Post Xbox One Two Scorpio, what should Sony do next? *spawn* (oh, and Nintendo?)

slim may be out between 2.5 - 3 years later. So not as late as the 4years in the example.
so how much longer does it have to be before it would be considered ok?
 
The revisions to consoles still generally ran and displayed the games equivalently, and the times that they didn't (hard drive capacity problems, etc) did cause problems.

edit: To keep on topic: Would screenshots on the box, online store fronts, and game demo reels have standard/enhanced versions?

Would there be a replay of the Killzone SF resolution lawsuit?
 
Last edited:
slim may be out between 2.5 - 3 years later. So not as late as the 4years in the example.
so how much longer does it have to be before it would be considered ok?
It's about what the previous option was a few months earlier. Someone buying a console for £400 doesn't begrudge someone buying the same 4 years later £200, nor someone buying a £500 phone is upset over someone getting a better phone for £500 years later. If the console price drops £50 just after you buy a console, you'll be mildly miffed and shrug it off, ut emotions will run a lot higher if the price drops 1/3, you'll be more than miffed. As long as the delta is small enough, it'll be accepted. Presenting a price drop and a higher performance would be a significant delta, well beyond anything the console space has ever seen. Personally I think people will be more bothered by a change in spec for the better than price. That is, buy a £300 console now and it drops to £200 tomorrow, Joe Consumer will explete, "oh fi!". But buy a £300 console now and the model released tomorrow is 30% faster, the expletive will be more colourful. That's because the money isn't really worth the same. The 30% better epxerience is for the life of the platform, whereas that £100 difference is however many hours work to pay for. I think, that thought just popping into my head and sounding very reasonable! Yeah, that makes sense. With 100 hours of gaming, the better box provides 30% better gaming x 100 hours, while the £100 difference is all of a few hours difference between the two customers.
 
I see no real point in a Slim with a slight clock increase. What purpose should that serve outside marketing with a potential negative impact on the current base.

In general I'm not against intermediate updates. It's more a question of the right timing.

*BUT* if MS would go Sony's path with an evolutionary step late 2016 that would mean delaying a real update which afaik should be possible late 2017 with HBM2,Zen, new GPU. If MS times such announcement right they might be able to take away Sony/Nintendo's momentum early 2017 until they can release theirs. I doubt neither console will demonstrate right from the start why people "must" buy them.
 
Would screenshots on the box, online store fronts, and game demo reels have standard/enhanced versions?

Screenshots would be so downsampled on the back. I can't even remember the last time I actually looked at the back either.

Game demo reels already use PC version. :3

Would there be a replay of the Killzone SF resolution lawsuit?
Share recording only goes back so far. ;)
 
I think I worked out XBO needing to get to 1200Mhz on the GPU side of things to hit 1.8TF. It'll still be behind on ROPs but man will that Esram have bandwidth at that speed LOL. That's a pretty high over clock.

Even so, the devs would have to still deal with the 32MB space, although I suppose for games with dynamic res (Halo 5 and Doom), they'd just be hitting 1080p more often (although who knows, +50% rendering speed for Halo 5 would mean ~900p *cough*, and maybe with a sufficient upclock on CPU, they can tweak the bloody animation rate.
 
Microsoft should go to Epic and ask how big ESRAM has to be for them to effectively run UE4 at 1080p. Then when Epic says they want 64mb, Microsoft can take that information and say, "¯\_(ツ)_/¯ "
 
Was just looking at the PS4 die.

Assuming that the edge next to the CPUs, labelled as "shared memory controller", represents half of the system's 256-bit GDDR5 bus then "128-bits" of it is taking up roughly 26 mm^2 according to my scruff mspaint measurements. The sram on the X1 should take up roughly 40 mm^2 at 14/16 nm if claims about scaling are accurate. If it's true that the memory controller scaling is far less effective than other logic scaling (something about the "physical layer" of the soc), then some of the die cost of the sram in a future system could be balanced against the area saved in reducing memory controller size.

For example, an X2 with 32MB of sram might be able to drop from a 256-bit bus to a 128-bit bus, incurring a relatively small die area penalty overall, while using less power and having more aggregate BW that was more resistant to "worst case" access patterns.

I've thought for a long time that the 32MB of sram in X1 paired with a 128-bit GDDR5 bus would have been a far more potent design, though obviously that would have cost MS the ability to have 8GB of ram. For a 2017 refresh though, 8GB of GDDR5X at 176+ GB/s paired with a 200+GB/s sram pool (or potentially 400+ GB/s) could support Fury X levels of performance while having a small external bus well suited to a 10 nm shrink.
 
But the thing is that you have to pitch this design against a comparable system. Ms does not exist in a vacuum.

Is it bad the XBO memory subsystem with the inclusion of Esram?
I think not, it gives them the ability of using a large pool of cheap memory (DDR3) and a good bandwidth to fed its GPU, it´s better handling contention and concurrency, and uses less power.

But with the inclusion of Esram, they traded logic for memory, and they ended up with a soc larger than Orbis. Durango and its memory it was around 90% of the cost of Orbis and its GDDR5, while being around 60% of its performance. To me this is a colossal Fuck up.

If you go for a cheaper design, it better be really cheaper to manufacture. People at large didn´t care for better acoustic characteristics or being less power hungry. On top of that they bundled Kinect, and price it 100$ more than the PS4, no wonder that they are selling half than its rival.


What you describe its nice on paper, two pools with loads of bandwidth, but it could be solved with 8/16GB of HBM2 in 2017. Vega its coming at year´s end, and apus with HBM next year.
 
But with the inclusion of Esram, they traded logic for memory, and they ended up with a soc larger than Orbis. Durango and its memory it was around 90% of the cost of Orbis and its GDDR5, while being around 60% of its performance. To me this is a colossal Fuck up.

I agree it was a fuck up, but don't agree with all your reasoning here.

They didn't necessarily trade logic for memory. Just because esram took up die doesn't mean that the X1 would have used that die area for CUs if it wasn't there. MS still had heat, power and noise targets that additional CUs would have blown through, they still at a concept level would have under-emphasised GPU power, they still would have gone big on Kinect as dash control (but not for gaming), and they would still have been aiming for 8GB from early on and not as a last minute bump.

X1 power was a result of MS's misreading of ... everything ... and not because they used esram. Esram unfairly gets the blame for MS's woes because it's the easy thing to point a finger at in 2013/2016 etc and say "durr" but it's a symptom of MS's approach, and not the cause of it.

Durango is also rather more than 60% of Orbis. GPU is around 70%, CPU is 110%, and there is die area taken up by other MS customisations (including random chunks of esram not related to the 32MB pool of video memory). A MS equivalent to Durnge might still end up being larger.

What you describe its nice on paper, two pools with loads of bandwidth, but it could be solved with 8/16GB of HBM2 in 2017. Vega its coming at year´s end, and apus with HBM next year.

HBM2 is likely to be vastly more expensive than mainstream memory types, and that certainly will cut into your die budget. Some of the 2017 APU slides from a few months ago were even suggesting that they might use HBM1 (128GB/s).

128-bit GDDR5X mixed with 40~80 mm^2 of esram seems like it would have a good chance of being cheaper than equivalent BW HBM2, plus you'd get to chose where you package everything.
 
Last edited:
I agree it was a fuck up, but don't agree with all your reasoning here.

They didn't necessarily trade logic for memory. Just because esram took up die doesn't mean that the X1 would have used that die area for CUs if it wasn't there. MS still had heat, power and noise targets that additional CUs would have blown through, they still at a concept level would have under-emphasised GPU power, they still would have gone big on Kinect as dash control (but not for gaming), and they would still have been aiming for 8GB from early on and not as a last minute bump.

X1 power was a result of MS's misreading of ... everything ... and not because they used esram. Esram unfairly gets the blame for MS's woes because it's the easy thing to point a finger at in 2013/2016 etc and say "durr" but it's a symptom of MS's approach, and not the cause of it.

Durango is also rather more than 60% of Orbis. GPU is around 70%, CPU is 110%, and there is die area taken up by other MS customisations (including random chunks of esram not related to the 32MB pool of video memory). A MS equivalent to Durnge might still end up being larger.

Sorry, I wasn´t trying to downplay Esram at all, just pointing that its inclusion put them in a competitive disadvantage against Orbis, at least from a cost-performance perspective.

We know from the leaked roadmap to Xbox 720 that Ms were comfortable with a fast and small pool of ram, they saw it as an extension of the 360 philosophy, economics and manufacturability force them to use Esram.

We don´t know if they settled from the beginning at 12 Cus or if it was the size of the APU the limit, both designs are around 350mm2. Maybe a combination of both



HBM2 is likely to be vastly more expensive than mainstream memory types, and that certainly will cut into your die budget. Some of the 2017 APU slides from a few months ago were even suggesting that they might use HBM1 (128GB/s).

128-bit GDDR5X mixed with 40~80 mm^2 of esram seems like it would have a good chance of being cheaper than equivalent BW HBM2, plus you'd get to chose where you package everything.

Well, I´d like to know the economics of HBM2, samsung and Hynix are expected to ramp up volume production at Q3, yes I guess it´ll end up in the pricier cards at first, but they don´t even need 4 stacks at 8GB, with 16GB would be enough.

Also an HBM based design would mean a really small pcb and case, and would use less power, i think

Anyway, just dreaming
 
Imo if they redesign they should pass on esram, mapping Durango to better/newer hardware should be trivial compared to running 360 games on the XB1.

On the other end something like the Alienware Alpha sales 499$ (I think it was lower during the holidays, like 449$)(and you can do better thus bigger, uglier and more power hungry for less slightly money).I think MSFT should push/promote that type of products. The next generation of those mini PC should see neat improvements especially if AMD manage move things out on the CPU and allow for decent low power 4 cores in that type of set-up. The power available will be used as seen fit by the user whereas it is not clear how and when the PS4 neo will be used. When plugged to standard FullHD it may even act as a standard PS4.

Parity on the highest end console SKU won't do much for MSFT, Sony has more publisher supports, people will have more of their friends on PSN than Xbox Live, ans SOny will still have the upper hand in the basis model race. I mean it is not like the Neo is going to reset the market the market for all I see the market is set already which is imho why Sony dare to do what it is doing.

My favorite strategy for MSFT would be to push a cheaper, low power xbox, and to counter Sony move (higher end SKU) by promoting adequate windows devices in the appropriate segment. I mean Steam managed to get some manufacturers to move toward the good direction (leaving aside SteamOS fate), what MSFT could achieve could pretty if they push in the right way.

Then if they want another console, make it counts wait a little for better IP, clearer market trend and if VR start to be the next thing (which would surprise me but it is another matter) then plan around it.
 
We don´t know if they settled from the beginning at 12 Cus or if it was the size of the APU the limit, both designs are around 350mm2. Maybe a combination of both

Yeah, it could well have been. As you pointed out, cost of APU + memory isn't that much less for X1 than PS4 (according to teardowns anyway) and so that would fit with both trying to come it at around the same price ... minus Kinect.

Unfortunately, back in 2011 if you wanted 8GB of RAM as an absolute certainty for your concurrent multimedia plans then you needed to either bank on DDR3 or on a 512-bit GDDR5 bus. Sony were prepared to be gaming focused and go with 4GB, and that paid off because they got to have 8GB anyway and swagger around like The Fonz after getting with Mrs Cunningham.

Well, I´d like to know the economics of HBM2, samsung and Hynix are expected to ramp up volume production at Q3, yes I guess it´ll end up in the pricier cards at first, but they don´t even need 4 stacks at 8GB, with 16GB would be enough.

Also an HBM based design would mean a really small pcb and case, and would use less power, i think

I'd offer a guess that ramping is relative, and that ramp for HBM2 is still going to much lower in volume and much higher in cost than DDR4/GDDR5 and perhaps even HBM1 at first. HBM2 is the dream, but FSVs, multiple stacks of stacked memory stacked on a layer of logic, with everything stacked on an interposer just has that air of "Danger Will Robinson!" that accompanies every new much wished for shift in the status quo. An APU die with a package and a bog standard bus to GDDR5(X) just has that kind of depressing air of realistic expectation to it ...

Anyway, just dreaming

Me too!
 

". it will be extremely powerful from the start, i think 5-6 times PS4 Neo, then something in the order of over 10 TFLOPS or so. "
If that's true then I really hope the partnership between MS and Oculus will allow for Xbox Next compatibility with the headset!
Also MS could help with the Oculus supply as they have store presence and supply lines all over the world
 
". it will be extremely powerful from the start, i think 5-6 times PS4 Neo, then something in the order of over 10 TFLOPS or so. "
If that's true then I really hope the partnership between MS and Oculus will allow for Xbox Next compatibility with the headset!
Also MS could help with the Oculus supply as they have store presence and supply lines all over the world

Yeah. Well first of all there a lot of inconsistencies in these rumours. ie 5-6 times supposed PS4 neo gpu specs would be over 20Tflops...not 10. Not sure if he just meant 5-6 times PS4.
 
Or 5-6 times Xbox One puts it at PS4Neo levels.
 
Back
Top