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

Performance comparison talk moved to more appropriate discusssion: https://forum.beyond3d.com/threads/speculation-gpu-performance-comparisons-of-2020-spawn.61885/

 

Long time no posting, but looks like this could be the best season for videocards in a very long time, if Radeon really comes back to highend fight...

i think Coreteks usually has some true leaked info but just mixes some pieces.

Beeing optimistic and trusting Lisa, if Radeon wants to compete in high end again, they can do it based on improved RDNA1 and promised 50% performance/clock uplift, but they need 80Cus and high clock for RTX 3080/3090. 3090 is just 15-20% more powerful than 3080 this time. NVIDIA could release quite more expensive 3070 16GB and 3080 20GB, or call them Ti/Super, and also maybe some 3090 Ti or Titan with 84SM, but 3090 already has 82SM, so if Gaming Ampere 8nM full is 84SM, not a lot of margin there, not even in clocks considering power consumption.
AMD may need not just a small IPC gain but also say 20% more clock than navi10, 2300Mhz or so similar to PS5 clock...so the question is can they do it without sacrificing performance/power sweet spot as they were forced to in recent past launches? Remember 5700 and 5600 are both quite more efficient and lower voltage than 5700 XT.
What if that new Radeon RDNA2 is GDDR6 8GB/16GB 256-bit RX 6800 navi22 60Cus, close to or a bit higher than 3070/2080 Ti performance, and the "big navi" Navi 21 80/72Cus is HBM2 16GB/12GB RX 6900? Navi23 is expected later, maybe christmas, and should be great midrange, quite faster than RX 5700 too for around 249 USD, but both navi 22 and 21 should be released in october before consoles. If they manage 50% P/W increase and high clocks, could reach 225% more performance than RX 5700 XT, and thats around 45% more than 2080 Ti, so just between both 3080 and 3090.

 

Thats probably one of his most unlucky ones AFAIK, and why i said "usually"...for a leaker, don´t know about his sources though.
If RDNA2 with close to RTX3070/2080 Ti performance exists, i can´t believe its "big navi", should be navi 22.

 

Ok, teasing is starting

https://hothardware.com/news/amd-big-navi-easter-egg-discovered-in-fortnite-map

 

It is one.
And, well, "close to".
Haha.

They're shitlording on social media alright.

 

They will sale again at cost (ie like Radeon Pro VII) to compete with new NV aggressive RTX3080, which is exactly what they don't want. What's the point to sale this expensive silicon at bargain price when it can generate huge margins with server CPU?
If RDNA2 needs HBM to compete with much cheaper RTX30 G6x, then it's another fail and I don't think that's AMD plan. On the other side, what solution do they have ? G6x is for now Nvidia exclusive and using G6 will give their products a huge bandwidth disadvantage... I'm curious to see what AMD will bring this time to re-enter the high-end market...

 

Atta boy.
Neither ICs nor backdrill PCBs for those are even remotely cheap.
It's cheaper, but not by much.

Ha.

 

16Gbps G6 (as used on the RTX 2080 Super but downclocked slightly to 15.5 to avoid PCB changes) compared to 19Gbps G6X is only a 16% disadvantage, not exactly huge. 16Gbps on a 384bit bus would allow them to reach the RTX 3080's bandwidth level.

The bandwidth per watt gain for G6X may be more significant, but we don't know if AMD is counting relying upon any memory improvements for its 50% RDNA1->RDNA2 performance per watt gain figure.

 

They're measured board to board so if there's any, they'd be accounted for.

 

But question is then how could they offer 16GB with 384-bit GDDR6 bus? or you mean AMD could posibly go the same route as nvidia with 384-bit so 12GB/24GB max even if they cannot count on DDR6X for extra bandwidth?

 

The answer is very simple: they don't.

Yeah.

 

dunno, i can see them going HBM for the highend, and don´t forget they used HBM for Apples Navi 12 very recently too, even if the rest of navi 1x are GDDR6. Navi 12 indeed has impressive p/w. 24GB GDDR6X are very expensive too...not sure about the cost comparison with 16GB HBM2, and 16GB are more than enough for many years in any game, any quality and any resolution imo.

 

384-bit for 80cu? RX 5600XT is doing ok with 192-bit and 14gbps, but its mainstream and 32cu...also, not sure if navi10 is bandwith limited at highest res, but HBM2 offers plenty and cost is not as high as in 2015 fury x days.

 

512b with slow GDDR6 ftw. R600 and Hawaii are both legends.

 

No way they are doing that. It's HBM for top SKU.

 

So, to double the performance of a 5700xt (80 CU), it's probably going to be about 300W assuming a 50% performance per watt improvement, if I did the math right (225W / 1.5 x 2 = 300W). That would be a straight up doubling of the 5700xt, using the 50% performance per watt purely as power saving. Make sense?

 

16x1 ? seems like a expensive way to get 896 GB/s

 

Yeah, but the reality is more hilarious than that.

Yeah just use HBM.