Xbox Series... M?

[LiveGamer]

Could you link the source for that please.

Could possibly see that being the case if it never already had disabled CU's for redundancy.

I can't speak to the first although I guess they can use defect series x's to play backwards compatible or less demanding games ?

For the second in terms of the actual screen power usage I don't think there will be a huge difference between 720p and 1080p. In terms of hardware they could lock it to 720p sure but I think they would go with 1080p to better match the full hd offering as a next gen portable

Can't seem to find the source. From what I remember they are using defective Series X chips that have 48 active CUs and upping the clock to 1979mhz to reach 12.155 TFs.

And I would love a 1080p screen but it makes sense to me to run the games at 720p when portable to conserve power.

 

[Silent_Buddha]

Can't seem to find the source. From what I remember they are using defective Series X chips that have 48 active CUs and upping the clock to 1979mhz to reach 12.155 TFs.

And I would love a 1080p screen but it makes sense to me to run the games at 720p when portable to conserve power.

Likely from here.

Xbox Series X SoC: Power, Thermal, and Yield Tradeoffs (anandtech.com)

XBS-X uses 26 WGPs but it could use 28 if MS wanted. 26 WGPs was chosen due to better yield.

XBOX Cloud gaming can use anything >= 24 WGPs. MS will then vary the clock speed in order to maintain the correct TFLOPs.

I had mentioned previously that MS were using XBS-X SOC that weren't suitable for the consoles in their XBS-X server blades.

Regards,
SB

 

[LiveGamer]

Likely from here.

Xbox Series X SoC: Power, Thermal, and Yield Tradeoffs (anandtech.com)

XBS-X uses 26 WGPs but it could use 28 if MS wanted. 26 WGPs was chosen due to better yield.

XBOX Cloud gaming can use anything >= 24 WGPs. MS will then vary the clock speed in order to maintain the correct TFLOPs.

I had mentioned previously that MS were using XBS-X SOC that weren't suitable for the consoles in their XBS-X server blades.

Regards,
SB

Thanks for this! Question is there any reason why the cutoff is 24 WGPs? As they move to lower nodes could they theoretically use even less WGPs with higher clocks without any issues?

 

[Silent_Buddha]

Thanks for this! Question is there any reason why the cutoff is 24 WGPs? As they move to lower nodes could they theoretically use even less WGPs with higher clocks without any issues?

If I were to hazard a guess, there's a couple things that immediately come to mind.

• One would be due to power consumption. The fewer WGPs the higher you'll have to clock the chip. That could take it beyond the knee of the power curve leading to exponentially higher power consumption.
  • In a data center that would be undesirable as not only would it increase the power consumption from the blade itself but would increase the heat output which would potentially lead to higher cooling costs for the data center.
• The other would be chip yield. It may be that the number of chips with fewer than 24 working WGPs is vanishingly small.
  • Basically at that point it may not be worth it to qualify another blade configuration just to account for a miniscule number of chips.

Regards,
SB

 

[Proelite]

If I were to hazard a guess, there's a couple things that immediately come to mind.

• One would be due to power consumption. The fewer WGPs the higher you'll have to clock the chip. That could take it beyond the knee of the power curve leading to exponentially higher power consumption.
  • In a data center that would be undesirable as not only would it increase the power consumption from the blade itself but would increase the heat output which would potentially lead to higher cooling costs for the data center.
• The other would be chip yield. It may be that the number of chips with fewer than 24 working WGPs is vanishingly small.
  • Basically at that point it may not be worth it to qualify another blade configuration just to account for a miniscule number of chips.

Regards,
SB

I actually think that data centers are equipped to cool much hotter blades than Xsx blades. Think of those 64-core servers chips...

 

[Silent_Buddha]

I actually think that data centers are equipped to cool much hotter blades than Xsx blades. Think of those 64-core servers chips...

They can, if needed. They would like to reduce reliance on high output cooling as much as possible, however, as that's one of the larger costs associated with running a data center.

With XBS-X blades, matching the performance of an XBS as closely as possible is more important than extracting the most performance possible from the SOC. So, reducing the heat output as much as possible is likely one of the goals. I'd be willing to bet that any XBS-X SOC that is capable of having all 28 WGPs enabled are used in the data center as those can be run at a lower clock speed. Meanwhile any XBS-X SOC that can run with 24 WGPs enabled are used as that represents a relatively small increase in heat output. Then anything below that is discarded due to a combination of the reasons I postulated.

Meanwhile, XBS-X SOCs that can run with 26 WGPs enabled but not 28 are only used if they don't have enough SOCs that run with 28 or 24 WGPs. Not sure how the AMD SOC handles non-power of 2 WGPs, but it's possible that ones with 25 or 27 WGPs enabled could possibly be used. I mention this as for data center use they mention >= 24 WGPs and not 24, 26 or 28 in a similar way to how the consoles can use 26 or 28.

Again purely speculation as I don't know anyone that works at a MS data center.

Regards,
SB

 

[LiveGamer]

They can, if needed. They would like to reduce reliance on high output cooling as much as possible, however, as that's one of the larger costs associated with running a data center.

With XBS-X blades, matching the performance of an XBS as closely as possible is more important than extracting the most performance possible from the SOC. So, reducing the heat output as much as possible is likely one of the goals. I'd be willing to bet that any XBS-X SOC that is capable of having all 28 WGPs enabled are used in the data center as those can be run at a lower clock speed. Meanwhile any XBS-X SOC that can run with 24 WGPs enabled are used as that represents a relatively small increase in heat output. Then anything below that is discarded due to a combination of the reasons I postulated.

Meanwhile, XBS-X SOCs that can run with 26 WGPs enabled but not 28 are only used if they don't have enough SOCs that run with 28 or 24 WGPs. Not sure how the AMD SOC handles non-power of 2 WGPs, but it's possible that ones with 25 or 27 WGPs enabled could possibly be used. I mention this as for data center use they mention >= 24 WGPs and not 24, 26 or 28 in a similar way to how the consoles can use 26 or 28.

Again purely speculation as I don't know anyone that works at a MS data center.

Regards,
SB

Sounds plausible and to circle back to @eastmen 's question. For Xbox to be able to change the CU counts like this for xcloud signifies they could potentially do the same for a portable Series S? He hypothesized lower CU counts could be used as they drop to 5nm and then up the speed to match the existing 4TFs. The lower node should counter any extra heat from the increased clock.

 

[Silent_Buddha]

Sounds plausible and to circle back to @eastmen 's question. For Xbox to be able to change the CU counts like this for xcloud signifies they could potentially do the same for a portable Series S? He hypothesized lower CU counts could be used as they drop to 5nm and then up the speed to match the existing 4TFs. The lower node should counter any extra heat from the increased clock.

The problem becomes cost. It's now taking multiple years for a new node to get to the point where a straight shrink (same # of transistors but smaller chip) is the same cost as the node you are coming from, much less getting to the point where you can save money with a shrink.

That's not how it used to be pre-7 nm. Going from 28 nm to 14nm was both a cost and area savings fairly quickly (it still took a while, but not years) once volume production started. Going from 7 nm to 5 nm is going to cost more even with volume production. So you end up with an area savings but a cost increase.

And that's only going to get worse with smaller nodes. It may come to a point when it is not possible to decrease cost with a shrink to a smaller node even with say, greater than 5 years from start of volume production of the smaller node.

So, might it be possible to make such a chip? It's possible. Will it be economically feasible for a portable console when most console gamers are far less willing to spend as much on a full sized console (much less a portable console) as they are on a smartphone?

Basically a portable XBS-S at a smaller node is likely going to cost more (and not just a little bit more) than an XBS-S. Remember it will also need batteries, a display, a more compact and tightly integrated PCB, likely with more layers, etc.

It's a conundrum facing both Sony and MS as they start to look towards a future next generation console.

Regards,
SB

 

[orangpelupa]

If they made Xbox series M also able to be used docked. Like a Nintendo Switch. Then it may have wide enough market reach. But may cannibalize Xbox series s....

 

[eastmen]

If they made Xbox series M also able to be used docked. Like a Nintendo Switch. Then it may have wide enough market reach. But may cannibalize Xbox series s....

I think by the time a Series M gets replaced the Series S will be $200ish while the Series M would likely be in the $400 range. So I wouldn't see that as being an issue imo. They may already be moving on to the Series s/x refreshes . I'd wager the series s refresh would be closer to 10tflops while the series x replacement would be closer to 20tflop. So people still serious enough for tv gaming might still opt for one of those.

I can tell you that I took a train for an interview earlier today and in the morning I was playing horizon on my pc and on the way to the interview I was playing it on my steam deck. It was a great experience and I think a lot of gamers would enjoy that using the xbox eco system

 

[Proelite]

6800U handheld with 12 CU GPU.

Xbox One has 12 CU GPU.

 

[LiveGamer]

6800U handheld with 12 CU GPU.

Xbox One has 12 CU GPU.

The 6800U is 3.4 TFs at 2.2 Ghz. That's overkill for a Xbox One. They could halve the CUs for a cheaper and more power efficient cheap. Or just wait a bit and get closer to 20CUs and lower the clocks to 1.5Ghz to match the Series S.

Either way exciting times ahead I feel!!

 

[eastmen]

The 6800U is 3.4 TFs at 2.2 Ghz. That's overkill for a Xbox One. They could halve the CUs for a cheaper and more power efficient cheap. Or just wait a bit and get closer to 20CUs and lower the clocks to 1.5Ghz to match the Series S.

Either way exciting times ahead I feel!!

It's all about that 5nm I'd wager. With 5nm AMD should be able to get the series s at 15 watts.

I'm more interested in if it would be smarter for MS to just hit 5nm with a refreshed series s/x and introduce series m. All of them zen 4+ rdna 3. If the reports of the increases for raytracing with rdna 3 are true it could be a very smart choice on MS's part.

 

[orangpelupa]

6800U handheld with 12 CU GPU.

Xbox One has 12 CU GPU.

WHOA it must have a much better cooling system than my 13 inch lenovo laptop hahaha damn

 

[LiveGamer]

It's all about that 5nm I'd wager. With 5nm AMD should be able to get the series s at 15 watts.

I'm more interested in if it would be smarter for MS to just hit 5nm with a refreshed series s/x and introduce series m. All of them zen 4+ rdna 3. If the reports of the increases for raytracing with rdna 3 are true it could be a very smart choice on MS's part.

What are the reports on improved Ray Tracing? I would think games would have to be coded woth those improvements in mind no? With base Series X|S still around that seems doubtful...

 

[eastmen]

What are the reports on improved Ray Tracing? I would think games would have to be coded woth those improvements in mind no? With base Series X|S still around that seems doubtful...

Its rumors and what not , you can look int the section of these forums devoted to it.

I don't see why they would need to be coded with the improvements in mind. Were all pc games re-coded when nvidia went from the 20x0 series to the 30x0 series ? Will they again be re-coded for the 40x0 series ?

 

[AzBat]

Jez seems to think that such a device may require dedicated support. Otherwise it will have to play games from the cloud. Granted he also says he knows nothing of such of device. He's just talking hypothetical. LOL

Would you be okay with an Xbox handheld that was mostly cloud-based, but maybe had a small subset of native games (like 20 games) that are popular titles like Halo multiplayer, Fortnite, Rocket League, Fall Guys, etc.

https://twitter.com/x/status/1542589533933473799

Theory is that it would launch** with a small subset of games, since devs would have to target the device specifically. Maybe more would come later, but there'd be no hard guarantee, since it would be up to devs.

https://twitter.com/x/status/1542591102129143813

Tommy McClain

 

[eastmen]

Jez seems to think that such a device may require dedicated support. Otherwise it will have to play games from the cloud. Granted he also says he knows nothing of such of device. He's just talking hypothetical. LOL

https://twitter.com/x/status/1542589533933473799

https://twitter.com/x/status/1542591102129143813

Tommy McClain

Would be very weird in my opinion.

If it would only support some games that would mean it would be comparable to xbox one or xbox series s. If its an xbox one why not just put all xbox one level games as native ? Heck just throw in the same apu as in the steam deck and it can easily play xbox one games and would likely do so on a low power budget if they are native. If it can play series s games then just have it play them natively.

Seems really odd.

 

[orangpelupa]

maybe the dedicated tier for XSM is optional? so new games with special mode for XSM can take advantage of XSM features and have XSM label (instead of XS label)

if its not optional... yeesh... that would be pretty abysmal IMO. as why devs even want to dedicate extra work for XSM?

 

[eastmen]

maybe the dedicated tier for XSM is optional? so new games with special mode for XSM can take advantage of XSM features and have XSM label (instead of XS label)

if its not optional... yeesh... that would be pretty abysmal IMO. as why devs even want to dedicate extra work for XSM?

It would be odd as the majority of games support variable resolutions. So even if the mobile device needed to render at a lower resolution it should be trivial to add in the lower bound resolution. But like I said it seems odd since AMD is so close to the series s in a mobile apu now. I think 5nm would let them hit it