AMD: Navi Speculation, Rumours and Discussion [2019-2020]

Status
Not open for further replies.
The savings wouldn't be also exactly 45% as you'd need to re-add the SerDes with enough bandwidth to the IO die, which would be significantly wider than what we see in Zen2 to be able to feed enough bandwidth.
That depends on the packaging tech. If they use 2.5D TSV, SerDes is likely unnecessary since it can be dense enough to carry all the links as if they are on-chip 32B/64B data buses. Though having a read at the Zen 2 ISSCC presentation, there seem to be an argument against interposers (?).

Zen 2 CCD is 74 mm2, with the IFOP taking ~8% of the space, meaning ~6 mm^2. Each gives 32B read + 16B write per clock. Assuming one IFOP per two GDDR6 channels, that'd gives 8 IFOPs for ~48 mm^2, providing 384-768 GB/s aggregate at 1-2 GHz.

Seem like a considerable saving e.g. for a hypothetical 2X Navi 10 product, shaving off perhaps >100 mm2 in the compute die. It seems to be even more relevant to CDNA products, which are touted to support advanced interconnect topologies (= more SerDes).
 
Last edited:
I remember this thread from three years before where I mentioned that the power considerations are mostly a direct result of the clocks AMD and nvidia can achieve with their respective chips.

It's the clockspeeds, stupid.

I think it's pretty obvious that the biggest reason for the difference is the clockspeed disparity. Running cards closer to their limits isn't new for AMD, but nvidia have opened up such a gap with Pascal that AMD fell way behind on performance to make up for it.

https://forum.beyond3d.com/posts/2015816/

AMD seem to be the one coming out on top now with clocks, just how much better they can do would decide how competitive they are at high-end because I still don't see them doing gigantic chips. Big Navi at 2GHz should've 5700 level of perf/W if not better. I also doubt high-end users would be bothered with going over 300W.

I wonder when it's arriving though, getting really tired of my ref. Vega blowing hard with low framerates in recent games.
 

Komachi and an Asian supply chain source seem to be teasing Navi 3X being a multi-die design, splitting into GCD (presumably Graphics Compute[?] Die) and MCD.

:yes:

If I were to guess, controllers and PHYs for display, memory and DMA would get spun off as MCD, and get manufactured a different process. Memory-side L2 stays on the monolithic GCD die, so as the gigantic L1-L2 interconnect.

Then the interconnect between the two dies would need only as much bandwidth as the GPU memory pool can deliver (maybe except when multi-GPU is enabled), which is totally viable in today's TSV packaging tech (equivalent to HBM class bandwidth).
Woohoo we're back to square one from over four years ago:

https://forum.beyond3d.com/threads/amd-navi-speculation-rumours-and-discussion-2017-2018.57684/

And you were there too :)
 
I don't think my Fury X will make it to 2022.

I'm expecting to be massively disappointed by this year's cards. Ah well, I guess the Steam library can wait and World of Tanks will keep me very happy :) So, let's see.

I can still do a lot of damage to my Steam library without an upgrade but it’s been 4 years since I’ve bought new toys. Need to scratch the itch.
 
Woohoo we're back to square one from over four years ago:

https://forum.beyond3d.com/threads/amd-navi-speculation-rumours-and-discussion-2017-2018.57684/

And you were there too :)
Hmm, it is fun to be proven wrong. :p

In RDNA 1 they have reduced the L1-L2 fabric complexity from a 64/16 to 5/16. I had wondered why they even bother to have 16 servers for only 5 clients in the design. Now in retrospect, it looks like an incremental change over Vega towards this, given teasers like "X3D stacking".

Imagine a possibility of L1 misses going straight to IFOPs, which are connected to chip stacks in the same package with beefy L2 caches and HBM stacks. Can't wait to see how the split would look like.
:runaway:
 
Last edited:
It's just notable for the fact that this is the first official public mention of an RDNA 4 chip.

It makes sense. RDNA 2 architecture, the engineering for it has long since been completed (in engineering time),

AMD has at least 2 separate large GPU architecture design teams. one would be working on RDNA 3 architecture, the other would be working on RDNA 4 architecture.
 
It's just notable for the fact that this is the first official public mention of an RDNA 4 chip.

It makes sense. RDNA 2 architecture, the engineering for it has long since been completed (in engineering time),

AMD has at least 2 separate large GPU architecture design teams. one would be working on RDNA 3 architecture, the other would be working on RDNA 4 architecture.
You forgot that they also need to develop CDNA, CDNA2 etc etc
 
You forgot that they also need to develop CDNA, CDNA2 etc etc

CDNA 1 was probably completed a long time ago. That was Arcturus. It's launching in the era of RDNA 2

RDNA 3, like CDNA 2, is well into development, probably close to being completed as far as silicon engineering, same with Zen 4 CPU core.

RDNA 4 is in design, just like Zen 5 is in design. These are for products hitting the market in H2 2023 and through 2024. So of course, in mid 2020, these architectures are in development now given AMD has multiple design teams for both CPUs and GPUs.


p.s. forget I ever mentioned an AMD Ray tracing co-processor bolted onto the CPU die. That was just an example of what might be possible in the 2023/24 time frame. Of course that won't happen.

I myself highly doubt there will be a need for a PS5 Pro in 2023 or 2024. Nor an upgrade of Anaconda on the Xbox side.

I'd rather see Sony and Microsoft shift hardware generations every 6 years instead of every 7.

I'd rather we get PS6 and Anaconda + Lockhart successors in 2026.
By late 2027, the 4TF Lockhart is no longer a development constraint, its successor is 20TF while PS6 is 40TF and high-end Xbox (Anaconda successor) is 50TF
 
CDNA 1 was probably completed a long time ago.
Kinda but not really.
RDNA 3, like CDNA 2, is well into development, probably close to being completed as far as silicon engineering, same with Zen 4 CPU core.
RDNA3 bits got taped out, and I think it's MI200 then Genoa, soon(tm).
These are for products hitting the market in H2 2023 and through 2024.
Way earlier.
By late 2027, the 4TF Lockhart is no longer a development constraint, its successor is 20TF while PS6 is 40TF and high-end Xbox (Anaconda successor) is 50TF
That's a lot of optimism.
 
Status
Not open for further replies.
Back
Top