But not nearly as much as you think. Especially not in consoles where limits are put into frame counts. If you Vsync a GTX 1080 at 60FPS bandwidth becomes basically irrelevant as it doesn't come close to being choked by it anymore.

Besides, at 4K, the bandwidth is negligible as the raw power of the GPU is the limiting factor by then. Even a 10% bandwidth increase often only results in 1-2% performance increase in 4K, if any at all.

2020 lauch.

Gpu: not a big jump, they will taut 10TF, will most like run like a RX580.
Cpu: big jump here, custom ryzen 8 cores.
Memory: 4GB ddr4 for system and dedicated 16 GB HBM on games, big jump in bandwidth.

Extras: 2TB HD, not true 4k in all games, will use "checkboarding" in most big games, will cost $499, full backward whit ps4.

It is a different standard than GDDR5. But it's also an extension of it with a few features taken from GDDR6.

I doubt GDDR6X is going to be a thing.

The age of something is ultimately irrelevant. It's either expensive to manufacture or it's not.
Otherwise you are delving into a logical fallacy.

HBM1 thanks to it's interposer is more expensive than conventional DRAM to manufacture and implement.

Has LPDDR5 even been ratified yet? It took years for the industry to transition from LPDDR3 to LPDDR4 even in the mobile space.

I agree. Bandwidth is becoming a big issue, especially at higher resolutions.
Hopefully AMD can actually make it's architecture more efficient by then.

I prefer everything to be in 4k. Even on a 1080P display. Why? Down-sampling.

And even if you are still running with an antiquated 1080P display, once you upgrade to 1440P or 4k, you get a free boost to visuals.

It really depends on the PC. If the PC isn't doing 4k on mid-range hardware in 2020, then next gen consoles aren't.

It's as simple as that.

But you start throwing next-gen bandwidth intensive rendering techniques... Then bandwidth becomes a factor.
I don't think people understand how intensive a next-gen Global Illumination implementation can be.

well, 10 teraflops is not great, but it´s a jump. from 1.84 to 10 teraflops, you have 5,4 times more calculations.

They have to think well about mem bandwith. Because HBM2 is expensive, and using DDR6 more bus width is more power needed.

CPU is a must. This article from Ubisoft shows that when full utilized, PS3 CPU is on the same level , even a little faster than PS4 (without using GPU for calculations).

http://www.redgamingtech.com/ubisoft-gdc-presentation-of-ps4-x1-gpu-cpu-performance/

1. At this point, and since they obviously will wait for 7nm fabrication. Having only 10TF is a joke and almost impossible. Look at it this way. Going from 28nm to ~14nm fabrication resulted in going from 1.2TF GPU to 6TF GPU. And you think going from 14nm to 7nm will not even double that 6TF but end up at 10TF? How? Why?
   
   
2. Considering Jaguar..... anything else is probably a big jump anyways.
   
   
3. The HBM thing again, consoles are just built that way. Using HBM instead of GDDR6 will literally mean they have to spend about 30-50% more for the same amount of memory. They will rather put that money into having more GPU. Makes no sense really. Its not like what they can get with GDDR6 and still be running well cheaper isn't good enough.
   
   
4. Doubt the 2TB thing. Think more like external HDD support from day one.

You should compare original ps4 to ps4 pro. You see a 1,8 TF jump to 4,2 TF, overall 2,3x increase. And this gave a higher power consumption when comparing the two consoles. Expect something similiar with Ps5 if they launch 2019-2020 with 7nm transistor technology. About 9 TF is my guess.

If they go for a higher TF number, they either has to increase the chip size or go with higher clock speed. Which just lead to higher cost for a bigger chip and more expensive cooling.

Why though? we could just as easily compare the OG PS4 to the XB1X. Afterall we are locking at the same architecture but with 40CUs(4 deactivated) on the PS4pro and 44CUs(4 deactivated) on the XB1x. The XB1X GPU is just running at a much higher clock.

I dont see how you can say all you have said and just ignore that the XB1X exists. Forget the cost saving shenanigans in the PS4pro. The XB1X represents whats possible. Look at it this way, using the exact same chip size in the XB1X but at a 7nm node will give you 88CUs, where if clocked identically would give you at least 12TF. And this is assuming its the exact same architecture. 

But it won't be, so those 88CUs will perform better and be more efficient, clock speeds will go up by at least 10-20%.... all that is already putting you in 14/15TF territory right there. 

Considering how easy it would be for them to hit 12-15TF GPU and have that for the next 6-8yrs you really think they will puss out on that and settle for 9-10TF? When they know they have competition in the Xbox? Not a chance.

Of course I would like more TFlops, but let´s be practical here. With a balanced memory hierarchy and 9 Teraflops, The PS5 would be just like the PS4 in full HD. I mean, most games (not all) run 30 fps 1080p with 1.84 Teraflops on PS4 with a crappy CPU. 

1,84 x 4 = 7,36 Teraflops. So an 8 teraflops machine would be able to run in 4k 30 fps in most games if data feeds the gpu acordingly. Now, If you tell me, I would like at least 30 % more perf than that for anti aliasing and better dinamic resolution if games run at 60 fps. That´s a little more than 10 teraflops. Would that feel like a new console generation?. No, but it will look nicer than a 1080p standar PS4 for sure. 

As long as they sell 2 to 3 times more consoles and have much better exclusives, I guess they will concentrate in a good lauch price regarding how much teraflops they loose in making that choise. After all they can sell a more powerfull PS5 pro to nerds like me that care about that shit. 

I'm not ignoring xbox one x. Xbox one x proves my point. Microsoft had to go with a bigger chip and more expensive cooling to get those extra Flops. So if we assume 7nm will be able to double the Flops with same power consumption we will get 12 TF. But this require a bigger chip and more expensive cooling.

How much improvement we will get from 7nm compared to 16nm we don't know exactly. TSMC (The chip manufacturer) has said 7nm brings about 60% power savings. But this is probably a bit optimistic.

And I don't expect sony to go that route microsoft went with xbox one x. They wanna keep the power consumption at the same level and keep cost down.

About Amd next architecture Navi, I'm not expecting much at all. Polaris was bad, Vega was a disaster and we already hearing Navi will be worst than Vega so don't expect much from Navi.