Xbox One (Durango) Technical hardware investigation

Status
Not open for further replies.
We have the numbers for SHAPE from the guy who worked on it, so we should easily know what kind of power would be required to replicate, that power easily exists in 8 core CPU's.

You dont have "tens of times more" which is what he said.

Also what are the shape "numbers"?? I must have missed those.

I want to say there was something about ~200 gflops but dont quote me on that.
 
You dont have "tens of times more" which is what he said.

Also what are the shape "numbers"?? I must have missed those.

I want to say there was something about ~200 gflops but dont quote me on that.

bkillian said you would need 500GFLOPs of power to achieve something similar.

So its probably, 50GFLOPS then?.
 
I'm looking into the dev kit thing, it's complicated - for one I don't even know how many dev kits there are:

I thought these were all of them:
1. alpha kits with HD7970s and 16 CPU cores, 12 GB of RAM - no ESRAM, move engines, SHAPE etc
2. beta kits - went out at the end of last year, have all the custom components like ESRAM and SHAPE etc, the ones with tiger camoflage and were very noisy?
3. final kits, went out July 2013

Then there's the question of whether final retail hardware has 12 GB of RAM or final kits based on retail silicon have 12 GB of RAM (and final retail hardware has only 8GB).

Okay but if it's been assumed that the retail kit is no different then what the dev kit possesses (indie devs-no limitations), then why would there be any final hardware kit at all? Why not just give devs the retail model and be done with it?
 
Okay but if it's been assumed that the retail kit is no different then what the dev kit possesses (indie devs-no limitations), then why would there be any final hardware kit at all? Why not just give devs the retail model and be done with it?

Good point maybe they have done that. Do dev kits generally have added ports or connectors?
 
If anyone seriously thinks that shape requires 5 TFLOPS to emulate, you are quite insane.

You cant emulate a PS3 or a 360 with any amount of flops. So I guess that'd be wrong.

I find it believable a 200 gflop specialized piece of fixed function hardware might be very difficult to emulate...
 
You cant emulate a PS3 or a 360 with any amount of flops. So I guess that'd be wrong.

I find it believable a 200 gflop specialized piece of fixed function hardware might be very difficult to emulate...

Thats a false equivalency but I will say it might be hard to emulate, but it depends at what level of accuracy you want. The difference between a single ASIC and a entire system is worlds apart.
 
The alpha kits had an extra 8 CPU cores to emulate some of the SHAPE block functions.
That wasn't enough though so they abandoned their emulation plans.

Forget the FLOPS, why would dev bother to "emulate" SHAPE functions back in early 2012? At that point you would think it would be basic graphics engine work, not secondary stuff like Kinect and non-standard audio, stuff that the SDK was probably without or in early phases of development.

Besides the GCN technology in the 7970, the rest of the alpha dev kit specs have little to do with the final XB1 hardware. Maybe bkilian can put this to rest?
 
Forget the FLOPS, why would dev bother to "emulate" SHAPE functions back in early 2012? At that point you would think it would be basic graphics engine work, not secondary stuff like Kinect and non-standard audio, stuff that the SDK was probably without or in early phases of development.

Besides the GCN technology in the 7970, the rest of the alpha dev kit specs have little to do with the final XB1 hardware. Maybe bkilian can put this to rest?

SHAPE has nothing to do with Kinect or "non-standard" audio. It does all of the basic audio processing you'd do in a game. I imagine they just stuffed extra CPU power into the dev kits because in the retail system the intention is to have audio offloaded from the CPU as much as possible.

The audio block has 4 processors. SHAPE is just one of them. Two of them seem to be very Kinect/system oriented and may not be useable by devs for much, if anything. One of them seems to just seems to do process management and maybe DMA. SHAPE is the last one that looks to be what devs will use to offload audio work from the CPU. It is leveraged using the same API that was used for audio on the 360. Using it should be transparent.
 
bkillian said you would need 500GFLOPs of power to achieve something similar.

So its probably, 50GFLOPS then?.
I said nothing of the sort. I said that if you used the entire 360 CPU, it would not be able to match what the fixed function audio hardware can do. That's pure output only, not emulating the hardware. You might be able to match it in numbers of channels and effects, but your effects would be lower quality.

And the alpha kits did not have intel CPUs in them.
 
SHAPE .......... does all of the basic audio processing you'd do in a game. It is leveraged using the same API that was used for audio on the 360. Using it should be transparent.

Very interesting, since I'm completely unable to find any usable information of the capability of the audio engine. It seems you have some information, so it would be appreciated if you could enlighten me regarding the following :

* Which fixed functions are available?
- 1 (and 2) pole filters? (what about the parametrization/transform?)
- Topology of the available filters (only direct form biquad)?
- Convolution, either in time domain or frequency domain?
* How much delay memory is available (for environmetal simulation)?
* Is it only a fixed function chip? If not - which instructions are available?
* SRC : How many taps for the FIR?
 
Last edited by a moderator:
There's also the possibility that devkits have 8GB and that devs can access more than just 5GB.

Sounds tricky, it wouldn't make sense to make an open statement saying that all consoles can be devkits granting indies equal privileges, and then force them to work around for not implementing them enough memory for debugging their projects. That's like you're giving equal rights but at the same time you're giving someone a short end of the stick.

Before the indie community had full access to the hardware but didn't have access to proper devkits. Now, in the xb1's case I'm thinking a uniform setup for both indies and developers; hence probably why Microsoft stated that indies won't be limited on memory usage compared to developers, because all of them will be debugging on the same unit.

My two cents.
 
I am not following you.

Retail units have 8GB, it's written in XBO official site
If devkits too have 8Gb then indies won't be treated unfairly since they don't get less memory then AAA devs.
If devkits instead have 12Gb then it's unfair for indies that can "only" use the 8Gb of retail kits.

What's the problem for indies?
 
Very interesting, since I'm completely unable to find any usable information of the capability of the audio engine. It seems you have some information, so it would be appreciated if you could enlighten me regarding the following :

* Which fixed functions are available?
- 1 (and 2) pole filters? (what about the parametrization/transform?)
- Topology of the available filters (only direct form biquad)?
- Convolution, either in time domain or frequency domain?
* How much delay memory is available (for environmetal simulation)?
* Is it only a fixed function chip? If not - which instructions are available?
* SRC : How many taps for the FIR?

I'm not a dev and I don't have access to any kind of technical docs. All the information that's out there is what was revealed in this leak: http://www.vgleaks.com/durango-sound-of-tomorrow/ and the contributions bkilian made to the following thread http://forum.beyond3d.com/showthread.php?t=63677 I seem to remember you participating in that one.

My understanding is that it won't offload 100% of the audio processing, but that it will handle nearly all of the basic functions an audio engine would have to provide. The XAudio2 API or the Windows Audio Session API can be used, which implies if you work as you did before the audio block should handle most things transparently.
 
If devkits instead have 12Gb then it's unfair for indies that can "only" use the 8Gb of retail kits.

What's the problem for indies?

Developers use 12 Gb devkits and indies would use 8 gb devkits, that's the problem. They're all developing with the same hardware in mind but indies would have to work in further partitioned space, which commercial developers don't have to.

that wouldn't make any sense on a comment stating that indies wouldn't have any memory limits contrary to developers. in reality they would be undercut 4 gbs on running software tools.
 
Developers use 12 Gb devkits and indies would use 8 gb devkits, that's the problem. They're all developing with the same hardware in mind but indies would have to work in further partitioned space, which commercial developers don't have to.

that wouldn't make any sense on a comment stating that indies wouldn't have any memory limits contrary to developers. in reality they would be undercut 4 gbs on running software tools.

Unless its all 12 gigs.
 
Developers use 12 Gb devkits and indies would use 8 gb devkits, that's the problem. They're all developing with the same hardware in mind but indies would have to work in further partitioned space, which commercial developers don't have to.

that wouldn't make any sense on a comment stating that indies wouldn't have any memory limits contrary to developers. in reality they would be undercut 4 gbs on running software tools.

I don't get it. Indies are going to be mad that they didn't have to pay $10K for a dev kit and were able to use the $500 console they bought because they have less memory to work with? The bigger indies will buy regular dev kits anyway. The smaller indies that can only afford a $500 console are unlikely to fully tap the hardware anyway. This really seems like a non-issue, if they full dev kits even have more memory.
 
I don't get it. Indies are going to be mad that they didn't have to pay $10K for a dev kit and were able to use the $500 console they bought because they have less memory to work with?

Over statement, we're just talking about an additional 4 gbs of memory.

The bigger indies will buy regular dev kits anyway. The smaller indies that can only afford a $500 console are unlikely to fully tap the hardware anyway. This really seems like a non-issue, if they full dev kits even have more memory.

So what's the point in making the Xbox 1 a devkit if in order to take advantage of it you need to buy an other one? that's basically superseding the purpose of Microsoft assuring all indies.
 
Last edited by a moderator:
Status
Not open for further replies.
Back
Top