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

Status
Not open for further replies.
I doubt AMD is interested in the phone market. Qualcomm and ARM between them have snapped up every available GPU spot in SOCs that aren't Apple. Imagination literally had to sell itself after it's nigh one customer (Apple) ended their contract, and are struggling (probably in vain, that 128 thread group processor sounds way too big) to get back into business after an investor group bought them. I suspect that after failing at custom CPUs Samsung is going for a cheaper, quicker hope for custom GPUs in their SOCs.

A smarter move for AMD would probably be to get into miniaturized SOCs/APUs like Intel's Y series. Foldable Windows devices are "on the way" and frankly the 2 foldable phones this year are kind of dumb, they need to be bigger and will thus have a larger TDP (who the hell cared if you get like, only 1" screenspace over ROG 2, need bigger dudes!). I could see AMD making money off 5 watt+ (sustained) SOCs for mini foldable tablets. They'd need a secondary "always on" CPU arch but those tend to be much simpler and cheaper to design than big high perf cores. RDNA itself should be quite scalable down to there, a single "block" would have only a 64bit bus, less already than some phone SOCs and more efficient, cut a WGP and bin for extreme energy efficiency over other laptop parts and there you go.
 
AMD can't launch SoCs under 15W.That's part of their Samsung RDNA deal.

I.e. AMD won't launch a 5W Y series competitor.
 
ToTTenTranz said:
AMD can't launch SoCs under 15W.That's part of their Samsung RDNA deal.

I.e. AMD won't launch a 5W Y series competitor.
Click to expand...
Source? Sounds awfully high limit especially when AMD has products on the market with lower cTDP than that
 
Kaotik said:
Source? Sounds awfully high limit especially when AMD has products on the market with lower cTDP than that
Click to expand...

I'd think so. And Anandtech confirms that Samsung can't compete in AMD areas, not the other way around (It's AMD that's selling, why would they lock themselves out of so much?). So it seems to confirm it's basically "smartphone only" and maybe android tablets for Samsung, and no smartphones but everything else for AMD: https://www.anandtech.com/show/14492/samsung-amds-gpu-licensing-an-interesting-collaboration
 
Last edited:
Correct as it's told in anandtech's article it's the other way around. Though if Samsung is interested in making a several million dollar deal with AMD, the non-compete is most probably going both ways.


As for TDP, I doubt the deal would include minimum cTDP values as you can probably put any value there.
 
Last edited by a moderator:
PSman1700 said:
https://wccftech.com/amd-radeon-rx-navi-21-gpu-2x-performance-5700-xt-die-size-rumor/

Damn, double the performance of 5700XT, that's 18 to 20TF of performance, using MS's analogy 2x the performance. I want that new AMD gpu :) Nvidia better be prepared.
Click to expand...

Sadly I'm gonna say "doubt" on this rumor. They basically just doubled the die size and said "it's twice as fast". But that's not how scaling to performance works, and how would double the die size and assumedly double the TDP work? HBM is too expensive for consumer GPUs to lower TDP, and even taking "full" advantage of 7nm+ a 400+ watt tdp sounds ludicrous, as does a 25% increase in power efficiency year over year just from arch if they wanted a still huge 300 watt tdp. I'd guess "Big RDNA" would end up as 72Cu/36Wgp/6 "function blocks" and 2 geometry procs. Switch to 16gbps GDDR6 and you'd only need to improve delta compression a bit to keep up with the added WGP per block. That'd nearly be the same "double the performance" but die size would only increase about 50% or so in area, at least based off 5700 to 5500 comparisons and including some die space for the new texture/raytracing processor. Not to mention power efficiency would still need to improve 10% or so just to get down to a 300 watt tdp.

Anyway, that's a hyptothetical example, but point is doubling die size seems silly and unlikely.
 
Frenetic Pony said:
Switch to 16gbps GDDR6 and you'd only need to improve delta compression a bit to keep up with the added WGP per block.
Click to expand...
I think they don't even need to improve the delta compression for this level of performance. E. g. Radeon RX 5700 (448 GB/s) reaches +50 % performance compared to Radeon RX 5500 XT (224 GB/s - half compared to the RX 5700). Radeon RX 5700 seems to be far from being bandwidth limited. Based on the behavior of RX 5500 XT it's possible to assume that Navi architecture should be able to reach RX 5700-level of performance even with 192bit bus (14gbps GDDR6). With 384bit + 14gbps GDDR6 memory performance level of RX 5700 ×2 should be achievable. Switching to 16gbps memory leaves room for additional +14 % performance (that's about the percentual difference between RX 5700 XT and RX 5700). A Navi-based GPU with ~200 % performance of RX 5700 XT using 384bit bus and 16gbps GDDR6 seems to be feasible.
 
PSman1700 said:
https://wccftech.com/amd-radeon-rx-navi-21-gpu-2x-performance-5700-xt-die-size-rumor/

Damn, double the performance of 5700XT, that's 18 to 20TF of performance, using MS's analogy 2x the performance. I want that new AMD gpu :) Nvidia better be prepared.
Click to expand...
They're even contradicting themselves in that.
In terms of performance, the Navi 21 GPU is said to be at least twice as fast as the Navi 10 GPU. The Radeon RX 5700 XT is the best case for the Navi 10 GPU and it comes close to the GeForce RTX 2070 SUPER, so the Navi 21 GPU could exceed the RTX 2080 SUPER's performance and even end up coming close to the RTX 2080 Ti.
Click to expand...
"At least twice as fast as the Navi 10" and then "could exceed RTX 2080 Super performance and coming close to RTX 2080 Ti".
RTX 2080 Ti is ~40-50% faster than RX 5700 XT (reference), something twice as fast as Navi 10 would run circles around it, not come close to it.
 
I think RTX3000 and next AMD definitely will be in excess of 30% of XSX, much and much more 2021. NV is going to go all out with their next thanks to AMD's progress in the GPU space.
 
PSman1700 said:
I think RTX3000 and next AMD definitely will be in excess of 30% of XSX, much and much more 2021. NV is going to go all out with their next thanks to AMD's progress in the GPU space.
Click to expand...
NVIDIA made the relevant decisions regarding their next gen long before Navi
 
trinibwoy said:
If RDNA2 on 7nm EUV can improve power efficiency by 30-40% then 2x 5700xt performance is in the realm of possibility.

That’s the stuff dreams are made of though.
Click to expand...

+ If my memory is right, at the of navi launch, there were rumors about the design of RDNA and manufacture were very hard for AMD, no ? Now, months later, I'm pretty sure they can gain power efficiency just by learning from this experience.
 
PSman1700 said:
I think RTX3000 and next AMD definitely will be in excess of 30% of XSX, much and much more 2021. NV is going to go all out with their next thanks to AMD's progress in the GPU space.
Click to expand...

AMD still have no high end. I'm pretty sure nVidia feels pretty safe. I hope they're wrong, but, you know...
 
Rootax said:
AMD still have no high end. I'm pretty sure nVidia feels pretty safe. I hope they're wrong, but, you know...
Click to expand...

I’m not sure the lack of high end competition from AMD matters that much these days. In the HPC space there’s insatiable demand for more performance and Nvidia has been pushing massive dies there for a few generations now. On the gaming side they pretty much maxed out die size and power consumption with Turing.

If AMD wants to take the lead they will have to earn it. Don’t think they can rely on Nvidia taking their foot off the gas.
 
Status
Not open for further replies.
Back
Top