Xbox One (Durango) Technical hardware investigation

[Jwm]

I thought they were saying 6x to 8x more powerful than 360? So maybe he was just picking the high range. With cloud it was 10x to 40x or something. But those were numbers coming out after the reveal.

Anyway, if we take the whole system and not the GPU isn't 8x really a believable number? CPU is no slouch plus it has SHAPE taking the audio load off of it, etc.

 

[AlphaWolf]

This thread needs about a thousand wavy arm emoticons.

 

[warb]

3x, 8x and now 10x....again.

Phil Spencer -

Phil Spencer -
http://gamingbolt.com/phil-spencer-xbox-one-10-times-more-powerful-than-xbox-360-cloud-gives-more

seems like MS's PRs might be far away from knowing the hardware. Maybe AMD is probably in a big pickle with microsoft's chips?

Or it's just not a very useful or reliable metric, to measure things in Xbox's. How many Xbox's was an Xbox 360?

He was more specific with 10x there.

 

[mboeller]

3x, 8x and now 10x....again.

Phil Spencer -

Phil Spencer -
http://gamingbolt.com/phil-spencer-xbox-one-10-times-more-powerful-than-xbox-360-cloud-gives-more

seems like MS's PRs might be far away from knowing the hardware. Maybe AMD is probably in a big pickle with microsoft's chips?

and the roughly 10x performance for compute is imho correct.

The X360 GPU cannot be used for compute (=GPGPU) therefore only the CPU is used for the comparison.

X360 CPU = 115GFLops

10 x X360 = 1150 GFlops or 1.15 TFlops.

1.15 TFlops is a little bit on the low side, the correct number should be 1,33 TFlops ( 11,5 times the X360 ) but well good enough for me from marketing.

 

[Proelite]

8x is the official GPU multiplier MS told devs during their Durango summit in Feb 2012.

 

[XpiderMX]

and the roughly 10x performance for compute is imho correct.

The X360 GPU cannot be used for compute (=GPGPU) therefore only the CPU is used for the comparison.

X360 CPU = 115GFLops

10 x X360 = 1150 GFlops or 1.15 TFlops.

1.15 TFlops is a little bit on the low side, the correct number should be 1,33 TFlops ( 11,5 times the X360 ) but well good enough for me from marketing.

Isn't MEMEXPORT some kind of GPGPU?

 

[Shifty Geezer]

The X360 GPU cannot be used for compute (=GPGPU)

It most certainly can. GPGPU originated as using GPUs to compute by structuring data as 2D texture/image buffers. As such, any GPU can be used for GPGPU. Changes have been made to make compute easier and more effective, so more jobs can be offloaded to the GPU, but quite how one calculates what constitutes a GPGPU FLOP or not for comparison is beyond my reckoning!

 

[Rangers]

8x is the official GPU multiplier MS told devs during their Durango summit in Feb 2012.

They've said 10X in recent times too.

If 1.2 TF is real, I dont have a problem with the overall concept of it being 10X as powerful, looked at "in totality". By flops, it wont be...

GPU flops is 5X+efficiency gains that can be very substantial. RAM is 16X. CPU? How do you quantify ESRAM?

If Wii U for example is really 160 GCN shaders, that means it is ~176 Gflops I think, and it keeps up with current gen, that would say a lot about GCN efficiency.

 

[(((interference)))]

Ironically, the ESRAM may end up being the much vaunted 'special sauce' in the design after all.

I'm very interested to see how the multiplat titles end up looking come November.

 

[Silenti]

Will not the real test come when engines are written specifically for the new hardware and all their nooks and crannies are known and taken into account rather than 1st gen titles where the engines may have to do double duty on the last gen hardware?

 

[Silent_Buddha]

Will not the real test come when engines are written specifically for the new hardware and all their nooks and crannies are known and taken into account rather than 1st gen titles where the engines may have to do double duty on the last gen hardware?

Most likely. From some of the bits dropped here and there by multiplatform game developers, they apprear to be designing to one target for both PS4 and Xbox One, what they call their Next Gen target. This makes sense as they have a limited development window in order to hit launch + first year. So limiting art assets, engine capabilities, etc. to one design target allows them to hit launch/first year targets. Going forward after the first year (or perhaps even late in the first year) is, IMO, when developers might feel they have the time to budget in differences (though I'm doubtful how many will actually do this).

So, I'm not expecting the launch titles (multiplatform) to show much differences.

The 1st-2nd year will be when PS4 might show some graphical advantages. If system design comes into play and developer's can leverage the ESRAM (and whatever else) we might see things even up again later in the generation. Or sooner. That's, of course, all dependant on how those design differences can be leveraged.

I fully expect that some things will always be slower just due to the base capabilities. But some things might actually be faster (any operations that can be done within the ESRAM pool without going to main system RAM, for example). The question becomes how that eventually works out and whether the advantages cancel out the disadvantages in an average scene such that you basically have relatively equal rendering performance.

Regards,
SB

 

[Strange]

Most likely. From some of the bits dropped here and there by multiplatform game developers, they apprear to be designing to one target for both PS4 and Xbox One, what they call their Next Gen target. This makes sense as they have a limited development window in order to hit launch + first year. So limiting art assets, engine capabilities, etc. to one design target allows them to hit launch/first year targets. Going forward after the first year (or perhaps even late in the first year) is, IMO, when developers might feel they have the time to budget in differences (though I'm doubtful how many will actually do this).

So, I'm not expecting the launch titles (multiplatform) to show much differences.

The 1st-2nd year will be when PS4 might show some graphical advantages. If system design comes into play and developer's can leverage the ESRAM (and whatever else) we might see things even up again later in the generation. Or sooner. That's, of course, all dependant on how those design differences can be leveraged.

I fully expect that some things will always be slower just due to the base capabilities. But some things might actually be faster (any operations that can be done within the ESRAM pool without going to main system RAM, for example). The question becomes how that eventually works out and whether the advantages cancel out the disadvantages in an average scene such that you basically have relatively equal rendering performance.

Regards,
SB

I seriously doubt the eSRAM + DDR3, even if being totally efficient and able to match or better GDDR5 (which is quite unlikely), is able to make up for the 6 CU deficit.

There are just too many stacked odds against it that even if the perfect storm comes along and everything falls into place, it feels a bit far fetched to say the entire package can match the competition.

A workaround is what it is, a workaround. It's rarely the case that they'll be as efficient and elegant as the straight up solution.

And in this case the design choices cascaded throughout the architecture design to present a complicated design that won't be as easy to work with.

 

[Silent_Buddha]

I seriously doubt the eSRAM + DDR3, even if being totally efficient and able to match or better GDDR5 (which is quite unlikely), is able to make up for the 6 CU deficit.

There are just too many stacked odds against it that even if the perfect storm comes along and everything falls into place, it feels a bit far fetched to say the entire package can match the competition.

A workaround is what it is, a workaround. It's rarely the case that they'll be as efficient and elegant as the straight up solution.

And in this case the design choices cascaded throughout the architecture design to present a complicated design that won't be as easy to work with.

It's a workaround in some aspects and not a workaround in others. For anything requiring more than 32 MB of data, you're still going to have to hit system RAM. If it fits into ESRAM and you can reuse that data multiple times you are going to have a huge latency advantage. Similar to a CPU if you are able to avoid hitting system RAM by staying within L2 or L3 caches. I suppose you can also consider CPU L1, L2, and L3 caches to be "workarounds" for just having faster memory. Except CPU's went well past the point of being able to have fast enough memory almost 2 decades ago when CPU multipliers were introduced because memory speeds could not keep up with CPU processing speeds.

CPUs in the past (and even currently still) show that fast low latency cache can trump having significantly faster access to main system RAM. It's also one of the reasons why there's a very small performance delta in general tasks between say DDR3 1333 and DDR3 2133. Take out that L3 and reduce the size of L2 cache and you'd see a rather significant performance delta between those two memory speeds. Enough that a significantly slower CPU with that L3 and slower system RAM would outperform a much faster CPU without that L3 and faster system RAM. One of the reasons Intel CPUs are so much faster than AMD CPUs is due to the lower latency combined with larger sizes.

The ESRAM doesn't exactly serve the same purpose as a CPU L3 cache, but it can function somewhat similarly if it allows you to avoid hitting main system RAM. Considering we haven't seen much of this out in the wild means that we have no way of judging it's potential impact except by looking at X360's use of the EDRAM which was far more limited.

Prior to the introduction of CPU caches in computers there was some debate as to whether it was worthwhile. It was argued by many that you could achieve better results with faster memory and/or faster CPU speeds than spending valuable space on a CPU cache. Now you basically cannot compete in the CPU sector without CPU cache.

Don't take this to mean that I expect ESRAM alone to allow for average rendering time to be relatively equal when fully exploited. But I'm not so close minded as to discount it entirely either. As well, there are other systems in play other than just the ESRAM.

Regards,
SB

 

[Gubbi]

The ESRAM isn't a workaround, it is a core feature of XB1.

XB1 and PS4 are made for different design points.

MS has put more an emphasis on long term cost reductions and lower power consumption while Sony has put more emphasis on getting the most performance out of the silicon.

Cheers

 

[(((interference)))]

Will not the real test come when engines are written specifically for the new hardware and all their nooks and crannies are known and taken into account rather than 1st gen titles where the engines may have to do double duty on the last gen hardware?

Of course, but it'll be better than nothing - which is what we have at the moment.

And ESRAM definitely wasn't included just to make DDR3 bearable - I've been told the designers realised that ESRAM could give the system a significant performance boost, especially for compute.

 

[Strange]

The ESRAM isn't a workaround, it is a core feature of XB1.

XB1 and PS4 are made for different design points.

MS has put more an emphasis on long term cost reductions and lower power consumption while Sony has put more emphasis on getting the most performance out of the silicon.

Cheers

Yes I understand they are made for different design points. However, if 8GB GDDR5 was available 2 years ago when they were forming the console, it doesn't seem too far fetched that Microsoft would just scrap the whole eSRAM and just use GDDR5.

GDDR5 simply didn't have the capacity to fit what they wanted to do with the console at that time and they had to find something else to solve their issues when using DDR3.

It doesn't seem logical to me that the eSRAM is a "core feature". If it is a core feature, it should be at the top of the design priority.

The Xbox one is designed to be a trojan horse to your living room so that it can do everything and replace your other stuff. Does the eSRAM fit this design? It feels very orthogonal to this design goal. Does 8GB ram fit this design? Very much so.

It's very hard to imagine that they designed the eSRAM into the Xbox one before they planted the 8GB DDR3 into the device.

Or they have very fucked up engineers, which I don't believe is the case.

 

[warb]

I believe there was eSRAM in there before there was 8GB DDR3.

 

[Rangers]

I believe there was eSRAM in there before there was 8GB DDR3.

Supposedly according to intereference who has a MS source, the 8GB DDR3 was implemented in late 2011, after it was decided the app OS wouldn't fit.

Presumably previous, the console had 4GB on a 128 bit bus? I guess that would have still necessitated ESRAM.

 

[Tchock]

If Wii U for example is really 160 GCN shaders, that means it is ~176 Gflops I think, and it keeps up with current gen, that would say a lot about GCN efficiency.

sorry for OT- but when did Wuu switch to 160 x GCN?!

----

I wonder low GDDR5 can power down at non load uses. It used to be pretty bad with AMD's 4000 series, then got progressively better- but I suspect that was because VRAM isn't linked with the system pool.

 

[Rangers]

Oops it was 160 VLIW 5 I guess. My mistake.

Would speak even better of efficiency gains in that case! Of course I'm not sure if Wii U has 160 or 320 shaders, but I suspect the former on the "always bet the low side with Nintendo" theory.