AMD: RDNA 3 Speculation, Rumours and Discussion

Status
Not open for further replies.
Were the Radeon 5000 series selling for well above MSRP?
At the time in my region yes, but it's not as if i have a time machine to travel back in time and tell my self to take printscreens and post it here as evidence, so take my word as is, don't to believe it? Fine by me, don't. Im not here to try to drag this into useless prolongued arguments with examples of both cases to show both sides were true on some different corner to the planet at some point in time.
Can't tell if AMD behaves by wait and see being optimistic to see if the final consumer somehow graces them with increased interest on their products but seeing they do nothing until their revenue dropped by 40% to wake up and call out to AIB's to cut their msrp just proves too litle too late but anyway, i won't waste more of my time on discussing my views, won't change what's past or what's to come.
 
On the topic of sorting there was this all the back on page 4 wonder if it is part of the RT improvements rumored. :unsure:
COMPUTE UNIT SORTING FOR REDUCED DIVERGENCE - Advanced Micro Devices, Inc. (freepatentsonline.com)
This has been floating around for a while. I can't work out from this whether there's any meaningful gain possible in a physical device.

It feels like I've been talking about this for 10 years:

AMD: R7xx Speculation

Conditional Routing was floating around back in 2004. Honestly, I don't feel motivated to compare this document and the Conditional Routing paper. I'll pay more attention when I see it in hardware.


The question around dedicated matrix hardware was touched on in the June tomshardware interview.
We asked whether AMD would include some form of tensor core or matrix core in the architecture, similar to what both Nvidia and Intel are doing with their GPUs. He responded that the split between RDNA and CDNA means stuffing a bunch of specialized matrix cores into consumer graphics products really isn't necessary for the target market, plus the FP16 support that already exists in previous RDNA architectures should prove sufficient for inference-type workloads. We'll see if that proves correct going forward, but AMD seems content to leave the machine learning to its CDNA chips.
 
On the topic of sorting there was this all the back on page 4 wonder if it is part of the RT improvements rumored
Oh you're getting a whole load of those.
The question around dedicated matrix hardware was touched on in the June tomshardware interview.
Yea you're not getting MFMA in client ever.
Better SIMD tricks for sure.
And then there's XDNA AIE if you want a nice low power option for mobile.
 
Sigh, we've been talking about the software (compiler) scoreboarding for months now. Can't believe others have only just noticed.


 
Still 2 days until announcement.
I wonder If they will announce the Full N31 or a cutdown version first, as they did with RDNA2.
They'll announce both but the availability gonna be shifted.
It's a teeny 308mm^2 die as far as stuff you care about is concerned.
2 days? I thought we had to wait until Nov 7th! 😲
It's 3rd aka this Thursday.
10th for Genoa.
And today is also their Q3 ER.
 
Still 2 days until announcement.
I wonder If they will announce the Full N31 or a cutdown version first, as they did with RDNA2.
How much binning can they realistically do with N31 before they get into N32 territory?
Full chip- 48WGP and 6MCD
Rumored cutdown- 42WGP and 5MCD (but shouldn't it be 40WGP?)

I guess they could do an oddball 44WGP and 6MCD.
Just depends on where exactly they fall performance-wise. We know Nvidia was trying to price a 3090Ti replacement at $899...

I could see a lineup looking something like this:

Navi31 (~2x Navi21)
$1400-1600 48WGP and 6MCD, AIB specials w/ 450-500w
$1199 44WGP and 6MCD @ 350w
$999 40WGP and 5MCD @300w

Navi32 (~1.6x Navi21)
$749 32WGP and 4MCD @250w
$649 28WGP and 4MCD @225w
$549 24WGP and 3MCD @200w

Navi33 (~Navi21)
$479 16WGP @180w
$429 14WGP @150w
$379 12WGP @ 130w
 
Last edited:
Status
Not open for further replies.
Back
Top