Back in 2014-2015 an AMD rep said that the cost of a CPU design was about 300M$, so a 700M$ figure 10 years later sound right.Those are fake numbers for design costs.
Like they're not real, this (wrongly) assumed the vendor does everything in house. Which no one does.
There is R&D marketing and othe things as well it's not like they have milked the design as of now also advanced packing I think Zen6 will have it so it will increase cost as well.Even if we take your design cost numbers as gospel (which I don't, as adroc points out that makes non real world assumptions) you're talking a few hundred million dollars divided across all the chips. How many Zen 6 CPUs will AMD ship in total? Sounds like a couple bucks per CPU, which again is inflated because we're using those bogus IBS numbers.
Yeah I know the die won't be ~100 mm2 I expect 70-85mm2As for the wafer cost figures, remember I did say quite clearly I was making up numbers so the math was easy/clear, but it doesn't affect my point. You might as well point out that Zen 6 dies won't all be 100mm^2 while you're at it...
What is your estimate of annual 9800X3D sales?Means the margins are obscene because the volume is tiny.
You can't just scale hybrid bonding capacity 10 times, that's a ton of equipment.
Ah right, your whole argument this time has been about the word delay. Fair enough.
Glad we agreed in the end that a hypothetical launch of X3D chips available from the start would mean a later theoretical launch than if they didn't launch x3D chips at the start. The only way to do this is to launch ( I used the word delay) the lunch of non X3D chips.
You're right, it does mean unsold inventory and it does mean a delay in revenues. Both quite unpalatable.
You do understand that DIY CPUs have obscene margins, don't you?
Means the margins are obscene because the volume is tiny.
No it means the volume is tiny.
You can't just scale hybrid bonding capacity 10 times, that's a ton of equipment.
So it's that cheap, meaning it can't be too slow and expanding capacity can't be too expensive.The cost premium to add V-Cache can be somewhere between $10 and $20 (including packaging)
Exactly - even with the highly desirable line like X3Ds - they are just waiting for Intel to launch a competitor next yearAMD has been bitten so many times that they are afraid to go out on the limb and anticipate success and place larger orders.
It's TSMC that would be increasing the capacity and they probably don't want to risk a large, sudden increase due to the fact AMD is their only customer using SoIC-X.What is your estimate of annual 9800X3D sales?
9800X - $479 (MSRP)
9700X - $359 (MSRP)
Which one will sell better?
Let's say scale it 2 times, what's your estimate of the cost - $1 bln? $5 bln? $10 bln?
AMD is a more privilege customer to TSMC than Intel though and is behind only Apple and Nvidia.It's TSMC that would be increasing the capacity and they probably don't want to risk a large, sudden increase due to the fact AMD is their only customer using SoIC-X.
It's still risky to build capacity for just a single customer.AMD is a more privilege customer to TSMC than Intel though and is behind only Apple and Nvidia.
Its not as risky when that customer is up and coming and gaining market share quickly.It's still risky to build capacity for just a single customer.
TSMC is very good at converting to Risk to Price PremiumIts not as risky when that customer is up and coming and gaining market share quickly.
i am pretty sure they had it cause they went to N2 skipping N3PTSMC is very good at converting to Risk to Price Premium
The question is does AMD have the stomach for this premium 🤔
There is tried and tested way to zero out that risk, TSMC is doing it all the time - pre-payment in full. This isn't N2 fab, costs should be far far lower.It's still risky to build capacity for just a single customer.
Rhetorical question...The question is does AMD have the stomach for this premium
It's TSMC that would be increasing the capacity and they probably don't want to risk a large, sudden increase due to the fact AMD is their only customer using SoIC-X.
How are we having this chat?It doesn't have to be either.
Suppose regular chip takes 2 months to complete and X3D takes 3 months. It can just mean that you start producing X3D at T - 3 months and regular at T - 2 months.
Or in other words, the first chips that come out (at wafer start + 2 months) as regular CPUs, send them to X3D conversion.
A month later, you will have chips coming out as both regular and X3D, and you can launch. No delay in revenue, no unnecessary inventory.
It doesn't have to be either.
Suppose regular chip takes 2 months to complete and X3D takes 3 months. It can just mean that you start producing X3D at T - 3 months and regular at T - 2 months.
Or in other words, the first chips that come out (at wafer start + 2 months) as regular CPUs, send them to X3D conversion.
A month later, you will have chips coming out as both regular and X3D, and you can launch. No delay in revenue, no unnecessary inventory.
Agreed. I wanted to add that I highly doubt AMD would go X3D cache-only any time soon, even on premium chiplet CPUs. There is still a pretty large market for non-X3D cache variants, and they are cheaper to make.Using your terminology. If T is the time the first that non-xcd chips can launch. While I don't agree with your lengths of time for each process I'll use them to make this easy. First batch in production starts at T-2. Second batch at T-1. First batch ready at T. You launch. It you now have to add x3D that first batch is ready at T+1 and is ready alongside the second batch ton be sold also at T+1.
I came across an article that said Apple may be using SoIC-X in not too distant future. But the article was not clear if it is the same implementation as AMD or different.
How are we having this chat?
I'm not saying you can't launch both at the same time, but the theoretical time you can do that is later than the theoretical time you launch without X3D because the process takes time. It's that simple.
Using your terminology. If T is the time the first that non-xcd chips can launch. While I don't agree with your lengths of time for each process I'll use them to make this easy. First batch in production starts at T-2. Second batch at T-1. First batch ready at T. You launch. It you now have to add x3D that first batch is ready at T+1 and is ready alongside the second batch ton be sold also at T+1.
You can't just hand wave this away in the way I think you dis in your response and start everything earlier. You can't just make the design process quicker so you can launch both at time T.
The length of time you have a delay (vs what you could have launched at) is the length of time it takes to process the x3D chips from a non x3D chip. In this example, one month.
This also assumes away the lower capacity of making x3D. Which means you need lots more time before you can launch both. It won't happen until there is lots and lots of packaging capacity, capacity that isn't available to AMD to my knowledge.
No, we don't. You make a lot of 10 word "drive by" comments but never fully explain yourself.You do understand that DIY CPUs have obscene margins, don't you?
It seems, from the marketing POV, there is a great value in introducing the new generation of the product by launching the strongest product first. Good impression from that initial introduction improves the sales of lower tier products of the same generation.
In case of AMD, X3D products are those strongest products.
Earlier time of introduction of a weaker product seems to have minimum to negative value.
Cases in point:
- Zen 4, non-X3D
- Zen 5, non-X3D
Zen4, really ?
You memory might be a bit defective, IMO.
The reason behind Zen 4 slow sales was mainly the horrible premium asked for AM5 + DDR5 compared to AM4 + DDR4.My memory is very good. Zen 4 launched H2 2022 and sales were very disappointing / underwhelming from original launch until 7800x3d, following year in April 2023.