Xbox One (Durango) Technical hardware investigation
- Thread starter Love_In_Rio
- Start date
- Status
Michellstar
Regular
of course i dont know, but i head in the post conference 200 GB/sec for sure
Well, i couldn´t watch the "hardware panel" on twich tv, anyway hoping an in depth hard reveal
french toast
Veteran
They couldn't have upped the gpu clocks, it has been implied that the 1.2 tf leak was correct.
Also it's been confirmed it's 32mb of 'sram', so no doubling up to achieve higher bandwidth either.
As someone else pointed out... Creative math in play.
We should be good grateful they didn't use creative math on the computing power either, adding the server farm power up and saying xbox one has 5tflops of computing power
Also it's been confirmed it's 32mb of 'sram', so no doubling up to achieve higher bandwidth either.
As someone else pointed out... Creative math in play.
We should be good grateful they didn't use creative math on the computing power either, adding the server farm power up and saying xbox one has 5tflops of computing power
of course i dont know, but i head in the post conference 200 GB/sec for sure
the 30 GB from cpu-gpu interconnection is a speculation from bkillian, we don't know how it work but they said "more than 200 GB/s system bus"
eSram is not for the system but it's only connected to gpu, so we don't know what this means in reality, here you can read only hypothesis
implied where? i've seen people saying a lot of things are implied in vids, that when you actually watch are nowhere near that clear cut. you're going to definitely need to prove that to me with a source.
you're probably just once again referencing the 768 operations thing which obviously has nothing to say at all about clocks.
me and xenios are keeping hope alive
We should be good grateful they didn't use creative math on the computing power either, adding the server farm power up and saying xbox one has 5tflops of computing power
you meant petaflops
Betanumerical
Veteran
system ram generally includes all the bandwidth for the GPU and CPU, so that in the Xbox One would include the eSRAM, DDR3 and the CPU <-> GPU interconnect. 200GB/s for that sounds about right, i think its actually 200.6GB/s.
HollovVpo1nt
Newcomer
I posted this in the distributed computing thread, but I think that at some point in time, Xbox One's total power was comprised of raw computational power + cloud. So Total Power = 1.2TF + Cloud. This would result into an always online system. The leaked VGLeaks documents did hint at an always online console. I think that due to recent backlash Microsoft had to get rid of that idea, leaving us with a 1.2TF base model.
What do you guys think?
What do you guys think?
It's two instructions per cycle, but each instruction can use memory operands, so you can have an AGU op, a memory op and an ALU op from one instruction, then you have SIMD (SSE/AVX), do those count as one op or four-at-a-time ?
The two instructions per cycle with memory ops is likely to be equivalent to the three instructions of the 360's CPU, but the OOOe machinery and improved memory pipeline is going to achive much higher performance.
Cheers
It's only been speculated as implied (which isn't much of a foundation to build on, admittedly); since they mentioned the 768 ops/cycle figure we know the 12CU part of the leak is correct. Hence, it may be speculated that if one part of the leak is correct then the other might be as well. Also, if clock was significantly upped, maybe they would have mentioned an actual performance metric (gigaflops) rather than just ops/cycle.
However instead they just toss out the ops/cycle number, presumably to try and impress those who aren't educated enough to relate that number to anything, like the 50% higher ops/cycle figure of PS4 for example.
So it's an inferrence based on an assumption. However it could reasonably be true, since MS holds any actual performance metrics so guardedly to their chest.
They might not have settled on final frequency yet. It all depends on how the initial silicon turns out. If >85% can operate at higher frequency without compromising stability and power constraints, they might up the frequency.
My money is on they won't
Cheers
admitting more or less 768 shaders lets people know they have less in that area than sony too, yet they did it.
but like i said if you tell me 12cu's all but confirmed i say true. it's the clocks i'm still holding out hope for as it's not confirmed yet.
anyways any upclock will still leave them <1.8tf, so they still probably wont be super keen to talk about it even if they upclocked to 1.4 or 1.6.
and the other upclock related thing is the mention of 200gb/s bw somewhere. you can say creative maths and it may be but we dont know. it does seem a little odd if vgleaks=official ms docs, and those docs said 170 gb/s not 200, that they'd change their accounting method in some other statement.
i'm not holding out great hope mind you, just hanging onto that sliver till it's officially quashed.
Adding up all bandwidth for all buses when listing system bandwidth isn't meaningful or useful. For example on a sandybridge-E based PC you have 4-channel DDR main RAM, something like 40 lanes of PCIe 3.0, dual QPI links and three levels of cache and on-chip ring bus for multiple CPU cores. Adding up all of that and I would expect total system bandwidth to hit well over a TB/s total aggregate. That's impressive as a paper spec, but that's about all it is.
You could tack on on-chip buses in PS4s APU to the 176GB of unified memory as well, there's the onion+ (garlic? whatever) interconnect for example which Cerny claims is "almost" 20GB/s in that garmasutra interview he did some weeks back. I don't consider that valid though. There are lots of on-chip buses and interconnects on chips like a GPU or APU, and just arbitrarily adding some of them to the total system bandwidth pool is not legitimate IMO. It stinks of PR department paper speccing, kind of like back in the atari jaguar days when it was claimed that the machine could "render" over 100 million pixels/s. Yeah, if you used the blitter to copy one-bit-per-pixel bitmaps to one memory location to another and used no system bandwidth for anything else (including throwing up an image on a TV...) you could theoretically do that. But in reality it was just lies, of course.
french toast
Veteran
Lol once the shaders had more or less been confirmed, then I'm gonna go along with vg leaks on this-800 MHz gpu, 1.6ghz cpu.
The bandwidth is just one of many questions I have put down in the other reveal thread.. feel free to have a crack at them :>
So the die is really on the large side. The RAM chips surrounding it should be housed in standard 9x14mm BGA packages. Compared to them, the die looks to be around 19x23 mm (I just counted the pixels in Photoshop)
Technicals details here:
http://www.anandtech.com/show/6972/xbox-one-hardware-compared-to-playstation-4
http://www.anandtech.com/show/6972/xbox-one-hardware-compared-to-playstation-4
I get an area of 2.5-2.75 times the RAM packages. That's an upper bound around ~330 mm².
Cheers
dagnastyep
Newcomer
Regarding the CPU:
http://www.engadget.com/2013/05/21/xbox-one-architecture-panel-liveblog/
A normal Jaguar can do 4 ops per core per cycle?
http://www.engadget.com/2013/05/21/xbox-one-architecture-panel-liveblog/
A normal Jaguar can do 4 ops per core per cycle?
- Status
Similar threads
