Xbox One (Durango) Technical hardware investigation

[Scott_Arm]

Question: Why do you need a "Data Move Engines" if all of the RAM is one unified pool on a single bus? If both the CPU and GPU have access to the full addressable memory space, why would you be moving data? Is this purely for moving data to the embedded ESRAM?

I'm guessing this was written by a non-english speaker.

 

[function]

Regarding 50GB optical disks, if a HDD is standard in every system then disk capacity doesn't actually matter.

 

[Rangers]

I thought someone on neogaf was talking about the Jag cores in durango/orbis having turbo up to 2.4ghz? Not sure though, but it sounded reasonable and it sure would be nice.

If so, this diagram doesnt mention it, maybe another clue of fake-ishness? Or not.

 

[DuckThor Evil]

I think it's legit, but we'll see...

The memory setup and 12 CUs don't really make me that happy. It's kind of funny that there was only 4 years between Xbox and 360 and 8 years between 360 and Durango and this is what we'll be getting?

I have to admit that my understanding of that memory setup in practical terms is limited, but it doesn't sound too impressive now does it? The rumoured Orbis seems significantly better to me.

What can the 32MB esram really do?

 

[french toast]

Yea the alleged 6 disk blueray multi changer also smacks of a 12 yrs wet dream...
Blitters...blueray multichangers..yea ok. :/

 

[Prophecy2k]

Question: Why do you need a "Data Move Engines" if all of the RAM is one unified pool on a single bus?

I'm guessing this was written by a non-english speaker.

Moving data between (fast) ESRAM and (slow) DDR3 main memory. I would assume it's there to provide speed up when having to tile given the low quantity of ESRAM (32MB).

It says nothing about the GPU archetecture outside of it being custom. Perhaps the "secret sauce" is in the GPU block (outside of the DSP & blitter).

 

[Scott_Arm]

Moving data between (fast) ESRAM and (slow) DDR3 main memory. I would assume it's there to provide speed up when having to tile given the low quantity of ESRAM (32MB).

I clued in just before I saw your post.

 

[Rangers]

i noticed our friend aegies commented on neogaf

I believe these are accurate, though, again, the extra hardware they don't explain will play a major part in what Durango does. I've also heard from reliable sources that the GPU CUs are super-efficient, with an aim for almost 100 percent usage at all times. The 360 hit around 60 percent, I understand. Like I've said previously, context is going to be really important for both systems.

 

[Bagel seed]

lherre seems to agree with what's been posted, more or less.

I think the DME will let tiling be easier to perform than it was on 360. That way the 32mb shouldn't be a limit as much as you'd think. With deferred engines seemingly being the norm nowadays they've probably made it in a way that multiple framebuffers won't be as troublesome to manage.

 

[DJ12]

Why would someone put hdmi input on it if they have just made it up.

 

[AlNom]

Question: Why do you need a "Data Move Engines" if all of the RAM is one unified pool on a single bus? If both the CPU and GPU have access to the full addressable memory space, why would you be moving data? Is this purely for moving data to the embedded ESRAM?

Maybe it has to do with data formats...

 

[xexuxjy]

Anyone want to comment on the chance of backwards compatibility with those specs.....?

 

[french toast]

i noticed our friend aegies commented on neogaf

Now that is more like it...efficiency is going to be the key....I hope this last detail is correct...could make the difference.

 

[anexanhume]

i noticed our friend aegies commented on neogaf

Yes, but I'd also like to hear a narrative about how Orbis can't have those same efficient CU's and the GDDR5 at 192 GB/s can't keep them fed. Kudos for efficiency, but they're still not being aggressive on TDP envelope.

 

[Rangers]

Yeah, lherre and thuway both are basically wink verifying it

http://www.neogaf.com/forum/showpost.php?p=46704759&postcount=501

I'm not sure if I have seen the diagram somewhere else ...

Seems real ladies.

 

[AlNom]

Now that is more like it...efficiency is going to be the key....I hope this last detail is correct...could make the difference.

Efficiency is coming down to the base GCN architecture.

 

[anexanhume]

Efficiency is coming down to the base GCN architecture.

Precisely. And there's no reason to think Orbis won't be using same GCN architecture. Now, is there something about 32MB ESRAM and 8GB DDR3 that can keep them fed all the time while 4GB GDDR5 can't, or is the secret sauce perhaps some sort of fancy scheduler?

 

[function]

Maybe it has to do with data formats...

Compress (and SCALE DAMMIT!) those render-to-textures as you copy out of embedded ram. Then you can laugh in the face of main memory bandwidth for infinite detail!

 

[french toast]

Efficiency is coming down to the base GCN architecture.

Do you think it could be a modified gcn 2?? After all microsoft would have access to the best shader technology amd could muster..and if that means less die size = more profit..

 

[MrFox]

Why would someone put hdmi input on it if they have just made it up.

DVR functionality or something?
I don't think an HDMI receiver is allowed to decrypt it and record it, that was the whole purpose of HDMI, after all. Maybe to record the output of personal devices, but that's equally crazy, they all have a data file to copy.

Maybe it allows to "chain" multiple boxes, so it can do what Sony did with GT5 on PS3, Only one box runs the game, but each box is rendering a part of the view. Having an HDMI IN would allow the box to combine them into a single output. That would satisfy hardcore gamers with money. That would make the "multiple SoC" and "Like a super computer" rumors much less crazy