a video titled cost of 7nm wafer, and in the middle there is a table showing cost of different technologies. 12/14nm was within 20% of cost of 7nm.
Going from logic die to N6/N7 SRAM die, the cost should be drastically lower than N7 logic die. due to fewer layers of metal and likely use of EUV resulting in fewer processing steps.
I'm not talking about the yield of the dice themselves but the yield of the stacking. We don't know what the yield is like through this process, but whatever it is, it is going to be significantly worse the more stacks that are added relative to 1 stack.
The question is: can one bad bond kill the entire chip or just 1 layer?
Most likely, it is only a single layer, so the yield is not going to be a big consideration.
BTW, it is funny that people keep mentioning 1 stack. 1 stack is a proof of concept, but AMD and TSMC are clearly not doing this for one stack.
No, it just means your costs go up to get working silicon which is the whole point. Companies could keep adding silicon to increase performance but they have to take into account not only the cost of the added silicon, but the hit in yields as the dice grow bigger. That's the whole reason AMD went chiplets in the first place. Stacking happens because continuing to grow horizontally on the substrate is no longer practical at some point both in power and cost, so stacking becomes the superior option,
That was my point. Intel had a poorly yielding process and only monolithic architecture making thing worse for larger dies. So throwing more silicon results in movement in wrong direction exponentially
AMD starts with very high yielding process and expands with even higher yielding dies.
but it will still be lower yielding than not stacking and stacking 4 hi will be lower yielding than 1 hi, etc.
At this point, a competent manager would ask you: Is it a material difference?
Most people in the know consider it to be a breakthrough technology, people NOT in the know are talking about yield difference between 1 and 2 stacks.
The reason SRAM is known for extremely high yields is the ease of implementation of redundancies.
AMD has technically been profitable since the beginning of 2018 but that notwithstanding, they still could have bought into risk production and just paid their debts will all of their income from chip sales. What you're saying, however, is that buying into risk production would have cut into their profits despite it giving them a superior product. Doesn't sound like a very good business decision, does it?
I think there is another limiting factor - manpower. There were layoffs, and AMD has been staffing back up,
It would be opportunistic to have, say a shrink of Zen 3 to N5, but AMD, with limited staff, has to be more methodical where to put their eggs, into which baskets.
BTW, AMD is doing risk production of SoIC with TSMC. So this one looked like a good bet to AMD leadership.
Nvidia also has sold a $3000 Titan graphics card and it was an absolute failure in terms of consumer success.
A/B testing. Through which NVidia found out it can charge $1,500 for a market leading consumer card.
Even Nvidia has limits as to what they can charge. We also have no idea how the 3090 would have sold if it weren't for the current market conditions and miners who are much less price sensitive than the regular consumers. You can't compare what graphics cards sell for now to pretty much any other computer part in history (except other GPUs during mining runs) as it is a unique situation and these cards are being driven up in price by people who aren't buying them as members of the target consumer market but are being bought as money making machines.
Intel used to charge $1000. And 5950x. is not that far from $1000.
I don't know where the limit is, but Zen 3D seems to be the ideal product to test the waters. Since it can go as far as 8 stacks high.
I doubt it will destroy Zen 3 but I guess that depends on your definition of destroy. I also said more than a year away from when ADL launches (actual launch, not announced).
There may be caveats, but after all the caveats, Alder Lake will be the best performing desktop CPU when you take a cross section of benchmark. And Intel will take the claim of best performing CPU from AMD.
(in absence of Zen 3D)
Again, I'd like to see your math behind this number.
About the AMD being able to add performance in $6 increments until it beats ADL?
The $6 is an extremely conservative estimate for the cost of 36mm2 SRAM die. I think Cutress floated it this one as well. It is more likely half that, and $6 including assembly.
Based upon your posts in this forum, I just don't think we will see eye to eye on how AMD will/should operate their business. We'll just have to wait and see how far AMD is willing to go and how much they are willing to spend to keep their 'mind share' as the gaming leader despite having only won this mind share less than a year ago and doing just fine without it.
It would be quite unimportant if you had a disagreement with a random poster on Internet. But you are not seeing eye to eye with AMD CEO about how to run their business: