Estimate a BOM delta for PS5 and XBSX

While I kind of like the mental image of Mark Cerny walking into an AMD engineering meeting and demanding AMD "take that crap out of my chip", it's generally more problematic removing logic blocks from existing designs rather than just ignoring the unwanted logic. You generally only do this if this if the issue if materially impactful on the overall transistor budget but if so, then you're already looking at a very large (and expensive) layout redesign. You don't do this lightly.

 

I stay behind my opinion that difference in BOM between the two will be ~15-20$.

Alot of SOC costs come from additional costs and not solely yielded mm² (~40%) so we could see 50mm² smaller chip but much higher clocked being very expensive once all is said and done, and not to far from bigger one.

As for everything else, probably too close to care (memory, SSD, packaging, cooling, PCB etc).

 

I could see Sony (arrogant Sony) selling the PS5 at higher price even if their box is less powerful.

 

This could be a good guess.

I just suspect the BOM difference wont be much or any in Sony's favor. The same as their probably wasn't much on One vs PS4 despite the former being markedly weaker.

The things I can really think of being a delta are:

-Soc, Sony's being cheaper.
-Cooling, Sony's presumably more expensive given the clock and Bloomberg article. HOWEVER that said, the Austin Evans video showed a hugeass heatsink for the Series X, and MS may overengineer it similar to how they did One X, so, who knows. We still nominally can charge Sony here maybe, Cerny mentioned super special cooling as well right?
-SSD, I guess Sony's more expensive? I'm not really sure, it's faster, but is that the silicon or the IO hardware?

I mean off the top that's two of three for Sony, not great.

Sony can engineer a "rougher", louder, less refined system and save there though, I think they pretty much did that with PS4Pro and it was probably the right move. This alone could probably make a big difference. We haven't seen PS5, but Series X looks large and refined/costly/quiet, similar to a One X perhaps.

The PRICE may be a different matter, with a slightly weaker GPU Sony may feel pressure to price lower, again, the bloomberg article.

 

Sony's SSD is potentially quite a bit cheaper. They can buy failed modules intended for 1TB units which will work for their 825Gb capacity and a custom controller, albeit with the overhead for design (a one-off cost), will be cheaper than licensing a controller which is a perpetual licensing cost.

There are a lot of variables on just this narrow issue.

 

Considering their controller is more powerful because it's 12 channels and 6 priority queues, this would make it more expensive in silicon area and custom might cost more than commodity... . And there's a lot of area of a nand chip which is for the interface, so having more chips cost more for the same total capacity. Maybe it breaks even from being smaller than 1TB? Or less?

It's 12x 512Gb chips, which adds up to 825 exactly. No idea about wear leveling margins.

 

What have I missed? Sony are not making NAND. How does this make any difference? The fabricated NAND has no concept of controller I/O channels. Sony will be buying NAND not complete NVMe solutions.

Sony's controller is - theoretically - more powerful because Sony have designed it that way at the expense of unneeded capability. If I design a quad-bus bus I/O architecture, that doesn't make the RAM more expensive only the bus/controller.

 

Smaller NAND chips costs more in $/GB if you want to make your controller really wide. There is decent bit of fixed cost optimized out for the NAND manufacturer in this case. It might be interesting if it will cost Sony more. A lot of low end SSDs come with 1 1Tb NAND chip and next gen NAND are going for 2Tb chips. If Sony wants custom made 512Mb chips in the long run, they will likely have to pay extra for it.

 

I am no longer in the industry [of buying any type of storage in bulk] but when I was we focused on capacity being on x times y, i.e. the NAND element part vs. the capacity we needed for any device. The industry is generally focused on producing storage for specific capacity points and there is a lot of product that falls slightly short, so if you're targeting 0.25 / 0.5 / 1Tb of NAND, drops below but may be suitable singular/multiply for 825G rather than 1Tb.

 

It depends on the brand of chip. A Micron "512gbit" is actually using a block size of 2048k+112k for wear leveling area, bad block remapping.

(old datasheet, but to illustrate)
• Organization
– Page size x8: 8640 bytes (8192 + 448 bytes)
– Block size: 256 pages (2048K + 112K bytes)
– Plane size: 2 planes x 2048 blocks per plane

It needs 12 chips to fill up the 12 channels, so it's obviously using 512gb chips ending up at exactly 825. There are no other possible configuration from any nand vendor.

512x1024x1024x1024 x12 /8 = 825G

 

Same bom, ps5s cooling solution is going to be very expensive.

 

Heat output on the PS5 is arguably less of a concern compared to the XSX than most people realize. At the end of the day there's no avoiding the fact that the higher-end system comes with a ~45% wider GPU and ~25% higher bandwidth which will linearly scale as much into power consumption so just the GPU alone will prove to be a challenge in terms of thermal management without accounting for architectural/process improvements or how aggressively binned these parts will be. That's not everything since even the CPU in the XSX is clocked higher as well and it has comparatively less liberty with downclocking too so that just means less options for the system to do thermal management which means that Microsoft would have to employ a more aggressive cooling solution than their competitor in the end.

 

Have you not looked at the actual system breakdown and cooling system employed by SeriesX?

 

The problem is that power consumption doesn't scale linearly with clock speed, especially if the clock speed used is beyond the "knee of the curve" WRT power.

It's safe to assume that the XBSX GPU core is below the knee of the curve. It's highly likely that the PS5 SOC is on the wrong side of the knee, hence necessitating variable clocks.

On top of that the XBSX SOC being large means there's more surface area available to transfer heat off the SOC compared to the smaller PS5 SOC.

Basically heat is a concern for both. We wouldn't be seeing that rather exotic V-shaped development box for PS5 if it didn't necessitate some interesting cooling solutions. In comparison the XBSX development boxes are housed in what are basically XBO-X boxes.

Regards,
SB

 

Which machine will be cheaper to make after first revision/ die shrink ?

 

PS5. With revision on newer node that TDP will fall down sharper and cooling requirement will be smaller.

 

Bloomberg article told us they are worried about it though. To refresh on that: https://www.bloomberg.com/news/arti...-with-playstation-5-price-due-to-costly-parts

To me there are two price scenarios vastly more likely than all others: They are the same price, or PS5 is less expensive by $50.

 

One can assume the PS5 is going to be cheaper, it looks like a 399 console, which makes sense as the PS4 at 399 was very successfull.

 

More than just TDP/cooling issues, a node shrink probably gives a little bit more frequency headroom, which lets them claw back the parametric yields they lost due to clocking so high. In the long term, PS5 has a definitive cost advantage. I just think that on release day, the advantage will probably be a lot smaller than people would think.