Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]

Status
Not open for further replies.
Efficiency. If you cut round dies out of a wafer, you lose all the silicon in the spaces between them. Rectangular dies are easy to cut and have no waste. The only other sensible geometry would be triangles (or hexagons) which could be cut with no waste, but that's an unnatural space to work in - functional units in a chip are always arranged in rectangles and rectilinear groups.
Imma need to see the area integration optimization equations

Show your work.

Worth 50% of your final grade.

:3
 
While the Zen cores maybe of similar size to Jaguar, they do utilize an L3 cache that will take up a decent size area.

Looking at core size is misleading, IMO. The Jaguars in PS4/Xbox One are organized in two quad core complexes. Each quad core complex, functionally equivalent to a CCX in the Ryzen CPUs, is roughly 26mm2, that includes the shared L2 and system glue.

A CCX is 44mm² in 14nm tech, it might be 35mm² in 10nm and 25mm² in 7nm, either way it seems like a clear upgrade path (to me anyway).

Cheers
 
Efficiency. If you cut round dies out of a wafer, you lose all the silicon in the spaces between them. Rectangular dies are easy to cut and have no waste.

You could get even better cost efficiency from circular dies using wafers large enough for the die space. If you're using ICs made from exotic (read, expensive) materials, this the only way to keep costs manageable. It doesn't scale well though. :nope: Nobody is setup to mass produce this way. Yet. :runaway:
 
Isn't it just simpler to create a rectangle since the transistors structure are linear by design. Plus... ordered grid design for repeated structures (memory etc.). Shortest distance between two parallel lines etc. (cue my t-shirt, which I will not be posting a selfie of).


Physically curved structures might not be so great during the patterning steps.
 
Last edited:
Looking at core size is misleading, IMO. The Jaguars in PS4/Xbox One are organized in two quad core complexes. Each quad core complex, functionally equivalent to a CCX in the Ryzen CPUs, is roughly 26mm2, that includes the shared L2 and system glue.

A CCX is 44mm² in 14nm tech, it might be 35mm² in 10nm and 25mm² in 7nm, either way it seems like a clear upgrade path (to me anyway).

Cheers

26mm2 on PS4/Xbox One?...is that the size on original 28nm consoles or slim 16nm consoles?..
 
The discussion was about launching on GF/TSMC "7nm".

right but my point was Zen will take up much more of die area than Jaguar on same node @ ~350mm2 die size.....so I think 8 core zen cpu in next gen consoles is out of the question..
 
right but my point was Zen will take up much more of die area than Jaguar on same node @ ~350mm2 die size.....so I think 8 core zen cpu in next gen consoles is out of the question..

Well, sure it will. The Zen CPU has significantly more transistors. But that's not relevant. The area available to the CPU on the die on the 28nm launch PS4 APU is what we're interested in. How small or big Jaguar is at smaller nodes is irrelevant because Sony won't be considering Jaguars for PS5. The issue is whether the same die area on a 7nm node will accommodate a 2x CCX Zen CPU, that accomodated an 8c two-module Jaguar CPU at 28nm.
 
Last edited:
I think the point being made is that an 8c Zen on a 7nm process node would be a similar size to what the 8 core Jag was at 28nm (i.e. when the PS4 launched). That's what is important, i.e. how many Zen cores can we fit in the same die area as the equivalent CPU on the OG PS4 @ 28nm (launch).

ahhh my mistake...I was only thinking relative size of cpu/gpu on die....and not taking into account the fact that the overall chip size (~350mm2) would still be the same.
 
Isn't it just simpler to create a rectangle since the transistors structure are linear by design. Plus... ordered grid design for repeated structures (memory etc.). Shortest distance between two parallel lines etc. (cue my t-shirt, which I will not be posting a selfie of).


Physically curved structures might not be so great during the patterning steps.
The only reason I brought up circular was for chip layout design. Everything is equidistant from the middle. Wasn't sure if there was some optimal chip layout that would ultimately reduce travel distance.
 
Both mid-gen chips are north of 300mm^2.

They'd need a sufficiently wide memory bus regardless of memory type anyway.

---

You could say they wouldn't be designing a PS4U. :oops:
 
Last edited:
Isn't it just simpler to create a rectangle since the transistors structure are linear by design. Plus... ordered grid design for repeated structures (memory etc.). Shortest distance between two parallel lines etc. (cue my t-shirt, which I will not be posting a selfie of). Physically curved structures might not be so great during the patterning steps.

It is because all design methodologies are predicted on 90 degree right angles. But.. who knows what could be achieved if those preconceived notions were tossed out of the window. Take the red pill!
 
It is because all design methodologies are predicted on 90 degree right angles. But.. who knows what could be achieved if those preconceived notions were tossed out of the window. Take the red pill!

Non-Manhattan routing is long-standing topic of research, since there are advantages in terms of wire length and the number of metal layers necessary for a diagonal connection versus right angle turns.
Routing methods that allow for increasing options for diagonals such as Y and X architectures exist.

It's not that they haven't been considered, but they seem to pose their own challenges given how long Manhattan routing has persisted despite the readily offered decrease in wire length and congestion.
 
26mm2 on PS4/Xbox One?...is that the size on original 28nm consoles or slim 16nm consoles?..
The original consoles. Die size is dictated by economics, bus width likewise to an extent. The bus width, dictating bandwidth, sets an upper limit to the amount of CUs, so all in all I'd expect a 300-350mm^2 die, 50-60 of which used for cpus, 256 bit bus (unless AMD's high bandwidth cache tech really shines).

Cheers
 
Status
Not open for further replies.
Back
Top