Keep in mind that there's every reason to think that the build costs for Sony and MS are pretty close, and Sony might even be higher, given the SSD and MS possibly having more leverage for bulk discounts, that MS has far more money to throw at gaining market share and far more reason to do it than Sony does, and that they still opted for 500/300. Clearly MS, at this point, is fairly confident that Sony is not going to be priced near 300, or they would have opted to take a bigger loss themselves. PS5 still does really well at 500/450. Definitely outsells the XSX. Maybe loses some market share in the US and UK to the Series S. Why throw money away?
How to sell next-gen consoles, Marketing, Positioning, and Pricing [2020]
- Thread starter steveOrino
- Start date
Johnny Awesome
Veteran
I'm also thinking Sony has the confidence to go $499/$449, but they might go $499/$399. They'll make up the difference in digital revenue fairly easily.
D
Deleted member 11852
Guest
We don't know what's inside PS5 cooling-wise yet, but I don't think the BOMs are close. Sony will almost certainly outsell Microsoft again (particularly as Microsoft rally don't seem to care about hardware sales) and if Sony do produce/sell more there footing begins will be a considerable economics of scale advantage.
I'm interested in what bulk buying advantage you think Microsoft have? Sony produce not only more consoles with optical drives in, but make Blu-ray and AV systems with those in as well. Their phones, TVs, high-end cameras and medical devices all pack solid-state memory and conventional memory. Sony are hundreds of millions of hardware units with these in annually so they probably have a lot of price-negotiating leverage.
Fundamentally, PS5's APU is smaller, so every wafer is proving more viable APUs which is a considerable saving. PS5's SSD is smaller and faster, which probably cancel each other out but this is the kind of thing that will cost reduce quickly as faster solid state becomes the norm.
In terms of the the economics of the launch, it goes beyond the BOM into user investment metrics. Sony are pushing the need for the new DualSense controllers, which are huge profit divers. Idiots like me will be picking up 2 extra controllers which basically amortising any PS5 cost subsidiation Sony would eat anyway. Plenty of folks will pick up at least one extra controller. Sure, Sony eat the profit 'loss' but now I'm a PS5 sure for six to seven years.
It's never been so true that Sony and Microsoft are taking different paths.
To compete against XSS could Sony (using the same SoC) create downclocked 7TF discless 512GB SSD version of PS5 with 8GB GDDR6 (entire 8GB available to devs) + 1-2GB DDR4 connected to the southbridge for the OS (similar to the 1GB in PS4 PRO)?
Sure switching between games and the OS might be slow but thats not a horrific sacrifice to get a system still with more memory bandwidth and larger GPU than the XSS, that Sony could manufacture for maybe $50 higher than the XSS (since the only sizable difference would be the SoC cost?)
(cooling & power at 7TF might only be an additional $10 extra over XSS)
Sure switching between games and the OS might be slow but thats not a horrific sacrifice to get a system still with more memory bandwidth and larger GPU than the XSS, that Sony could manufacture for maybe $50 higher than the XSS (since the only sizable difference would be the SoC cost?)
(cooling & power at 7TF might only be an additional $10 extra over XSS)
Last edited:
I'm not sure why I would put someone on an XSSX blade versus giving them tenancy on a 128 core server, the costs are all out of line. Unless I missed something there are no virtualisation hooks in the XSSX GPU to allow the GPU to be shared across multiple VMs per host. So now I have a single tenant per blade, a situation that is normally reserved for instances with significantly more GPU on tap for CAD/CAM users et al (and the GPU is not certed for any of those apps so no).
I'm aware of Microsoft's comments I just can't see where the hell it slots in, for the record I sold virtualisation and workstation solutions into the CAD/CAM for over a decade so I'm not pure guessing here. For the majority of my customers the only justifications for very low density virtualisation were security or remote access, as the cost per seat was significantly in excess of buying them a big tower box per desk given the very low levels of concurrent usage. For all the bloviating you see from Citrix et al about how their remote solutions are "just like a desktop" the pricing and savings they tout are from deployment scenarios where users are screaming about web pages taking 5-10 seconds to laod on a good day.
I guess it's a question of bundled discounts vs bulk discounts.
The most expensive parts of the BOM are:
APU, Memory, SSD. And MS needs these for their Azure builds regardless of xcloud. The deal with Seagate is probably not coincidence, they are also likely a supplier for them on Azure. Memory provider will now need to provide for Azure because of xcloud, and AMD is a major partner of theirs.
So if you're going to piggy back on Azure bulk orders for Epyc and the supporting contracts for memory and storage; it could very well to be to your advantage; the bundling may provide some preferred rates that even though they didn't hit bulk numbers on, it's made up through bundling.
The other parts they may not do as well on.
I think it would be close between the two. Even if it were, I still think Sony will be willing to price at 499 to show that they compete at the same performance level as XSX.
There should also be trade-in deals at the time? I sort of forget if they allow for TIV Day 0
Last edited:
Cheaper. Azure allows you to rent a large variety of facilities ranging from a single Dual Core Intel up to DGX. There may be some services that are better aligned for GPU compute and others that require CPU compute. For the most part these GPUS are fairly competent at both, a single instance is enough for either if you're spooling up jobs.
But it's not good at either in the context of an Azure data centre, I have many better options both price and performance wise for CPU, GPU and Hybrid Compute than a mid range desktop class CPU with a mid range desktop class GPU. I can't find any of this stuff on the Azure pricing calculator yet but that's likely merely a reflection of them not having gone live yet.
I don't know what the price point will be for just leveraging the XSX blades. They may not have enough reservation to do actual 'reservation'.
These units are dedicated towards Xcloud, so I suspect they can only be leveraged for job processing and not actual tenancy.
Given the type of customizations performed; I suspect they intend to make it do ML processing.
Last edited:
2 questions:
1. Do not Sony's much higher clocks (by much higher I do not mean in absolute terms, but in relative to power and yield) have a detrimental effect on the yield?
2. As to the SSD. I can only go by numbers available for retail (or posted rumors) but you don't think a DRAM-less, budget, 2 lane PCIE 3.0 over PCIE 4.0 connection is going to cost substantially less than a custom SSD which has a retail equivalent of a 4 lane, PCIE 4.0 7.0GBps drive? I don't see how the 175GB difference is going to make that up.
Yea it should relative to being lower on the clocks. What the yield amount is unknown. But you can imagine having say 1500 Mhz as a clock rate you'll get nearly 90%+ parametric yield for instance. Which is once again why we see Series S have such a low clock rate.
I think the main issue is that devs are only considering one SKU at launch, and so any product that comes later would need proper patching and development to run the day 1 games.
XSS doesn't have this problem because it's been there long enough that developers have to consider it for all games.
D
Deleted member 11852
Guest
Without stating the obvious (sorry!) this is a problem only if it's a problem ¯\_(ツ)_/¯
7nm+ is a mature process, nothing in PS5 is cutting design-edge and if there is a yield issue, they'll fix it one time and Sony will get more APUs per wafer. Fundamentally any yield issues aren't Sony's problem, this responsibility (unless Sony are insane) is split between AMD and TSMC so fundamentally, you'll back to units per wafer.
We've not seen PS5's board yet but I'm expecting Sony to see semiconductor cells hooked up to their custom controller, i.e not an embedded commercial drive package. E.g. what is in PS5 is probably tailored exactly for it, there are no extraneous functions and they are paying nobody to use it. Again, if Sony make more PS5s than Microsoft they'll get solid state storage cheaper and each console uses less so it's going further. Faster cells will cost more, but not that much more - the expense is typically the controller - this is the price premium you're paying for.
Johnny Awesome
Veteran
Yeah. It's too late for Sony to come to the lower-end SKU party. They ended up right in the middle, which is probably good enough given their brand power. Kinda reminds me of early PS3 days when it wasn't quite as powerful as the X360 until they figured out how to program Cell properly. In the end I'm sure Sony devs will find the shortcuts needed to close the performance gap between PS5 and XSX enough for gamers not to notice or care.
D
Deleted member 11852
Guest
Wasn't that shortly after Sony released PS4?
Microsoft has a data center and Surface hardware business that massively dwarfs anything Sony is doing, and they're going to be putting a very sizable number of XSX SoC based server blades into their Azure data centers to support xCloud.
Ok. So I have this correct. You are saying that the yields being effected by the high clock rate is a problem for TSMC and AMD? Meaning, I could clock my desired chip as high as I wish and any yield problems arising from this are paid for by TSMC and AMD? That does not compute to me. You are also stating this can be fixed? Only need to fix it once? I was under the impression that having a lower target clock increased usable chips per wafer. The reverse being true as well. That this was a process issue and not a design issue.
They've said that one XSX server blade can run and stream 4 Xbox One games simultaneously, so there has to be some virtualization hook; it is definitely not one tenant per blade. There's video encoding and ML hardware that can be leveraged, and some of MS's data center work is around being able to seamlessly offload specific compute tasks to different, more efficient for that task, hardware, so you wouldn't want to rent a XSX blade, but that XSX blade could be leveraged if you're doing ML work or video processing work even if most of your compute is happening somewhere else. Much of that is probably going to be internal MS stuff for a bit though.
D
Deleted member 11852
Guest
I'm 100% certain that 99.9% of the Azure server infrastructure is neither using GDDR6, nor solid state storage. It's using server grade ECC RAM and RAID-array HDDs.
