Xbox Series X [XBSX] [Release November 10 2020]

What if you only do so while you build up your collection of worthy SOCs, so you have the temporary SOCs while scaling up? You know, like how you probably have that one pair of socks with a hole or two in them that you just cant get rid of because you might need them on laundry day or if your good socks get wet somehow.
 
I understand this and it doesn't change anything. Server infrastructure is built to maximise efficiency which is why you don't save a few quid by using a chip with some broken cores. It's cheaper to toss it and put it a fully working processor, now you can support 1 or more extra users. The I'll-save-that-it-might-come-in-handy mentality is not compatible with how servers operate. :nope:

Right.

So MS stopped deployment of Xbox one servers that required a costly external video encoder. They are now transitioning to a platform where they can run 4 instances of xbox one on a single blade and no longer need the external video encoder.

They will be using chips that may require slightly more voltage or can take advantage of the better cooling afforded. They aren't going to go scraping the bottom of the barrel for defective parts but just like every server chip ever created there will be different tolerances for what they will accept. From other companies they would be diffrent bins.

MS will be rolling out XSX stuff for around 3 years until they start replacing it with newer hardware.

We also aren't talking about saving a few quid. We are talking about being able to have a timely roll out for their new cloud platform using chips they can only source from one supplier
 
We also aren't talking about saving a few quid. We are talking about being able to have a timely roll out for their new cloud platform using chips they can only source from one supplier

Sure, but the only way Microsoft are deploying chips with fails cores in servers is if they have yield/capacity problems because a borked chip is better than no chip.
 
I understand this and it doesn't change anything. Server infrastructure is built to maximise efficiency which is why you don't save a few quid by using a chip with some broken cores. It's cheaper to toss it and put it a fully working processor, now you can support 1 or more extra users. The I'll-save-that-it-might-come-in-handy mentality is not compatible with how servers operate. :nope:
What if they have chips with disabled units but overclock the processor back up to spec (or higher)? Ultimately the primary purpose would be to serve cloud streaming of Scarlett games, so it needs to be at least as capable as retail in theory, no?

e.g. 48CUs @ 2GHz = 12.3TF

I do assume they're cherry picking fully enabled units for xcloud though so that performance is as perfect as can be for streaming.

The current XO xcloud units actually have overclocked CPU cores, FWIW.
 
What if they have chips with disabled units but overclock the processor back up to spec (or higher)? Ultimately the primary purpose would be to serve cloud streaming of Scarlett games, so it needs to be at least as capable as retail in theory, no?

e.g. 48CUs @ 2GHz = 12.3TF

This is similar to the scenario that Mark Cerny used in his The Road to PS5 talk to explain why two theoretical identically setups in terms of teraflops - one with more CUs at lower clocks, one with less CUs at higher clocks - would produce different performance. So.... ¯\_(ツ)_/¯
 
Last edited by a moderator:
Sure, but the only way Microsoft are deploying chips with fails cores in servers is if they have yield/capacity problems because a borked chip is better than no chip.

Alternatively you have a goal of X chips that you want to have for your server farm. Using current production of fully function chips it'll take you Y years to fully populate your server farm. And you'll be able to grow that farm at Z usage capacity (service requests served) per month/year based on production of fully functioning cores.

Alternatively while X and Y will still be same in the following scenario, you can use defective chips to increase your Z usage capacity. It isn't optimal and it'll still take the same amount of time to fully populate a server farm, but you'll be able to service more requests while you work to fully populate that server farm.

Of course that comes at the cost of higher energy cost per service request, but it allows you to service more requests more quickly while you are ramping up your server farm.

Regards,
SB
 
Add whatever management / orchestration solution that now has to keep track of which blades are 100% and which are not and assign users / streams to blades. You want things as identical as possible, just to simplify your life.
 
Add whatever management / orchestration solution that now has to keep track of which blades are 100% and which are not and assign users / streams to blades. You want things as identical as possible, just to simplify your life.

can just be flagged in the OS of what the hardware is capable and it wont spin up as many instances.
 
Of course that comes at the cost of higher energy cost per service request, but it allows you to service more requests more quickly while you are ramping up your server farm.
And this is why I say this is Microsoft's nightmare scenario because you measure server cost in terms of task/users per server against average running cost, most of which as you say is power. Green energy, which Microsoft are into, is more expensive that fossil fuel power. If you have 120 cores in a blade and even a dozen have borked cores, you're running at a loss so the idea of putting broken cores on mass as an actual strategy is insane unless Microsoft don't care about cost.

Microsoft are a publicly traded company owned by people who do care about costs and their server business, on which they do report in detail, is competing with Amazon and Google. Microsoft cannot afford any perception that their farms are not efficient, reliable and economical and that does not change just because this is a niche service for one of Microsoft's own business units.
 
If xsx core is capable of running four instances of an xone s, is it still (quite probably) more efficient than xone s if it only runs 3?
 
There is probably no real costs saving because you end up spending money elsewhere. A 3/4 functional core running less instances doesn't require the level of RAM that a fully functional core would require. So what do you do? Waste money on unnecessary RAM or fund separate production lines and QA/QC costs?

Using only fully functional chips may increased cost per chip by a few dollars, but it streamlines downstream manufacturing and production.
 
And this is why I say this is Microsoft's nightmare scenario because you measure server cost in terms of task/users per server against average running cost, most of which as you say is power. Green energy, which Microsoft are into, is more expensive that fossil fuel power. If you have 120 cores in a blade and even a dozen have borked cores, you're running at a loss so the idea of putting broken cores on mass as an actual strategy is insane unless Microsoft don't care about cost.

Microsoft are a publicly traded company owned by people who do care about costs and their server business, on which they do report in detail, is competing with Amazon and Google. Microsoft cannot afford any perception that their farms are not efficient, reliable and economical and that does not change just because this is a niche service for one of Microsoft's own business units.

Absolutely, but not necessarily applicable in this case.

We know that MS intends to use the SOC for multiple purposes. We know that they are intent on pushing their streaming gaming service. We know that they've burned money in the past in order to build up a foundation for the future. The Xbox itself is a case of burning cash short term for long term gains.

It is entirely within the realm of possibility that they see a short term loss in terms of inefficiencies worth it for a long term potential of getting as many users as quickly as possible with the limited monthly/yearly SOC production they will have.

Obviously if XBSX flops there will be more supply of fully functioning chips available to build up the server infrastructure. Likewise if there streaming ambitions flop there will be less need and more supply for their consoles.

But noone goes in planning for failure, you want to plan for success and then if you don't succeed you do what you can to salvage things.

Microsoft can afford to be aggressive because they have cash to burn. You've mentioned investors and investors have also complained about MS sitting on a pile of cash. While not optimal in the short term this is something they can present to investors as a use of cash with long term potential for pulling in greater amounts of revenue. Minecraft is an example of this sort of investor strategy. Viewed by many outside the company as a waste of money, it's certainly paying off in terms of consistent high revenue.

Regards,
SB
 
It is entirely within the realm of possibility that they see a short term loss in terms of inefficiencies worth it for a long term potential of getting as many users as quickly as possible with the limited monthly/yearly SOC production they will have.

Again this is a nightmare scenario. The only scenario where there would be a strategy or need to put broken CPUs into servers is if Microsoft could not produce enough good CPUs for Xbox and xCloud production and the CPUs that were in production were exhibiting high failure rates.

This would be a disaster. How would this even happen? :runaway:
 
If xsx core is capable of running four instances of an xone s, is it still (quite probably) more efficient than xone s if it only runs 3?

I wonder how many instances of an Xbox Series S each blade can run. With the rumor being that the Series S will be 4TF and have 10GB of RAM, three instances fit nicely into a blade. (Four instances of XBOs would need 32GB of RAM on a blade.)

It would make also sense to for them to have tiers of performance:

1. Xbox One
2. Next Gen 1080p streams (Series S)
3. Next Gen 4k Stream (Series X)
 
I wonder how many instances of an Xbox Series S each blade can run. With the rumor being that the Series S will be 4TF and have 10GB of RAM, three instances fit nicely into a blade. (Four instances of XBOs would need 32GB of RAM on a blade.)

It would make also sense to for them to have tiers of performance:

1. Xbox One
2. Next Gen 1080p streams (Series S)
3. Next Gen 4k Stream (Series X)

I assume the CPU on series S is the same as X give or take, with the being the case then even with less gpu and memory it would be hard for a full Apu to deliver anything more than a single instance.

At that point does it make sense to offer?

You may be able to reduce the memory on the motherboard but you loose the flexibility to offer the 4k series X stream.

You could encode a series X stream as 1080p if that was the requirement.
 
Again this is a nightmare scenario. The only scenario where there would be a strategy or need to put broken CPUs into servers is if Microsoft could not produce enough good CPUs for Xbox and xCloud production and the CPUs that were in production were exhibiting high failure rates.

This would be a disaster. How would this even happen? :runaway:

Just like any new console generation. You can NEVER have enough SOCs at launch. :p

Regards,
SB
 
I wonder how many instances of an Xbox Series S each blade can run. With the rumor being that the Series S will be 4TF and have 10GB of RAM, three instances fit nicely into a blade. (Four instances of XBOs would need 32GB of RAM on a blade.)

It would make also sense to for them to have tiers of performance:

1. Xbox One
2. Next Gen 1080p streams (Series S)
3. Next Gen 4k Stream (Series X)

I would wager they would just have lockhart blades.
 
Back
Top