Baseless Next Generation Rumors with no Technical Merits [pre E3 2019] *spawn*

[AlNom]

The general impact of SMT on games may be a ways different from that. And no benchmark of dual-threads is going to give any idea what 3 or 4 threads could achieve. Conceptually, what if Zen 3 has an extra SIMD unit per core for running maths tasks? May be possible to keep that populated and active as a third thread.

Somewhat tangential, but it'd be curious to see how AMD might implement more than 2 threads while IBM has been doing SMT4 since Power 7 and SMT8 with Power8, although those have oodles of on-die cache to help facilitate the many threads. IBM's guide on SMT4: https://www.ibm.com/support/knowledgecenter/en/linuxonibm/performance/tuneforsybase/smtsettings.htm

Maybe a jobs/tasks engine heavily multithreaded ala Cell's reqs. would be applicable in the context of games, if you get my meaning, although I guess it'd be somewhat more flexible with full cores here. e.g. Core 0 - Main thread + render thread + OS mini-thread, Core 1/2/3 - 12 more threads (SMT4) for workloads descended from the heavens to play with if a developer so chooses.

(I'm not bothering with the 3-way rumour, just a hypothetical context with all threads available)

 

[PSman1700]

Just a question, now with the next generation of consoles coming, and AMD's new Navi/Zen2 parts that offer greater value in the pc market. Are we going to see graphics like the movie Avatar (2009), or am i completely wrong there? Was just thinking, the baseline performance is going from 1.8TF to 10TF with better CPU and ram.

 

[Jay]

Would anyone want an APU in servers? Other than virtual consoles, at which point you'd only be running one game on a machine.

Penultimate last question you can see Phil pretty much 100% support what I expected to happen. Didn't realise (or remember) he'd actually said it. Maybe this shouldn't be in baseless rumor though as it's from the horses mouth and may be worth keeping in mind when discussing Scarlett?
https://www.microsoft.com/en-us/Investor/events/FY-2019/Barclays-Global-Technology-Media-Spencer

PHIL SPENCER: Yeah. If you go and you watch that video again, one of the things to take notice of is the silicon we're using to stream these games is actually the silicon from our console. And it turns out that consoles have very compatible kind of design criteria to what a blade in a server looks like. You want high power to energy use, low-price, cooling is an issue, all the things that these devices care about in the home work really well on a server blade.

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. So as a company like Microsoft, we can dual-purpose the silicon that we're putting in. We have a consumer user for that silicon and we'll have enterprise use for those blades as well.

It's all in our space around driving down cost to serve, and your cost to serve is made up by two things, how much was the hardware and how much time was that hardware monetized? So we can monetize that hardware over more cycles in the day, in the 24 hours, through game streaming and other things that needs CPU and GPU in the cloud, and we will drive down the cost to serve in our service.

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.

 

[Globalisateur]

XB1 was designed for games and TV.
Scarlet is being designed for games and Azure servers.
Why can't they design a console just and only for games ?

So the design as we move forward is done hand-in-hand with the Azure silicon team

Won't the design be flawed because of that ? A bit like XB1 was because of the TV TV TV ?

XB360 was designed only as a gaming machine, I think.

 

[AlNom]

He explains why. You’ll have to explain your assumptions on the implications of why you think the things they are considering are not desirable or not necessarily compatible. I have some questions because I'm genuinely curious as you seem to have waved aside all that was said in the interview without really explaining yourself clearly.

What is a server? What is important to servers? Is power consumption not a concern for a console?

What does machine learning require? What does that mean for the GPU? Is double rate FP16 the worst thing for a gaming console?

How much RAM should they have? How much bandwidth do you need for gaming, how much would you need in a server environment? Should the current gen have only gone with 4GB RAM?

Do you want them to throw away App switching & Resume or other OS functionality because those aren't needed in a traditional gaming console? e.g. What is needed for a universal streaming/sharing app?

What are the cost implications?

What is a Nintendo console?

What is it that you believe them to be building spec-wise based on the information given

edits9000:

Given how thermally & yield challenged the PS360 designs were for the longest time, are you proposing that console companies make the same design pushes without being conscientious of price & inflation (nearly 15 years later) while considering a significant enough performance delta?

In today's environment, we can also see how fab node advancements aren't what they used to be, and that has implications.

We might throw in the precariousness of Microsoft's position in the market where current Xbox sales have stymied around the 40M mark where the competition can enjoy at least 2x the volume production. That will have implications for making volume contracts.

 

[iroboto]

XB1 was designed for games and TV.
Scarlet is being designed for games and Azure servers.
Why can't they design a console just and only for games ?


Won't the design be flawed because of that ? A bit like XB1 was because of the TV TV TV ?

XB360 was designed only as a gaming machine, I think.

The last paragraph of the interview has the answer you seek

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.

a real competitive advantage could mean a lot of things in PR land, but usually it means having leverage in an area over the competition because the competition won’t be doing this.
It should mean either their streaming service is cheaper or better. Or the console hardware will be cheaper or better as a result of the dual purposing helping to subsidize the hardware.

In the end the work with azure team is still about bringing a better gaming experience forward. So it’s not like TV. As Phil notes the console chips are well suited for server life. So if that part is already covered, then the customizations might be elsewhere for azure.

 

[Jay]

Won't the design be flawed because of that ? A bit like XB1 was because of the TV TV TV ?

I'm not sure I agree with that being what flawed the design as you put it.
Even if they wasn't aiming for it to also be a media device, do you think 4GB of GDDR5 total would have been a better option for a games only console?
From the initial design phase that could've been the option as it was for the PS4.
At the time the only reasonable way to be sure of 8GB may have been their approach.

There's many ways that working with Azure team could benefit a console, may include something that you otherwise wouldn't, but as its going to be used in other workloads also, may offset cost. AI acceleration for example.

As long as primary focus is games which it will also be in the cloud I don't see the problem. Goals don't seem to be opposed.

Edit:
Or using the basis where all this started
Zen 3, 3 thread may not make financial sense compared to zen 2 for only a console.
But for console and cloud use, it may make sense and therefore the console benefits by getting it.

 

[Shifty Geezer]

Won't the design be flawed because of that ? A bit like XB1 was because of the TV TV TV ?

XB360 was designed only as a gaming machine, I think.

It depends on what the synergies are between the different device applications. If what's needed for servers is also what's needed for gaming, you get benefits of scale and whatnot. May even get more 'lossy' hardware on the console side because more is recovered from the hardware design through the cloud business. Perhaps XBN will have extra special networking robustness??

It should mean either their streaming service is cheaper or better.

I think Globby is looking simply at the console experience rather than the platform. If you don't care to use the cloud service, whatever benefits Azure team involvement brings to improve that is on no benefit for you. Though if doesn't compromise anything, it's not a problem. More as AlBran asks, lots of questions really about what a console-for-games design would need to be and how that contrasts with a console-for-games-and-servers needs to be. Without a solid argument, the fears seem unjustified based on an assumption that spreading a design too thin across different targets will always be a negative.

 

[eastmen]

XB1 was designed for games and TV.
Scarlet is being designed for games and Azure servers.
Why can't they design a console just and only for games ?


Won't the design be flawed because of that ? A bit like XB1 was because of the TV TV TV ?

XB360 was designed only as a gaming machine, I think.

What flaw did the xbox one have because of tv ? The additional $3-5 bucks for a second hdmi ? It was an 8 gig console with a small pool of fast eram . When the ps4 was in development it only had 4 gigs of ram and luckly right before announcement they were able to add in another 4 gigs. The ps4 could have been seriously flawed from the jump if they didn't get lucky

The benefit of Azure servers is two fold. 1 ) They are going to look at making efficient designs since they will want to be able to scale these as blade servers so heat / power usage is important and 2) they are going to be building hundreds of thousands of blade servers based on the next xbox hardware for cloud streaming. So Ms will have a pricing advantage there by buying a lot more of every component . MS can also justify the cost because zen processors with lots of ram and a strong gpu can be used by other clients for cloud rendering or even just word online when they aren't being used for gaming.

 

[iroboto]

I think Globby is looking simply at the console experience rather than the platform. If you don't care to use the cloud service, whatever benefits Azure team involvement brings to improve that is on no benefit for you. Though if doesn't compromise anything, it's not a problem. More as AlBran asks, lots of questions really about what a console-for-games design would need to be and how that contrasts with a console-for-games-and-servers needs to be. Without a solid argument, the fears seem unjustified based on an assumption that spreading a design too thin across different targets will always be a negative.

Yea, that makes sense. I guess I was looking at it like shopping at a grocery store and another at like Costco/Sams/Price Club. Looking at ways to exploit at single chip design used for a multiple purposes allowing for maximization of Cost/performance as a result.

But that’s just one perspective it could be of many. Lately i am thinking the customizations being around making multi chip performance being better. So if we’re seeing loads spread across multiple chips, how we could improve that. I don’t think that had a lot of feasibility in the console space, but that type of thing shouldn’t impact performance locally

 

[MrFox]

The Cell was presented like that, and the PR was very similar. Designed to be as good for gaming as for supercomputers.

 

[iroboto]

The Cell was presented like that, and the PR was very similar. Designed to be as good for gaming as for supercomputers.

Yea I get the concern but we aren’t talking exotic architecture here. It’s x86 to x86. We are talking about hardware that will likely be 99% the same as what will go into the PS5. The server side stuff primary role is streaming Xbox games. The secondary role while idle is for enterprise compute on azure to maximize idle time for profit.

They should be prioritizing customization in that way; and our considerations for outlier performance penalties should be in consideration of these realities. We also know that the methodology in which they can profile performance on their chip customizations prior to burning to silicon should be able to identify these setbacks. They aren’t going in blind.

The probability that MS will not make a competent high performance console for its price point is very low.

Whether it is the king of performance is a separate topic entirely. Expecting that MS and Sony will launch next gen like they did this gen, with Sony at better price points and with 30% more performance is wishful thinking ignoring how and what they accomplished with X1X.

 

[function]

If you're going to be using the same silicone in "none entertainment" workloads then you probably need to be looking at having turbo options and power balancing modes. And that will hopefully make its way in some form into consoles too.

Too much power being left on the shelf with current fixed clock speeds.

Doesn't worry me that they're looking to use the silicone outside of gaming. Zen is as much a server chip as a consumer one, and Vega was also designed as an enterprise GPU/accelerator. I can see a fast APU as offering high performance and high flexibility.

 

[BRiT]

The Cell was presented like that, and the PR was very similar. Designed to be as good for gaming as for supercomputers.

To be fair, it was better for gaming than it was supercomputers.

 

[Shifty Geezer]

Playing Devil's advocate...

1 ) They are going to look at making efficient designs since they will want to be able to scale these as blade servers so heat / power usage is important...

Power consumption isn't so important for consoles. There's a potential worry there that console power will be slightly hampered to enable a more power efficient, lower overall performance server unit.

2) they are going to be building hundreds of thousands of blade servers based on the next xbox hardware for cloud streaming. So Ms will have a pricing advantage there by buying a lot more of every component...

Actually, hundreds of thousands ain't that many. I had thought bulk-buying would be a bonus, but a more successful console sells tens of millions. A few hundred thousand more isn't going to tip the buying balance much.

Another concern/curiosity I have is long-term value of this hardware for servers. Where the design may be valuable for launch, three years on, won't servers be wanting something better? Maybe that's a good thing, causing a more considered upgrade for the console hardware as the cost to process-shrink pay more dividends in running servers than in selling slightly cheaper, cooler consoles.

 

[BRiT]

Another concern/curiosity I have is long-term value of this hardware for servers. Where the design may be valuable for launch, three years on, won't servers be wanting something better?

That fits right in with the mid-gen console update cycle. Win. Win. Win.

 

[iroboto]

Actually, hundreds of thousands ain't that many. I had thought bulk-buying would be a bonus, but a more successful console sells tens of millions. A few hundred thousand more isn't going to tip the buying balance much.

Total addressable market for streaming is 2B+ IIRC. 1B for gaming? I dunno they said the tam during the announcement of xcloud. They will have much more than hundreds of thousands.

 

[DJ12]

While I must say it will be interesting if zen 3 was in the next Xbox, but as zen 2 isn't even out yet, and zen 3 is on a slightly newer process 7nm+ it's unlikely that a console launched in 2020 will have it.

Plus o believe this rumour started as someone added it to Wikipedia lol.

Amd seem totally in bed with Sony though so I call bs on this one.

 

[iroboto]

Amd seem totally in bed with Sony though so I call bs on this one.

That’s not how the semi custom department at AMD works.

And MS spends $1B per month on data centre builds and have chosen EYPC to provide their CPU. MS has a lot of money with AMD and should have very good negotiation power. I’m not sure I understand the argument on why Sony would get more favourable terms over MS.

 

[Jay]

Amd seem totally in bed with Sony though so I call bs on this one.

Everything was so reasonable, but then this made me sigh..
Even if Sony was working close with AMD, I don't see how that affects MS.