Sony Reportedly Working on Redesigned PS5 SoC on 6 nm for 2022

[Raevenlord]

It's not only graphics cards and CPUs that are best kept on the edge of manufacturing processes; in truth, one could even say that consoles have more to gain from these transitions when it comes to their manufacturers' financial outlooks. This happens because usually, consoles are subsidized by manufacturers in that their actual retail price is lower than manufacturing costs; this works as a way for console players to increase their platforms' attractiveness and user base, so they can then sell them games and subscription services, where the big bucks are actually made. We knew this already, but Microsoft's head of Xbox business development, Lori Wright confirmed it yesterday at the Apple vs Epic Games hearing. Lori Wright is quoted as answering "We don't; we sell the consoles at a loss" when asked whether Microsoft does or does not turn a profit on Xbox Series S | X hardware sales.

Considering the similarities between the Xbox Series X and PS5's SoC, it's very likely that Sony doesn't make a profit on console hardware sales either - or if it actually does, it's nothing actually meaningful. This is part of the reason why consoles are usually actually in the forefront of manufacturing processes' advancements, as it's a way for console players to quickly reduce the BoM (Bill of Materials) for their consoles. Since the specifications don't change within a console generation (discounting Pro models, which both companies have taken to launching some years into their generations), they choose to take advantage of process advancements due to the transistor density increases that allow for both lower silicon area for the SoC, and lower power consumption - which sometimes enables them to develop slim versions of their gaming consoles.





This brings us to the core of this news piece, in that Digitimes reports from industry sources that Sony is aiming to take advantage of TSMC's 6 nm manufacturing process for their PS5 consoles come 2022. The move from the current manufacturing process - TSMC's 7 nm - down to N6 would mean an up to 18% increase in transistor density, meaning that the SoC would use a smaller area for the same computing resources, thus improving Sony's margins on each PS5 console sold. Considering the company already moved some 7.8 million consoles since its launch, even a $10 difference would mean a cool $78 million either entering Sony's coffers, or simply not leaving them in the first place. This might also open the door for a price reduction on the console, of course, but considering the demand/supply ratio, Sony might as well just keep the money in its pockets.

View at TechPowerUp Main Site

 

[Valantar]

Given the high costs of cutting-edge nodes these days it sounds rather unlikely that this would net them any major savings. Given that the PS5's APU is ~23.367*13.184mm=308.070528mm², an 18% decrease in area (and assuming no major rejiggering of components on the die) would make it ~252.6mm² and thus ~11.938*21.159mm. With the 7nm wafer yielding ~131 good dice and ~41 dies with some error (mostly likely salvageable with extra CUs on the die), divide that by a wafer cost of ~$10 000 and we have ~10 000/172=$58.14 per die currently. Assuming similar yields on 6nm as 7nm, that gives ~176 good dice and 44 faulty ones, or a total of 220 dice on a 6nm wafer. At the same wafer cost that would result in a price of 10 000/220 = $45.45 per die - but 6nm is bound to be more expensive than 7nm. Let's say it's 20% more expensive per wafer - that's $54.55/die. That's a savings of $3.6/die - not much at all. Of course that wafer price is purely made up, but it goes to show just how quickly these savings get eaten by cost increases.

If anything, a die shrink is likely to happen due to other savings:
- Thermals: The PS5 runs hot, has a pretty bad cooling design overall (it's quiet, but only due to a stupidly overbuilt and likely expensive cooler, due to its inefficient design). This can let them shrink the cooler, make it cheaper.
- Power draw: Even a 10% drop in power for the same performance can net them notable savings in speccing a lower output PSU.
- Ditching liquid metal for conventional thermal paste?
Etc., etc.

 

[HisDivineOrder]

I thought I read a story recently about Sony redesigning the PS5 to reduce their reliance on certain parts that are now hard to come by. If anything, I expect they're doing that and just taking advantage of a mild die shrink at the same time.

 

[Valantar]

I thought I read a story recently about Sony redesigning the PS5 to reduce their reliance on certain parts that are now hard to come by. If anything, I expect they're doing that and just taking advantage of a mild die shrink at the same time.

Hm, that sounds difficult - unless there are some very specific shortages that aren't widely communicated? It's not like they can design themselves away from using silicon chips or substrates after all. Though I guess it might be that they're short on specific ICs or something like that?

 

[defaultluser]

Hm, that sounds difficult - unless there are some very specific shortages that aren't widely communicated? It's not like they can design themselves away from using silicon chips or substrates after all. Though I guess it might be that they're short on specific ICs or something like that?

The biggest cost reduction everyone is expecting them to do is drop the GDDR6 chips down to 8; this year, those are as rare as hen's teeth (but in anther year, I could see the availability become wide enough to actually buy a 6000-series GPU at msrp!)

A die shrink on 6nm would mean a relatively simple die revision - it's just an optimization of their 7nm rules. There is no way they are going to get 5nm at a reasonable price anytime soon (so this is the best compromise for cost reductions on an expensive system).

Remember, the PS3 was such a monster, they went through multiple die shrinks before they made it affordable...PS4 didn't have to worry about that!

 

[Valantar]

The biggest cost reduction everyone is expecting them to do is drop the GDDR6 chips down to 8; this year, those are as rare as hen's teeth (but in anther year, I could see the availability become wide enough to actually buy a 6000-series GPU at msrp!)

A die shrink on 6nm would mean a relatively simple die revision - it's just an optimization of their 7nm rules. There is no way they are going to get 5nm at a reasonable price anytime soon (so this is the best compromise for cost reductions on an expensive system).

Remember, the PS3 was such a monster, they went through multiple die shrinks before they made it affordable...PS4 didn't have to worry about that!

There's no cost reduction in halving the amount chips if capacity is the same though - double density chips literally have twice the silicon inside, and that's what costs money. And I haven't heard of any RAM makers moving GDDR6 to a smaller node recently, so any savings there sound unlikely. They'd also struggle to maintain bandwidth with double density chips, unless the chips were just two chips packaged as one, in which case I can't see how there would be any savings at all.

This generation of consoles is in a way a return to normal - while the previous gen used hardware that was lower midrange even at launch, most consoles have been pretty cutting edge before that. (Though the PS3 did manage to be noticeably slower than the X360 even though it launched a year later, for some weird reason.)

 

[defaultluser]

There's no cost reduction in halving the amount chips if capacity is the same though - double density chips literally have twice the silicon inside, and that's what costs money. And I haven't heard of any RAM makers moving GDDR6 to a smaller node recently, so any savings there sound unlikely. They'd also struggle to maintain bandwidth with double density chips, unless the chips were just two chips packaged as one, in which case I can't see how there would be any savings at all.

This generation of consoles is in a way a return to normal - while the previous gen used hardware that was lower midrange even at launch, most consoles have been pretty cutting edge before that. (Though the PS3 did manage to be noticeably slower than the X360 even though it launched a year later, for some weird reason.)

Yeah, the initial press releases said they were going to use `16 1gb gb dies, but if ypure going stacke packages (just looked at tear downs)d, it's virtually the samprice!

\

 

[Valantar]

Yeah, the initial press releases said they were going to use `16 1gb gb dies, but if ypure going stacke packages (just looked at tear downs)d, it's virtually the samprice!

\

Not only with stacking packages, but with double density packages as well. Why? Because it's all based on the same silicon. A 2gb package just has twice the silicon of a 1gb package. It's not like 2gb chips are magically 2x the density of 1gb chips and thus have the same silicon area, after all. Sure, packaging costs are 1x instead of 2x, but those are negligible in the grand scheme of things. There is very, very little price scaling for increased capacity in memory.

 

[Raevenlord]

There's no cost reduction in halving the amount chips if capacity is the same though - double density chips literally have twice the silicon inside, and that's what costs money. And I haven't heard of any RAM makers moving GDDR6 to a smaller node recently, so any savings there sound unlikely. They'd also struggle to maintain bandwidth with double density chips, unless the chips were just two chips packaged as one, in which case I can't see how there would be any savings at all.

This generation of consoles is in a way a return to normal - while the previous gen used hardware that was lower midrange even at launch, most consoles have been pretty cutting edge before that. (Though the PS3 did manage to be noticeably slower than the X360 even though it launched a year later, for some weird reason.)

The consensus on that seems to be on the programming difficulty of its CELL CPU. The world wasn't ready for highly-paralellized workloads running on ten cores.