Question 'Ampere'/Next-gen gaming uarch speculation thread

[Ottonomous]

How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.

 

[Panino Manino]

Again, no one will ever need so much power to play games, so whatever "Big" Navi delivers is enough. But as I knew too well there's no way for AMD to catch up with Nvidia. The ludicrous idea that AMD was on it's way to pass Nvidia within two generations was pure insanity.

 

[Glo.]

And how many ALUs does it have?

 

[Bouowmx]

2x FP32 is marketed with 2x CUDA cores.

Damn, how many jebaits did this train ride have.

 

[Stuka87]

Give it a rest already.

You have attacked all of us for months about how all of us were flat out wrong on everything. Now that we have been proven right, you are telling us to take a rest?

How about instead you admit you were wrong, or apologize to all the people you have been attacking?

EDIT: Minor typo fix

 

[Hitman928]

Nvidia website listing 3080 FE price as $699 so I was wrong, looks like that will be the actual price. That's really good pricing for the 3080 but it also seems way out of place with the 3070 at $499 and the 3090 at $1499.

 

[TESKATLIPOKA]

2x Cuda FP32? So TMUs and ROPs probably won't be doubled. I have to wonder how It will affect the performance.

 

[DooKey]

Nvidia website listing 3080 FE price as $699 so I was wrong, looks like that will be the actual price. That's really good pricing for the 3080 but it also seems way out of place with the 3070 at $499 and the 3090 at $1499.

Gotta make room for a 3080 Ti later on.

 

[badb0y]

I kinds missed the end portion there, did they give an indication of when we can expect independent reviews?

 

[DXDiag]

How about instead you admit you were wrong, or apologize to all the people you have been attacking?

Don't let the NVIDIA performance numbers that you claimed are so low blind you out of balance. Take a moment to admit reality and that you were wrong and move on.

 

[Stuka87]

Seems like Kopite nailed everything down.
PCB, Process, G6X, Cores. Everything...

Only thing that stands out is that, why are they so affordable?

I don't know, I think saying an $800 GPU is "affordable" is crazy. Especially when parts of the world have upwards of 20% unemployment right now.

 

[Glo.]

2x Cuda FP32? So TMUs and ROPs probably won't be doubled. I have to wonder how It will affect the performance.

Isn't it the same as Dual CU in RDNA1 being able to execute two wavefronts?

And the number of ALUs is still the same?

 

[Capt Caveman]

I hope so, I don't mind paying that price once. I've had my 1080ti for a while now. I would intend keeping the new card the same or longer. I just dont want to buy all new Mobo / case.

My guess, just like the 1080ti(which I have), it will come down the road and bridge the 3080 and 3090.

 

[StinkyPinky]

I don't know, I think saying an $800 GPU is "affordable" is crazy. Especially when parts of the world have upwards of 20% unemployment right now.

Well these are enthusiast class GPU's. Plus they are powerful enough that I suspect you can use them for years, a bit like the 1080ti.

The 3070 and below are more than mainstream GPU's and they seem reasonably priced for the performance IMO.

 

[DXDiag]

Isn't it the same as Dual CU in RDNA1 being able to execute two wavefronts?

Now we are trying hard, aren't we?

 

[DooKey]

Well these are enthusiast class GPU's. Plus they are powerful enough that I suspect you can use them for years, a bit like the 1080ti.

The 3070 and below are more than mainstream GPU's and they seem reasonably priced for the performance IMO.

Yep, if it turns out true that the 3070 is a 2080 Ti class GPU for $499 then I'd call that a huge win. Will sell like hotcakes.

 

[Sable]

Nvidia website listing 3080 FE price as $699 so I was wrong, looks like that will be the actual price. That's really good pricing for the 3080 but it also seems way out of place with the 3070 at $499 and the 3090 at $1499.

Leaves room for some supers to slot in...

 

[amenx]

Big marketing mistake to call the 3090 that. Calling it Titan would at least 'justified' its price in the eyes of many. Especially after he called the 3080 the 'flagship'.

 

[sze5003]

My guess, just like the 1080ti(which I have), it will come down the road and bridge the 3080 and 3090.

Yea I'm hoping it's not too far out. But the price gaps are weird. $1500 for a 3090 and $700 for a 3080. Where does that leave the 3080ti? 8-1k range?

 

[Stuka87]

Well these are enthusiast class GPU's. Plus they are powerful enough that I suspect you can use them for years, a bit like the 1080ti.

The 3070 and below are more than mainstream GPU's and they seem reasonably priced for the performance IMO.

Yeah, but there is a distinct difference between "priced well for the performance" and "affordable".

 

[DooKey]

Big marketing mistake to call the 3090 that. Calling it Titan would at least 'justified' its price in the eyes of many. Especially after he called the 3080 the 'flagship'.

Maybe a Titan is coming later?

 

[Bouowmx]

There's got to be a GeForce RTX 3090 Lite in the works. Everything the same, except 12 GB memory.

 

[uzzi38]

In a weird sort of turnabout, AMD now has the perf per Flop advantage.

Jokes aside, Pricing for the 3070 and 30870 is better than expected. I'm impressed by this launch, but it's clear that that's for a reason

 

[Saylick]

Yeah, that advertised CUDA core count is suspect... RTX 2080 Ti was 4352 cores and the 3080 should be about 35% faster than the RTX 2080 Ti if we go by this plot:

Either 2x the core count leads to 35% more performance, which would be terrible scaling, or there's simply the same number of cores/ALUs but higher IPC and higher core/mem clocks are the real reason for the performance gain.

EDIT: I'm thinking Nvidia is now advertising both FP and INT cores as CUDA cores now, which they did not do for Turing. Also, Nvidia advertise the cards as using "Shader-TLOPS", not "FP32 TFLOPS", which is interesting.

 

[CakeMonster]

A 3080Ti could be 1 year out, its unfortunately a gamble to wait for it, the whole situation could be different by then.

 

[Konan]

In a weird sort of turnabout, AMD now has the perf per Flop advantage.

Jokes aside, Pricing for the 3070 and 30870 is better than expected. I'm impressed by this launch, but it's clear that that's for a reason

How do you figure that?