Series X Refresh and Next Generation Xbox Hardware

[Globalisateur]

Being available doesn’t make it cheaper for them however. XSX had conservative clocks to increase yield. Silicon size may reduce sure, but it only is beneficial if you can actually get more usable chips per wafer out of it.

That's not what the massive heatsink suggests. At launch the mass of copper was about 50% more than PS5 heat sink (both with 7nm APU). PS5 heatsink was actually quite light but bigger (more volume), which is why it was very efficient (and didn't need vapor cooling).

Also the whole cooling setup was very complicated and expensive with that second mini board. The fact that the CPU clocks are a bit higher than PS5 suggest their yield could be actually worse than PS5.

 

[iroboto]

That's not what the massive heatsink suggests.

Heat sinks aren’t suggestive of yield. Xbox one has a massive heat sink it didn’t need.

Power output and frequency requirements are part of the yield calculation.

 

[iroboto]

Yeah, supply chain is hard. Just because you can get one part cheaper doesn't make the TCO go down. Amazon's second largest consumer of EC2 is their own supply chain sourcing department, which is continuously solving the travelling salesman problem for just about every SKU they carry. Where do they source it from, how is it transported, what tarrifs might apply, where is it warehoused, and who is buying it and where do they generally live? I happen to know the guy who runs the department, and despite how much money they spend on compute (tens of millions per month) they make it back a hundredfold or more every year just making sure supply chain is actually working at peak efficiency.

And it has to be said repeatedly, this is not Microsoft’s domain expertise. They are not churning out hardware products all the time for a variety of products where they can take advantage of a massive logistics pipelines and channels to exploit cost advantages.

 

[davis.anthony]

I think it should be fairly obvious that XSX GPU clock rates were really conservative given how high desktop RDNA2 GPU's clocked.

If they weren't that much slower in XSX for yield purposes, then what other reason?

 

[iroboto]

I think it should be fairly obvious that XSX GPU clock rates were really conservative given how high desktop RDNA2 GPU's clocked.

If they weren't that much slower in XSX for yield purposes, then what other reason?

Right, nailed it.

A comparable would be that 5pro is on 6nm.

52CUs should clearly be able to make it as high as 60CUs in clockspeed. All things equal the XSX would use less power here.

series consoles are very conservative.

 

[RDGoodla]

A comparable would be that 5pro is on 6nm.

Is Pro on 4nm?

 

[iroboto]

Is Pro on 4nm?

there are several sites reporting it’s very highly likely to be 6nm.
I don’t know any PC components on 4nm

 

[Charlietus]

there are several sites reporting it’s very highly likely to be 6nm.
I don’t know any PC components on 4nm

I don't know how it could be. It's too small and apparently, it's also silent.

And Nvidia is on 4nm.

 

[fehu]

Zen 5 too

 

[Subtlesnake]

there are several sites reporting it’s very highly likely to be 6nm.
I don’t know any PC components on 4nm

I don't see how Sony is going to be able to match the 7700 XT with perhaps 50W less power on a significantly worse process.

 

[iroboto]

Spec Analysis: PlayStation 5 Pro - the most powerful console yet

Tech specs for PlayStation 5 Pro have leaked - Digital Foundry verifies them and delivers analysis and further information.

www.eurogamer.net

but the evidence points to PS5 Pro running on the same 6nm process as the Slim. PS5 Pro only has limited clock speed increases (or actual decreases potentially) and the size of the GPU architecturally has not doubled in the way it did with PS4 Pro

 

[iroboto]

I don't know how it could be. It's too small and apparently, it's also silent.

And Nvidia is on 4nm.

which gpu?

The XSX is smaller, and is only has 8 less CUs, and is silent as a whisper. It's definitely doable.

 

[Subtlesnake]

Spec Analysis: PlayStation 5 Pro - the most powerful console yet

Tech specs for PlayStation 5 Pro have leaked - Digital Foundry verifies them and delivers analysis and further information.

www.eurogamer.net

Rich's analysis doesn't make any sense to me. You're using the same process (6nm) as the existing Slim but suddenly it's expected that you can find an extra 45% more performance for the same power, when none of the RDNA 3 derivatives have shown that kind of architectural improvement? I guess it's not impossible, but I don't see how you can say it is likely, especially when 4nm isn't going to be more expensive on a per transistor basis.

 

[Globalisateur]

Rich's analysis doesn't make any sense to me. You're using the same process (6nm) as the existing Slim but suddenly it's expected that you can find an extra 45% more performance for the same power, when none of the RDNA 3 derivatives have shown that kind of architectural improvement? I guess it's not impossible, but I don't see how you can say it is likely, especially when 4nm isn't going to be more expensive on a per transistor basis.

Just the 5% overclock from PS5 (with 66% more CUs) and the size of the console (smaller than PS5 launch) would suggests 5nm max. 4nm possible. APU is likely smaller than PS5 7nm launch APU too, like PS4 Pro.

which gpu?

The XSX is smaller, and is only has 8 less CUs, and is silent as a whisper. It's definitely doable.

RDNA1.5 vs RDNA3.5 + added RT + AI upscaling. PS5 Pro is using different CUs with much more cache and custom silicon vs XSX APU. I can't believe you would even make such naive comparison.

 

[davis.anthony]

I don't see how Sony is going to be able to match the 7700 XT with perhaps 50W less power on a significantly worse process.

Because power draw doesn't affect performance in such a linear way.

A 50w reduction could only reduce performance by 2-3% which isn't going to be all that noticeable.

 

[Charlietus]

Tom handerson has leaked pretty much everything about the pro, and he says the power draw is around 250 watts and that it's quiet:

PS5 Pro won't ramp up your electricity bill or get too loud, says insider

New PS5 Pro details suggest the console will stay cool and quiet and won't be a massive power draw

www.theshortcut.com

You aren't getting 70% more CU's in there without a major spike in power consumption. The slim is around 200 watts, and the pro is just around 50 watts more. 6nm is unlikely, unless rDNA 4 is so much more efficient.

 

[iroboto]

RDNA1.5 vs RDNA3.5 + added RT + AI upscaling. PS5 Pro is using different CUs with much more cache and custom silicon vs XSX APU. I can't believe you would even make such naive comparison.

The CUs only have dual issue otherwise it’s 2 more dual CUs per SE.

The improved RT units or support for sparsity won’t increase the size of the chip significantly. Cache. Maybe. But we don’t know the cache numbers 5pro is working with. L2 cache is still small due to it being 256bit bus.

AI is just done inside the CUs, no extra silicon required there.

From my perspective of things, I’m not drawing the same conclusions as you here.

 

[Subtlesnake]

The XSX is smaller, and is only has 8 less CUs, and is silent as a whisper. It's definitely doable.

The XSX is also clocked much lower.

Because power draw doesn't affect performance in such a linear way.

A 50w reduction could only reduce performance by 2-3% which isn't going to be all that noticeable.

Well we see a ~15% performance difference for just a 20W power difference between the 7800 XT and 7700 XT. Though yes, the 7700 XT is smaller and clocked higher so it's further up its power curve.

However I agree that you could do the scaling at 5nm. But then going back to 6nm seems like a lot to ask.

 

[davis.anthony]

Well we see a ~15% performance difference for just a 20W power difference between the 7800 XT and 7700 XT. Though yes, the 7700 XT is smaller and clocked higher so it's further up its power curve.

However I agree that you could do the scaling at 5nm. But then going back to 6nm seems like a lot to ask.

In this thread about 7700XT undervolting a

In Time Spy and Fire Strike Extreme, after setting the card's undervoltage settings properly, the FPS difference between 208W (-10%) and 262W (+15%) was:

3.6% fewer FPS in Time Spy (cores at 2580-2630 MHz @-10%, ~120 MHz higher @+15%)

2.7% fewer FPS in Fire Strike Extreme (cores at 2670-2720 MHz @-10%, ~130 MHz higher @+15%)

With 20.6% less power used

There's a lot of headroom in terms of power scaling.

 

[Kaotik]

And Nvidia is on 4nm.

4N is based on N5, not N4

RDNA1.5 vs RDNA3.5 + added RT + AI upscaling. PS5 Pro is using different CUs with much more cache and custom silicon vs XSX APU. I can't believe you would even make such naive comparison.

XSX/S is RDNA2, not 1.5 like PS5