Next Gen GPU architecture GCN, RDNA, Navi 10/20 (PS5 Navi Hybrid, Xbox Navi Pure) *spawn*

Status
Not open for further replies.
CPU offloading, yep.
GPU tasks? Only thing I could think that could come close to that is when kinect was unbundled and there was hope that some of the functions would be opened up to devs. Can't remember if some could be used for things like path tracing etc.
Do remember was bit more flexible than it was be used for, but wasn't exposed to devs. But nothing like bridging power gap or anything like that.
 
I am really looking forward to the improved audio of next gen. If it doesn't improve I'll never forgive Cerny :) his gotten my hopes up.
 
I do recall the last bit.
Definitely a lot of discussion of CPU offload. GPU offload don't recall so much, perhaps more discussion around move engines than SHAPE doing rendering loads.

OK, the move engines I do remember getting some (undeserved) hype.
 
Mostly threads between Gaf and IGN during this time. Even a few members (not naming no names) here spreading misinformation. Point being, SHAPE original intent was for audio and kinect purposes, not a last minute ditch effort on improving XB1 overall TF performance.

giphy.gif
 
I am really looking forward to the improved audio of next gen. If it doesn't improve I'll never forgive Cerny :) his gotten my hopes up.
We'll definitely get better.
I reckon with things like the following vid and good object path traced audio combined we will finally get good 3D audio.
 
Reality check?
From what we’ve seen so far, the clock speed of a Navi-equivalent GPU won’t be a massive amount faster than existing console hardware, but even a difference of, say, 10% to 15% in GPU compute power could mean a lot of smaller improvements across the board that lead to better-looking games. It would bring the console much closer to the heftier PCs we actually use to develop on, or the high-end gaming PCs that are reserved for gamers who invest thousands of dollars in their hobby.
 
Can’t be just 15% faster, that difference would be just in 7nm let alone all the other improvements over a six year old gpu.
 
But those people of the land were (are) here as well. Let's not kid ourselves...
I don't recall any talk of Shape being used for GPU acceleration, or even much acceleration at all. It was discussed repurposing it for audio as opposed to Kinect. We knew the hardware - Tensilica DSPs - and things like move engines were on the whole recognised as relabelled AMD GPU parts. I recall some hoping to believe there was more special about them, but not any talk about graphics workloads on the DSPs; not even ray tracing (which they wouldn't be suited for).
 
I don't recall any talk of Shape being used for GPU acceleration, or even much acceleration at all. It was discussed repurposing it for audio as opposed to Kinect. We knew the hardware - Tensilica DSPs - and things like move engines were on the whole recognised as relabelled AMD GPU parts. I recall some hoping to believe there was more special about them, but not any talk about graphics workloads on the DSPs; not even ray tracing (which they wouldn't be suited for).

When I stated "But those people of the land were (are) here as well. Let's not kid ourselves...," was in prior reference to those B3D members showing up on other boards misrepresenting certain information on SHAPE (move as well).
 
I don't recall any talk of Shape being used for GPU acceleration, or even much acceleration at all. It was discussed repurposing it for audio as opposed to Kinect. We knew the hardware - Tensilica DSPs - and things like move engines were on the whole recognised as relabelled AMD GPU parts. I recall some hoping to believe there was more special about them, but not any talk about graphics workloads on the DSPs; not even ray tracing (which they wouldn't be suited for).
Can you remember what the talk was about in regards to possibly being able to do on them was?
Did they also model room using time of flight sensors or was that just for body?
Was nice having belkin (sp) about because I don't remember ever seeing anything about it doing magic on gpu, he seemed disappointed that it was locked out and never opened up to devs.
Those were the days....
 
Can you remember what the talk was about in regards to possibly being able to do on them was?
Audio. ;) Spatial audio in particular (HRTF).
Did they also model room using time of flight sensors or was that just for body?
It scanned the entire scene. MS provided libraries to track bodies. You could create a 3D room scan but I think that was done independently. 3D vision didn't require SHAPE though. Possibly SHAPE helped with denoising and filtering, but on the PC, Kinect vision was all CPU based. Although there was also processors in the Kinect itself that may have been doing all the signal processing, now I think about it, so I'm not sure what SHAPE's involvement was.
Was nice having belkin
bkilian
 
xbat said:
Possibly early dev kit? Otherwise 10 TFlop dream is dead.

The phrasing's pretty vague though:

"From what we’ve seen so far, the clock speed of a Navi-equivalent GPU won’t be a massive amount faster than existing console hardware, but even a difference of, say, 10% to 15% in GPU compute power could mean a lot of smaller improvements across the board that lead to better-looking games."

He only refers to the clockspeed of Navi not being much faster than existing console hardware, and then goes on to state 10-15% would be enough to bring plenty of smaller improvements.

The XBoxOneX GPU is 1172MHz. If the PS5/XBoxNext GPU's are around 1400MHz, the "clock speed... won’t be a massive amount faster than existing console hardware." We could see 72CU's and 12.9TF and the statement would still hold true.

As for the 10-15% figure, I'm curious what that means. I think one possibility is that he's just throwing a number out there. But if I put on my tinfoil hat, I wonder if he's treading carefully because that's roughly the performance gap between the XBoxOneX and Lockhart? Because the only aspect of the Lockhart rumour that I don't buy is the notion of it being less powerful than the XBoxOneX.

Is a 40CU GPU clocked at 1172MHz really going to be that expensive to surpass in 2020? Even if Lockhart and Anaconda are created via binning, there's scope for something like a 44/64CU and 1200/1500MHz difference between them. In that case, even though it would play second fiddle to Anaconda's 12.29TF, Lockhart's 6.76TF would still slap around the XBoxOneX.
 
Status
Not open for further replies.
Back
Top