Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]

[function]

If MS are going with a high end scropoio they're going to need an entry level system too.

X1 is prehaps better suited for a scrap at the low end than PS4. I wouldn't be surprised if a 14nm X1 SoC was actually smaller than a PS4 14nm SoC due to the esram scaling perfectly while the huge GDDR5 interface doesn't. Should be more defect tolerant too. They can also use DDR3L to save power with little to no additional engineering while PS4 will remain on vanilla GDDR5, and DDR3 prices are remaining in the gutter (note to self: buy another 16GB for PC!). Throw in a cheaper heasink and PSU, ditch the hybrid drive, ditch the HDMI in and associated silicon, possibly kill all the Kinect specific hardware and silicon area, and the X1 should lean up pretty well.

Just focus like a laser on the gamer who wants an affordable system, and present it as a 360 BC stepping stone into UHD Scropium VR.

XScro should be presented as a next level system, not a Bone.5.

 

[function]

I wonder...The rumored released date is for end of 2017, is it possible that Scorpio would be a direct reaction to the previous Neo leak?

30 march 2016 to December 2017 would be ~21 months. Is it possible to plan the release of a whole new console in such a short time? Particularly if you use an AMD supercharged known architecture...

Could be, or it might be that MS wanted their "big numbers" and decided that a Zen APU would need to be the foundation of that. That would limit them to the second half of 2017.

Interestingly, a 6TF system would seem to be in that interesting ground where either GDDR5X or 2X HBM2 stacks could fit. Do we know if 16 Gb GDDR5X chips are going to be knowing around by the second half of next year?

 

[mpg1]

I wonder...The rumored released date is for end of 2017, is it possible that Scorpio would be a direct reaction to the previous Neo leak?

30 march 2016 to December 2017 would be ~21 months. Is it possible to plan the release of a whole new console in such a short time? Particularly if you use an AMD supercharged known architecture...

Phil Spencer already talked about new xbox hardware before the neo leaks so they were definitely already in the stages of developing a new console. It's just a question of how much are the neo leaks influencing their decisions now. Especially in terms of things like specs/announce dates.

The 5-6Tflop rumor is interesting because it sounds exactly like the numbers you would target if you wanted to one-up the PS4 neo (assuming those rumors are accurate).

 

[mpg1]

Could be, or it might be that MS wanted their "big numbers" and decided that a Zen APU would need to be the foundation of that. That would limit them to the second half of 2017.

Interestingly, a 6TF system would seem to be in that interesting ground where either GDDR5X or 2X HBM2 stacks could fit. Do we know if 16 Gb GDDR5X chips are going to be knowing around by the second half of next year?

I still think it's possible they could go with 12GB over 16GB.

 

[mpg1]

Developer: We're working on an Xbox One VR game for 2017

http://arstechnica.com/gaming/2016/05/developer-were-working-on-an-xbox-one-vr-game-for-2017/

 

[function]

I still think it's possible they could go with 12GB over 16GB.

I wonder if we'll see split memory.

Much of the memory in PS4Bone is taken up with either relatively low BW dash activity or basically used as a HDD cache. 3GB on dashboard, 1~3 GB caching data you don't need yet, but might do soon because you're streaming off a frikkin 5400 rpm laptop drive with ~70 MB/s sequential read.

4GB of DDR4 for the dash and another 4GB DDR4 for the cache if we're going to stuck with mechanical 2.5" drives might be a good move. No point wasting HBM2 on that stuff...

 

[3dilettante]

I wonder...The rumored released date is for end of 2017, is it possible that Scorpio would be a direct reaction to the previous Neo leak?

30 march 2016 to December 2017 would be ~21 months. Is it possible to plan the release of a whole new console in such a short time? Particularly if you use an AMD supercharged known architecture...

It's 1-2 financial quarters to get chips back from manufacturing, depending on what stepping is good enough. Then months to ship components, manufacture, and distribute for launch.
That might be a year off the time frame that can't go into the chip design.
The APUs that we do know took longer than a year to get nailed down.

There's a lot of discussion about X design being in reaction to a leak or press article today, when architectural decisions and system designs are rolls of the dice years in the making. If any given design is a "reaction" it would probably be deciding between projects that had something going on already.

AMD probably wouldn't mind a little of that. It's got HPC APU slides with big GPUs and new cores that could be used as a basis, and a sequence of GPU architectures it might want to use.
If a war of designs results, it gets paid either way--with the caveat that it can only put so much work into false starts before its other projects become constrained.

Conspiratorially, AMD would benefit from a cascade of leaks that freaks out its console customers into paying into a succession of hand-over-hand design leapfrogs.

 

[Grall]

yup and I don't think that is a coincidence either..

And then you imagine Sony bumping clock speed to say, 1.3-1.5GHz range, and watch what happens to neo's Tflop rating... Sub-1GHz is really low for 14nm, assuming neo actually features a shrunken SoC.

 

[3dilettante]

And then you imagine Sony bumping clock speed to say, 1.3-1.5GHz range, and watch what happens to neo's Tflop rating... Sub-1GHz is really low for 14nm, assuming neo actually features a shrunken SoC.

Considering sub-GHz speeds slow does bake in some assumptions like the GPU is Polaris and has been architecturally revamped to take advantage of the node, and that Polaris is really intended to reach for those clocks.

One side effect of being near the same clock as 28nm is that FinFET can drastically cut power consumption. If Neo is in part the result of an inability to really shrink Orbis without leaving dead silicon, then it might be a form of PS4-slim. Even with a doubled GPU, the APU might be decently cooler-running if it doesn't push clocks and the shortcomings of the 28nm implementation for Orbis are addressed.

If it turns out Sony has to respond with more performance, having clock headroom that significant can allow for a Neo adjustment to get closer to a new Microsoft console as long as the chip and possibly 1-2 board revisions are validated for full-range. It would require a cooler/case change rather than spinning up a design.

Whether the rumored bandwidth increase is sufficient for whatever Neo might be a question mark.

 

[Jay]

I bet if you asked the average person about a PS4.5 or PS4K or even just a new PS4 you'd be surprised how many people would not have heard about it. And that's after several months of popular internet rumors.

Formal announcements bring mainstream press which is an entire different level of exposure.

yes but ms only needs to formally announce it around the time neo is due out.
then they say our machine is 6months away.
general public doesn't know about the neo either, ms could piggy back off their launch.
would only work if ms has a compelling machine though.

 

[Grall]

Even with a doubled GPU, the APU might be decently cooler-running if it doesn't push clocks and the shortcomings of the 28nm implementation for Orbis are addressed.

I hear ya, but, like, a massively powerful GPU like GP102 runs at 1.8GHzish and draws ~165W with boutique RAM hooked up to it, less than launch PS4 did. A scaled-down chip for neo should therefore reasonably still be able to clock well and only have a fairly modest power draw. Or so I'd like to think anyhow.

 

[mpg1]

yes but ms only needs to formally announce it around the time neo is due out.
then they say our machine is 6months away.
general public doesn't know about the neo either, ms could piggy back off their launch.
would only work if ms has a compelling machine though.

They could announce Scorpio before Neo is out but Microsoft may want go one step further and announce it even before Sony announces neo. They may want to give pause to gamers who may pre-order Sony's new console. If neo is coming out this fall it means Sony could announce it anytime including as soon as E3 which means Microsoft would have to announce it there or just before.

 

[3dilettante]

I hear ya, but, like, a massively powerful GPU like GP102 runs at 1.8GHzish and draws ~165W with boutique RAM hooked up to it, less than launch PS4 did. A scaled-down chip for neo should therefore reasonably still be able to clock well and only have a fairly modest power draw. Or so I'd like to think anyhow.

The GP104 does not draw from the same architectural line as what Neo would. GCN so far hasn't had that clock target, and some unsourced rumors indicate AMD did not expect GP104 to clock as high as it did. Some articles on GP104 indicate Nvidia didn't quite expect how highly it clocked, either.

If Neo is situated as a sort of PS4 slim-with-benefits, GP104's operating power could be over 2x too high. Just drawing what the PS4 draws now wouldn't get a slim model.

 

[AlNom]

I feel like we should temper expectations on clocks as well. nVidia just has the higher R&D budget.

edit:

And... I suppose there's Sammy/GF vs TSMC.

The two different A9's may or may not be a hint. The TSMC chip seemingly performed better than the one produced by Sammy, whereas the latter was a touch smaller.

 

[Nisaaru]

I wonder if we'll see split memory.

Much of the memory in PS4Bone is taken up with either relatively low BW dash activity or basically used as a HDD cache. 3GB on dashboard, 1~3 GB caching data you don't need yet, but might do soon because you're streaming off a frikkin 5400 rpm laptop drive with ~70 MB/s sequential read.

4GB of DDR4 for the dash and another 4GB DDR4 for the cache if we're going to stuck with mechanical 2.5" drives might be a good move. No point wasting HBM2 on that stuff...

Would it really make sense to use 8GB HBM2 and then 8GB DDR4 in a console? If you're already using a HBM2 2.5D interposer setup it easier to avoid the 2nd memory bus completely. Less complex design, less heat, smaller PCB and less parts which might be even cheaper in the end.

A 2-bus setup could be usable for CPU latency reasons but would that really be a relevant decision factor for MS?

 

[Silent_Buddha]

I wonder...The rumored released date is for end of 2017, is it possible that Scorpio would be a direct reaction to the previous Neo leak?

30 march 2016 to December 2017 would be ~21 months. Is it possible to plan the release of a whole new console in such a short time? Particularly if you use an AMD supercharged known architecture...

Highly unlikely due to a number of factors, including that Phil Spencer has hinted at it before there was even a sniff of a rumor about PS4 Neo.

However, the important thing is, if they are partnering with Oculus for VR then the XBO doesn't meet the minimum specifications needed. A new Xbox would be needed regardless.

Developer: We're working on an Xbox One VR game for 2017

http://arstechnica.com/gaming/2016/05/developer-were-working-on-an-xbox-one-vr-game-for-2017/

And that gives credence to the rumor that Microsoft is partnering with Oculus for VR. Either that or the developer is making a VR version for the Xbox platform purely based on rumors... Yeah, no developer is going to survive if they develop for a platform based on rumors and not information provided to them by a hardware manufacturer.

If anything Sony did would have prompted this type of response it likely would have been the PSVR. Since if Microsoft wanted to do VR and looked at Oculus as a VR partner, they were going to need a more powerful console in order to get proper support for it.

Regards,
SB

 

[MrFox]

Highly unlikely due to a number of factors, including that Phil Spencer has hinted at it before there was even a sniff of a rumor about PS4 Neo.

October 2015: Sony CTO interview.
http://www.dualshockers.com/2015/10...e-release-of-a-ps4-with-improved-performance/

With the PS3 architecture and the Cell, it was impossible to expand the machine beyond the capacity of the hard disk. On the other hand, with the PS4 adopting a conventional X86 architecture, it’s easy to achieve flexible performance enhancements while using the same game assets.

That’s why providing a standard performance version of the PS4 and a high performance version of the console side-by-side is, according to Ito-san, an idea that might be considered.

 

[Globalisateur]

All those leaks coming from Polygon and Asstechnica? What a coincidence!

 

[Grall]

Just drawing what the PS4 draws now wouldn't get a slim model.

I don't think Neo would be both faster AND slim. And TBH, current PS4 is fairly svelte as it is. If Sony would just straighten out the ridiculously slanted front and back sides, it would be way smaller still.

Pizza box PS4 Neo! Who's with me?!

 

[rockaman]

All those leaks coming from Polygon and Asstechnica? What a coincidence!

Is Ars bad for this stuff? Serious question. I don't really go there.