AMD Radeon Navi Launch Timeframe Leaked - Releases After 3rd Gen Ryzen

[grec]

We already have the results.

The new 1660 (220$ US official price) is like 8% faster than RX 590. Considering the RX 590 was something like 6-10% faster than RX 580, we have that the new GTX 1660 is something around, at least 15% faster than RX 580.

this GTX 1660 is about 21% faster than GTX 1060 6gb. it's by the other hand , around 13% slower than GTX 1660ti.

That said, i don't think Navi RX "680" will beat RTX 2060. Hope it does, but it wont.

If we take the most basic of sensible assumptions, essentially using known facts, then a direct 7nm port of the RX580 with GDDR6 would result in a card at least 35% faster. That's assuming no architectural speed improvements at all. This is a minimum baseline we can take from the progress of physics, essentially. So at an absolute minimum, the RX680 will tie with the RTX2060, if the card is designed to the same outwards specifications as the RX580(Same core count, bus width, IPC).

If AMD come out with a card weaker than the RTX2060, they have either released something heavily cut down from Polaris10 specs wise, or they have completely blown their use of 7nm and are completely finished as a GPU company from top to bottom by creating an architecture that according to fundamental laws of physics would have to be worse than their previous couple.

But Navi already has at least a couple of designs in major gaming consoles, so something tells me that's not the case.

 

[Piskeante]

If we take the most basic of sensible assumptions, essentially using known facts, then a direct 7nm port of the RX580 with GDDR6 would result in a card at least 35% faster. That's assuming no architectural speed improvements at all. This is a minimum baseline we can take from the progress of physics, essentially. So at an absolute minimum, the RX680 will tie with the RTX2060, if the card is designed to the same outwards specifications as the RX580(Same core count, bus width, IPC).

If AMD come out with a card weaker than the RTX2060, they have either released something heavily cut down from Polaris10 specs wise, or they have completely blown their use of 7nm and are completely finished as a GPU company from top to bottom by creating an architecture that according to fundamental laws of physics would have to be worse than their previous couple.

But Navi already has at least a couple of designs in major gaming consoles, so something tells me that's not the case.

Are you really expecting AMD to release a chip that can outperform the last generation for about 35% at least just with the reduction in size?? omg.... AMD has never done that before. From R9 380x to RX 580 (28nm to 14nm), they could only add something around 15% faster changing both the size and the architecture. you are way too much optimistic. Way too much.

Let me share the numbers with you. RX 580 has always been on pair with GTX 1060 6gb. This has been like this for a lot of time, even with driver optimizations. Well, the RTX 2060 is like 53% faster than GTX 1060 6gb at 1440p. More if we consider 1080p. Do you really think that AMD can pull up a chip that can outperform Polaris by that margin? I don't believe it, simply can't believe it.

That's why the supposed RX 680 will not be better than RTX 2060. RX 680 will probably fight to sustain RTX 2060 competition at lower prices, but not in performance. No way AMD can get there. No way.

 

[ET3D]

If we take the most basic of sensible assumptions, essentially using known facts, then a direct 7nm port of the RX580 with GDDR6 would result in a card at least 35% faster.

Can you state which sensible assumptions and known facts these are?


Because the known fact is that 7nm can provide a 25% speed increase over 14nm. How do you get to 35%?

 

[grec]

All game benchmarks I see put the RX480 as 50%+ faster than the 380X, despite a less than 10% increase in CU count.

Although AMD used Samsung/GloFlo 14LP as opposed to TSMCs high performance 16FF (Which performs much closer to 12LP) for Polaris upto the RX590 (Whereas Navi is TSMC 7nm).

But, for a more recent comparison, the VII outperforms Vega64, which has ~5% core count advantage, by on average around 35% at 4K max, largely thanks to the bandwidth increase besides the clock speed increase, and we can expect a similar double in bandwidth from 8Gbps GDDR to 16Gbps GDDR as we got from double HBM bus width. While that 35% alone wouldn't roundly beat the RTX2060 in every game off the bat it certainly brings it in contention especially in DX12, and that bandwidth increase means a lot more here, and Polaris already had weak clock speeds for a 14nm part thanks to the weak node they had to use to satisfy GloFlo contracts.

Edit: Here is my source for the 35% figure, I think we can assume Vega20's ridiculous heat density and similar issues held it back a little there, definitely seems thermally throttled or possibly limited by the pipeline depth being unoptimised for these clock cycle durations.