You showed the decrease in wafer cost yourself. I am not comparing it to the current Polaris cost.How can it not be more expensive when it is bigger in die size? It has at least 30% more xTors.
You showed the decrease in wafer cost yourself. I am not comparing it to the current Polaris cost.How can it not be more expensive when it is bigger in die size? It has at least 30% more xTors.
At the current Polaris price.It still will be more expensive to make. 10$ vs 13$.
And that is THE BEST CASE SCENARIO.
No, not at current Polaris Price.At the current Polaris price.
Wafer cost dropped more than 30%, to compensate for some lower yield and wasted space because of the bigger die size, but the process is mature and these are not huge chips.
Is that 575 theoretical or actual dies?No, not at current Polaris Price.
At the time Polaris was introduced, 14 nm wafers cost 6500$. From each one you got around 575 Polaris 11 dies. Each die cost you 11$.
Vega 12 made on 14 nm Wafer, at this point, where each wafer costs 4500$, each die costs AMD 13$.
If it would be made on that wafer when it cost 6500$ - the GPU die itself would cost 19$, almost twice as much, as Polaris 11 die.
I know you want that GPU to be cheap. But you are expecting GTX 1060 6 GB performance class GPU, with sub 75W power consumption/efficiency to cost under 150$, when GTX 1060 6GB costs twice as much.
It does not work this way.
Yes, but it is 2 years later.575 is the maximum number you can get from 300 mm wafer, with 123 mm2 die size.
The same case is for Vega 12. Maximum 333 working dies you can get from 300 mm 14 nm wafer.
And?Yes, but it is 2 years later.
Yields would not be so good two years ago, which is the Polaris comparison point.And?
When AMD released Polaris GPUs, 14 nm yield was at 92-94%. Currently it fluctuates between 95-98%Yields would not be so good two years ago, which is the Polaris comparison point.
It was actually 14 CUs two years ago.
That means Polaris 11 cost about $0.86 per CU. Vega 12 would cost about $0.70 per CU.When AMD released Polaris GPUs, 14 nm yield was at 92-94%. Currently it fluctuates between 95-98%
Worst case scenario: from that 300 mm wafer you get 575 working dies, and 24 dead dies. How good yield it gives you?
You forgot that HBM memory subsystem makes it more expensive, than RX 460, or any Polaris 11/21 GPU?That means Polaris 11 cost about $0.86 per CU. Vega 12 would cost about $0.70 per CU.
And Vega CUs would be more powerful than Polaris CUs.
It was actually 3 Vega CUs = 4 GCN3 CUs.
$12.15 for Polaris 11 and $14.07 for Vega 12.
But I guess your current yield was for Polaris, which is smaller.
Interestingly, the RX 460 4GiB launched at $119.
The Vega PCBs should be cheaper to make and fill.
No. But AMD does not enjoy good margins on Vega.You forgot that HBM memory subsystem makes it more expensive, than RX 460, or any Polaris 11/21 GPU?
Big Vega - should be alright. Vega Mobile is high margin part.No. But AMD does not enjoy good margins on Vega.
I don't care if it's high margin for Apple.Big Vega - should be alright. Vega Mobile is high margin part.
Don't expect low price for GTX 1060 6GB level of performance at much better efficiency.I don't care if it's high margin for Apple.
Yes, $400 for a P2000 is way too much if you don't need it for engineering.Don't expect low price for GTX 1060 6GB level of performance at much better efficiency.
Its desktop market. Nobody is stupid to sell that kind of performance, on old node, high manufacturing costs, very cheaply.
AMD can charge for this GPU 199$, and it will be a very good value, looking at what is on the market.
Im not talking about Professional versions of those GPUs.Yes, $400 for a P2000 is way too much if you don't need it for engineering.
I am talking about any 75W GPU which is not just a little faster than an RX 560.Im not talking about Professional versions of those GPUs.
Gaming oriented Vega 20 will cost 200$, in best case scenario, on desktop. Professional version will be around 400$, at the least, replacing WX 5100 as a WX 5200.
I have no idea how you worked that out...AMD should only get low margin on the HBM chips. Most of those profits would go to the memory vendor.
The packaging should not get good margin because it just competes against the PCB.
I am not talking about Apple laptops where cost does not matter.I have no idea how you worked that out...
Vega is a premium product, and will only land in premium Laptops.I am not talking about Apple laptops where cost does not matter.