Next Generation Hardware Speculation with a Technical Spin [2018]

[anexanhume]

So this guy claims to have insider knowledge about MS's strategy. Basically a discless streaming box and a traditional physical box, both using Zen 2 and AMD's next gen "upcoming gen" gpu and releasing at the same time as Sony at 2020.

The discless box is a Thurrott rumor and the other stuff is a rehash of digital foundry and AdoredTV rumor/speculation.

Edit: didn’t realize this was Brad Sams. Makes sense they’d repeat their own rumor

 

[Xbat]

Well looking at those Navi performance rumours it looks like 1080ti performance will be the upper limit for next gen?

 

[Love_In_Rio]

Well looking at those Navi performance rumours it looks like 1080ti performance will be the upper limit for next gen?

Is the optimal objective.1080 ti is what? 250 watts at 16nm?.Suppossing it would be half that at 7nm we would be in the 125 watts at 7nm, the console consumption region for the GPU.But if AMD gets same performance with that wattage they will have given a big jump in efficiency. If the architecture is still GCN we will only get Vega 56 performance.

 

[vipa899]

1080 performance, if were lucky. Cpu somethig midrange by desktop standards, 16 to 24gb ram total?

 

[anexanhume]

Phil dishing the goods here:

The thing that’s interesting for us as we roll forward, is we’re actually designing our next-gen silicon in such a way that it works great for playing games in the cloud, and also works very well for machine learning and other non-entertainment workloads. As a company like Microsoft, we can dual-purpose the silicon that we’re putting in.

We have a consumer use for that silicon, and we have enterprise use for those blades as well. It all in our space around driving down the cost to serve. Your cost to serve is made up by two things, how much was the hardware, and how much time does that hardware monetize.

So if we can monetize that hardware over more cycles in the 24 hours through game streaming and other things that need CPU and GPU in the cloud, we will drive down the cost to serve in our services. So the design as we move forward is done hand-in-hand with the Azure silicon team, and I think that creates a real competitive advantage.”

Spencer concluded by talking about the issue of latency while streaming games. He mentioned that he’s already streaming games while he’s on the road around the world, and the test servers for Project xCloud are still in Washington, but Azure has a global scale.

 

[Xbat]

1080 performance, if were lucky. Cpu somethig midrange

It's what I've been expecting, earlier in the thread where I said expecting 1080ti performance was being very hopeful in next gen consoles people seemed to be disappointed. For me the CPU is the massive step up from current gen.

It's also why I believe temporal upscaling is going to be a big thing next gen. There's enough power to wow just not enough to wow at native 4k in my opinion.

 

[Shifty Geezer]

Spencer concluded by talking about the issue of latency while streaming games. He mentioned that he’s already streaming games while he’s on the road around the world, and the test servers for Project xCloud are still in Washington, but Azure has a global scale.

That's not a good answer. It's open to the imagination, with people thinking he's playing action games seamlessly. What games is he playing, and what's the latency like? We know there are physical limits regards transfer times, and we know that part of the way to reduce latency is to move servers nearer users rather than route across thousands of miles.

We know streaming games works because others have already been doing it for years, and we know it's not ideal but an okay experience, particularly for some genres. Spencer's comment doesn't build on that any - are they lower latency? Or is the experience similarly a compromise?

 

[iroboto]

are they lower latency? Or is the experience similarly a compromise?

This will largely be about how close MS can get the hardware to edge for Real Timr applications.

 

[Jay]

That's not a good answer. It's open to the imagination, with people thinking he's playing action games seamlessly. What games is he playing, and what's the latency like? We know there are physical limits regards transfer times, and we know that part of the way to reduce latency is to move servers nearer users rather than route across thousands of miles.

We know streaming games works because others have already been doing it for years, and we know it's not ideal but an okay experience, particularly for some genres. Spencer's comment doesn't build on that any - are they lower latency? Or is the experience similarly a compromise?

Everything your saying may be true, but I doubt that was the time and place for a better answer.
All they've done is announce that their going into that market, and now that a portion is up and running.

May not hear anything actually worth while from them until they start public tests.

 

[ultragpu]

Well looking at those Navi performance rumours it looks like 1080ti performance will be the upper limit for next gen?

Well, a 1080 ti is 11.3 tf so for an equal AMD gpu it would need to add 3.6 tf which makes it almost 15 tf if Vega 64 a 12.6 tf gpu is on par with a 1080 at 9 tf. Yeah that's definitely high balling. A 1080 performance is more likely after all at ~12 tf. Would be lucky to hit 14 tf if they're willing to invest a quality cooler.

 

[mrcorbo]

The discless box is a Thurrott rumor and the other stuff is a rehash of digital foundry and AdoredTV rumor/speculation.

Edit: didn’t realize this was Brad Sams. Makes sense they’d repeat their own rumor

The rumor of the discless Xbox for this gen is fairly recent, but the rumor about the streaming-only box for next -gen goes back a ways.

 

[mrcorbo]

That's not a good answer. It's open to the imagination, with people thinking he's playing action games seamlessly. What games is he playing, and what's the latency like? We know there are physical limits regards transfer times, and we know that part of the way to reduce latency is to move servers nearer users rather than route across thousands of miles.

We know streaming games works because others have already been doing it for years, and we know it's not ideal but an okay experience, particularly for some genres. Spencer's comment doesn't build on that any - are they lower latency? Or is the experience similarly a compromise?

I wonder if any of this can be mitigated by using more processing power on the cloud server side. There's precedent in the netcode for fighting games, which are among the most latency-sensitive games (if not the absolute most latency-sensitive games). To what degree can you, with machine learning, develop algorithms to compensate for latency and create the feel of a more responsive experience?

 

[cheapchips]

Well, a 1080 ti is 11.3 tf so for an equal AMD gpu it would need to add 3.6 tf which makes it almost 15 tf if Vega 64 a 12.6 tf gpu is on par with a 1080 at 9 tf. Yeah that's definitely high balling. A 1080 performance is more likely after all at ~12 tf. Would be lucky to hit 14 tf if they're willing to invest a quality cooler.

Don't we get to take off some AMD flop penalty for better console flop usage.

 

[Tkumpathenurpahl]

That's not a good answer. It's open to the imagination, with people thinking he's playing action games seamlessly. What games is he playing, and what's the latency like? We know there are physical limits regards transfer times, and we know that part of the way to reduce latency is to move servers nearer users rather than route across thousands of miles.

We know streaming games works because others have already been doing it for years, and we know it's not ideal but an okay experience, particularly for some genres. Spencer's comment doesn't build on that any - are they lower latency? Or is the experience similarly a compromise?

Streaming's come on leaps and bounds over the last few years. I don't know what PSNow's like, and xCloud is still up in the air ( ) but I can happily play Spider-Man via Remote Play, at my friend's house, which is roughly a 20 minute drive.

This, on my smartphone, from my WiFi connected PS4Pro. Having the consoles located in server farms will be a further improvement. If they can beef up the server hardware to render at 60 or 120 fps, so dropped frames matter less, that'll further improve it. If the local streaming consoles are capable of frame interpolation and upscaling, that'll further improve it too.

I used to be pretty skeptical about game streaming, and I'll always want to play the likes of Spider-Man on local hardware, but the tech's already there to make my sub-optimal situation viable in the suburbs of a fairly minor British city.

Little boxes with a mostly stable, barely laggy 30fps will be good enough for most people.

Well looking at those Navi performance rumours it looks like 1080ti performance will be the upper limit for next gen?

Does it necessarily? Even if those rumours are true, we only really have confirmation about cards which seem designed to appeal to a broad range of the market, whilst conceding they have nothing to go toe to toe with the RTX2080 in 2019.

 

[mrcorbo]

Phil dishing the goods here:

The pull quote about the ability to leverage the silicon being developed for Project xCloud and, by extension, Project Scarlet for non-gaming workloads is interesting. It makes we wonder in what ways that might affect the design and what goes into it. Are we just talking standard x86 CPU and GPU processing units or will there be more custom hardware that maps well to machine learning?

 

[iroboto]

I wonder if any of this can be mitigated by using more processing power on the cloud server side. There's precedent in the netcode for fighting games, which are among the most latency-sensitive games (if not the absolute most latency-sensitive games). To what degree can you, with machine learning, develop algorithms to compensate for latency and create the feel of a more responsive experience?

We have some real information to work with here.

Thoughts on the cost savings if MS is trying to make an all in 1 chip for servers and consoles? Yields /binning must be done creatively to be good for cost efficiency.
What are we looking at as well, for customization for azure? I'm not even sure what that means. A chip setup that is good for passive cooling? but can be retrofitted for active cooling as well?

 

[vipa899]

Don't we get to take off some AMD flop penalty for better console flop usage.

Its probally not that easy, they would get more performance out of 10TF turing then 10TF out of amd Vega/navi in a console.

whilst conceding they have nothing to go toe to toe with the RTX2080 in 2019.

2019/2020 will probally be close to the rtx 2000's successor. Amd competing with 2018 close to mid/high end range hw in 2019/2020 isnt all that stunning. We need more competition from amd and maybe intel.

 

[AlNom]

Thoughts on the cost savings if MS is trying to make an all in 1 chip for servers and consoles? Yields /binning must be done creatively to be good for cost efficiency.

The only thing we can glean for certain is support for packed math. Otherwise....¯\_(ツ)_/¯ Just Monika

Machine learning for drivatars (Forza) /shadows (Killer Instinct) hm..................

What are we looking at as well, for customization for azure? I'm not even sure what that means. A chip setup that is good for passive cooling? but can be retrofitted for active cooling as well?

Server racks are cooled a lot differently, no? Like a field of fans blowing across the racks?

 

[iroboto]

Server racks are cooled a lot differently, no? Like a field of fans blowing across the racks?

Yea passive cooling ? Did I use that term incorrectly?
It doesn't have it's own cooler and relies on the rack

 

[AlNom]

¯\_(ツ)_/¯