Xbox One (Durango) Technical hardware investigation

[mosen]

Yes last gen, Sony's talk of super computer architecture and harder to develop for but worth it is the exact same argument that is being presented here.

I was trying to say that Sony said same thing even this gen, there is no need to comparing XB1 to last gen arguments about PS3. XB1 is much easier to develop compered to XB360. XB1 has the same API as Windows and it shouldn't be hard to develop. Go and read CD PROJECT interview.

The Xbox One is pretty easy to understand because not just the hardware is similar to the PC, but everything like the SDK, the API is really similar to what you would find on a PC

http://www.lazygamer.net/general-ne...ill-get-more-powerful-is-easier-to-work-with/

But right now it's harder to master.

 

[Deleted member 11852]

but you can introduce new features faster. The developer has a save state of a OS, that is worth a lot.

You may be able to introduce features faster (this is arguable) but games will take years to adopt them because no developer substantially into development is gong to go back and rewrite code to use an API that is no longer compatible with the code they've written. That tales time and time costs money.

For example, suppose next week Microsoft roll out a new version of DirectX with massive performance improvements, but which also introduces a bunch of incompatibilities with the current API. Do you really think the guys working on The Division or Destiny, are going to go back and re-write tons of code? Not a chance. The games that will take advantage of that API will be those barely started and won't be out for a year or more.

With the PS4, the is not that stable, it could break or produce bugs with future iterations. That is why updating the system software is so risky.

And yet amazingly, DirectX, OpenGL and Sony across four consoles (PlayStation 3 and 4, PSP, Vita) have managed to improve and deploy updates to the graphics APIs without mass software failures.

Still, MS won't break compatibility until it is absolutly necessary. But if an algorythm can be replaced by a faster one, it has not that big impact if it doesn't behave as the older one. The developers can than (if they want) update to the new version and check if still everything is running, or they can just use the old state.

Software API functions change a lot, particularly on consoles during the early years where optimisations are being rolled out after real world usage can be analysed. What developers need is that the output doesn't change (unless a function is expanded) from the input and that the operation of the function doesn't have a resource impact elsewhere, i.e. shifting a function from using CPU resources to compute, could break things.

 

[Betanumerical]

I was trying to say that Sony said same thing even this gen, there is no need to comparing XB1 to last gen arguments about PS3. XB1 is much easier to develop compered to XB360. XB1 has the same API as Windows and it shouldn't be hard to develop. Go and read CD PROJECT interview.



http://www.lazygamer.net/general-ne...ill-get-more-powerful-is-easier-to-work-with/

But right now it's harder to master.

There's a world of difference between talking about how you modified a PC architecture to make it better and talking about how your console is super computer like, when it doesn't even bring any advantages or not many compared to a conventional solution.

The messages are very different and it seems to be pushed by multiple people (Albert Penello, etc) saying how complicated and hard to use their architecture is and how we should wait for the next big thing to bring the performance in line with the competitor.

 

[Shifty Geezer]

but you can introduce new features faster...

There's a lot of assumptions going on in this conversation, which doesn't even strike me as on topic. We don't have details of PS4's APIs or software implementations to make comparisons. It's probably best if we don't try to draw comparisons based on missing details.

There's a world of difference between talking about how you modified a PC architecture to make it better and talking about how your console is super computer like, when it doesn't even bring any advantages or not many compared to a conventional solution.

The messages are very different and it seems to be pushed by multiple people (Albert Penello, etc) saying how complicated and hard to use their architecture is and how we should wait for the next big thing to bring the performance in line with the competitor.

All PR works the same : If it's difficult, it has hidden power; if it's weak or easy, it's easily accessible and developer friendly. We shouldn't pay any attention to PR spin applied to the details. We can simply deal with interpreting the details ourselves.

 

[mosen]

There's a world of difference between talking about how you modified a PC architecture to make it better and talking about how your console is super computer like, when it doesn't even bring any advantages or not many compared to a conventional solution.

The messages are very different and it seems to be pushed by multiple people (Albert Penello, etc) saying how complicated and hard to use their architecture is and how we should wait for the next big thing to bring the performance in line with the competitor.

What I know for sure is that the competition called their product a supercharged PC architecture and if their products is a "supercharged PC architecture" in your point of view then XB1 is a supercharged PC architecture, too. Actually XB1 has more modification than competition and conventional solutions on PC.

• XB1 has 30GB/s coherent bandwidth between CPU and GPU ( and other 15 additional engines).
• XB1 has 8GB of unified RAM and every data could be coherent between CPU and GPU on it.
• XB1 had eSRAM which could have some benefits for XB1 in long term.
• XB1 has SHAPE.
• XB1 has 4 DSPs which 2 of them are for speech recognition on Kinect and 1-2 years from now part of them could be freed for developers.
• XB1 has two graphics command processors which could help developers to have better usage of GPU.
• XB1 has 4 DMEs with higher bandwidth than normal DMAs in AMD GPUs, and 2 of them could be used for compression/decompression textures in different formats for use in CPU (or other usages).
• XB1 has 3 Display planes.
• XB1 has faster CPU than competition (but it has some overhead compared to competition right now).

But If you think it's a PR talk from both side then take it easy, all of PR talks are the same. There is no need to bring some strange comparisons to this thread.

 

[Brad Grenz]

It has more modifications in most cases because it required more modifications to keep performance from tanking. You might as well be bragging that one road has more bridges than another when that's only true because a more circuitous path was chosen that crossed more rivers. Unless you're writing a book about Madison County most people will probably tell you the more direct route with the higher speed limit is preferable.

 

[Davros]

[*]XB1 has SHAPE.
[*]XB1 has 4 DSPs

Are those 4 dsp's independant of shape ?

 

[Billy Idol]

I really hope neither of the consoles have a 'supercomputer' like architecture...because supercomputers are typically quite slow wrt to its relative performance...they are fast because they throw tons of slow processors on the problem and leave it to the scientist to make use of it.

So, imo, 'supercomputer' like is not a certificate of speed...and thus is cringe-making when reading such statements.

 

[mosen]

It has more modifications in most cases because it required more modifications to keep performance from tanking. You might as well be bragging that one road has more bridges than another when that's only true because a more circuitous path was chosen that crossed more rivers. Unless you're writing a book about Madison County most people will probably tell you the more direct route with the higher speed limit is preferable.

During operation, an external processor, reading and writing the shader core context state (which is required when saving and restoring GPU context state) can be severely limited (1) by the external processor's own read/write capabilities. (2) It can also be limited by the bandwidth between the external processor and the GPU. These two limitations can result in extremely long context switching times. This problem also exists for the other non-shader resource components associated with the GPU context state save and restore process.

http://www.google.com/patents/US20130135327

(1) XB1's CPU has 30GB/s bandwidth (to DDR3).
(2) XB1 has 30GB/s of coherent bandwidth between CPU and GPU.

They have chosen this modifications because of their benefits. This is true for every other modified part of the XB1.

Are those 4 dsp's independant of shape ?

Yes.

 

[Rangers]

There's a world of difference between talking about how you modified a PC architecture to make it better and talking about how your console is super computer like, when it doesn't even bring any advantages or not many compared to a conventional solution.

They just mean how the CPU/GPU are on the same die and interconnected very fast compared to a PC. HSA type stuff.

Yeah it's kinda flimsy, but that's what they referring too.

 

[Brad Grenz]

http://www.google.com/patents/US20130135327

(1) XB1's CPU has 30GB/s bandwidth (to DDR3).
(2) XB1 has 30GB/s of coherent bandwidth between CPU and GPU.

They have chosen this modifications because of their benefits. This is true for every other modified part of the XB1.

And the PS4 has 20GB/s coherent bandwidth plus an additional 10GB/s that bypasses the cache to avoid stalls created by a cache flush needed to maintain coherency. The PS4 solution is more customized in this case, not less.

Obviously the customizations in both designs have intended benefits, the Xbox One just had more problems to solve based on strategic decisions that had been made.

 

[dobwal]

And the PS4 has 20GB/s coherent bandwidth plus an additional 10GB/s that bypasses the cache to avoid stalls created by a cache flush needed to maintain coherency. The PS4 solution is more customized in this case, not less.

Obviously the customizations in both designs have intended benefits, the Xbox One just had more problems to solve based on strategic decisions that had been made.

How is it more customized when it's the same design found in Kaveri? Onion and onion+ share the same 10 GBs (20 GBs total) in both directions.

All we know is that MS favored larger amounts of on chip memory than larger bandwidth to off chip memory.

What's so special about gddr5 that its easier to source customized x86 chips than it is to source chips large enough to handle a 8 GB design?

 

[Brad Grenz]

The work on PS4 is what lead directly to Onion+ being included in Kaveri.

 

[shredenvain]

Both consoles Socs are unique and have no equals in the consumer Apu market.
Both have been customized by the system architects and Amd.
I think both are great designs and both have great future potential because they are Socs alone.
I dont understand why people insist that any custom aspects of the X1 are there to assist because the system has a flawed design. What is the flaw exactly? The use of esram? Having less alu than the competition? Im pretty sure that the Xbox One would have esram even if they decided to use gddr5 for the main memory. After all last gen they had fast graphics ram for main memory and still included edram. As far as the gpu alu goes no 2 consoles have ever had equal gpus to my knowledge. Just because one system has more alu doesnt mean the other is flawed. Both the X1 and Ps4 have been well designed with each teams goals in mind. The both have less gpu power than we have come to expect from new consoles. They are also closer in design than any 2 systems before. Still even though they are closer than ever it seems the differences between the 2 have become a bigger deal than ever to alot of people.

 

[Brad Grenz]

I didn't use the word "flaw" to describe anything. But it's a fact that the Xbox One APU has audio customizations intended to handle Kinect voice commands. It's a fact that it has more DMA Engines because the multiple RAM pools require moving data around more often. It needed 3 display panes to accommodate the multitasking concept of the OS. The number of modifications isn't the measure of a design. The purpose and capabilities created by those modifications are what is relevant. One design can be both more customized and less capable at the same time.

 

[mosen]

And the PS4 has 20GB/s coherent bandwidth plus an additional 10GB/s that bypasses the cache to avoid stalls created by a cache flush needed to maintain coherency. The PS4 solution is more customized in this case, not less.

Obviously the customizations in both designs have intended benefits, the Xbox One just had more problems to solve based on strategic decisions that had been made.

ON PS4 Onion+ bus which shares Onion's 10GB/s and bypasses the GPU caches, is coherent. Onion (20GB/s, 10GB/s read and 10GB/s write) Which snoops CPU L2/L1 caches isn't coherent. Onion+ and Onion run over the same I/O controller so they can't access it at the same time. Also CPU bandwidth is 20GB/s on PS4.

On XB1, the GPU has a coherent read/write path (30GB/s) to the CPU’s L2 caches and to DRAM. The CPU requests do not probe any other non-CPU clients, even if the clients (GPU for example) have caches. Coherent read-bandwidth of the GPU is limited to 30 GB/s when there is a cache miss, and it’s limited to 10–15 GB/s when there is a hit. CPU bandwidth is 30GB/s on XB1, they beefed up CPU bandwidth.

The PS4 solution is exactly like the Kaveri.
http://forum.beyond3d.com/showthread.php?t=64406
http://share.csdn.net/uploads/5232b691522ba/5232b691522ba.pdf
http://pc.watch.impress.co.jp/img/pcw/docs/632/794/html/06.jpg.html
http://pc.watch.impress.co.jp/docs/column/kaigai/20140129_632794.html

I didn't use the word "flaw" to describe anything. But it's a fact that the Xbox One APU has audio customizations intended to handle Kinect voice commands. It's a fact that it has more DMA Engines because the multiple RAM pools require moving data around more often. It needed 3 display panes to accommodate the multitasking concept of the OS. The number of modifications isn't the measure of a design. The purpose and capabilities created by those modifications are what is relevant. One design can be both more customized and less capable at the same time.

You don't need 2 more DMA for data transfer (AMD PC GPUs use 2 of them) and two of them are useful for texture streaming. Two of display planes are for games and one of them is for system (PS4 has one for game and one for system). Kinect uses most of audio block right now but in future they could free up part of the resources (DSPs) for games (SHAPE is accessible by games).

 

[Betanumerical]

ON PS4 Onion+ bus which shares Onion's 10GB/s and bypasses the GPU caches, is coherent. Onion (20GB/s, 10GB/s read and 10GB/s write) Which snoops CPU L2/L1 caches isn't coherent. Onion+ and Onion run over the same I/O controller so they can't access it at the same time. Also CPU bandwidth is 20GB/s on PS4.

On XB1, the GPU has a coherent read/write path (30GB/s) to the CPU’s L2 caches and to DRAM. The CPU requests do not probe any other non-CPU clients, even if the clients (GPU for example) have caches. Coherent read-bandwidth of the GPU is limited to 30 GB/s when there is a cache miss, and it’s limited to 10–15 GB/s when there is a hit. CPU bandwidth is 30GB/s on XB1, they beefed up CPU bandwidth.

The PS4 solution is exactly like the Kaveri.
http://forum.beyond3d.com/showthread.php?t=64406
http://share.csdn.net/uploads/5232b691522ba/5232b691522ba.pdf
http://pc.watch.impress.co.jp/img/pcw/docs/632/794/html/06.jpg.html
http://pc.watch.impress.co.jp/docs/column/kaigai/20140129_632794.html

You seem to have missed this.

The work on PS4 is what lead directly to Onion+ being included in Kaveri.

 

[mosen]

You seem to have missed this.

How should I know that it's true?!

 

[liquidboy]

I didn't use the word "flaw" to describe anything. But it's a fact that the Xbox One APU has audio customizations intended to handle Kinect voice commands. It's a fact that it has more DMA Engines because the multiple RAM pools require moving data around more often. It needed 3 display panes to accommodate the multitasking concept of the OS. The number of modifications isn't the measure of a design. The purpose and capabilities created by those modifications are what is relevant. One design can be both more customized and less capable at the same time.

"fixed-function overlays hardware" which enables this GPU multi-plane-overlays was introduced in Dx11.2 to accommodate a trend in future GPU's having this fixed-function HW..

XB1 and its design incorporates this and yes it delivers the smooth switching between app/game/system . BUT there are also QUALCOMM SoCs in dev machines that have GPU-OVERLAYS support and with Dx11.2 we can do similar things as what XB1 can do with D2D/D3D swapchain switching..

Was this gpu-overlays feature introduced into Dx11.2 because of XB1 or of future trends ?! The \\build2013\ presentations lean towards the later but who really knows .....

 

[Shifty Geezer]

I think Brad's point is that XB1 has three planes instead of two because of the OS overlays. The DX feature you point to talks of two planes and native 2D UIs over 3D renders, requiring only 2 planes. XB1's customisation added a third plane for the OS docked app feature.

Although I've no idea where this conversation is headed now or what it's trying to achieve. People are making platform comparisons without a clear indication (to me at least) as to what those comparisons are trying to show. Is there a question being answered here?