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.

 

[eek2121]

If Nvidia has a 3070Ti ready to go for $599 with 16GB to counter Navi, I don't understand how it could be a good option. It will basically be a 3070 with slightly higher clock speeds and additional 8gb ram. But for $100 more you could have a much faster card with less ram. It would be kind of similar situation to GTX 780 6GB and 780Ti 3GB.

That just seems like a confusing product stack. Unless AMD really throws NVidia a curveball I don't see a 3070 Ti coming out soon.

Instead I expect we see the same thing as with Turing where there's an eventual Super line across the various models where we get better dies with that have fewer parts disabled.

The 3070 could always get a bump from going to GDDR6X. Depending on when that launches prices might allow for a 16 GB version. Or maybe they'll find someone to make 1.5 GB memory chips and we'll get a 12 GB model which is probably more than enough for what the card would need to be.

NVIDIA has both a 3070ti and 3080ti on hold. Both have increased memory and slightly higher clocks.

The 3070ti will be competing with the RX 6700 XT and the 3080ti will compete with the RX 6800 XT. The RX 6900 XT will still compete with the 3090 IF it gets released.

I have seen a rumor from a few different sources indicating that AMD’s new architecture may be approaching 3 Ghz boost clocks. I have no idea on how truthful it is, though I know that 2.3 Ghz is achievable.

At any rate, it is a fantastic time to be a gamer! I can’t wait!

 

[Tup3x]

I didn't say it didn't. Neither did I claim it previously.

They increased compute cores without boosting anything else so the gain isn't as big as expected.

They did boost L2 bandwidth. The actual boost in game depends on what it's actually doing but games lean more on FP than INT. For pure FP processing the boost should be basically double.

 

[Head1985]

If Nvidia has a 3070Ti ready to go for $599 with 16GB to counter Navi, I don't understand how it could be a good option. It will basically be a 3070 with slightly higher clock speeds and additional 8gb ram. But for $100 more you could have a much faster card with less ram. It would be kind of similar situation to GTX 780 6GB and 780Ti 3GB.

I expecting full GA104die will have 3584/7168SP.They keeping it for 3070TI and release it after big navi or later as refresh.Anyway it will be 20% faster(with gddr6x) than stock 3070 and will fill the gigantic hole between 3070 and 3080.Because it looks like 3080 will be 40% faster than 3070.
There is just no way ampere 104die have same SP number as turing.3070 is just cutdown die and full die have 3584/7168SP.

 

[AtenRa]

I didn't say it didn't. Neither did I claim it previously.

They increased compute cores without boosting anything else so the gain isn't as big as expected.

Well, they did increased the SM Shared L1 cache from 96KB to 128KB and also doubled the L1 bandwidth so Ampere L1 is larger and 2x faster vs Turing.

edit: Tup3x beat me to it.

 

[eek2121]

I expecting full GA104die will have 3584/7168SP.They keeping it for 3070TI and release it after big navi or later as refresh.Anyway it will be 20% faster(with gddr6x) than stock 3070 and will fill the gigantic hole between 3070 and 3080.Because it looks like 3080 will be 40% faster than 3070.
There is just no way ampere 104die have same SP number as turing.3070 is just cutdown die and full die have 3584/7168SP.

I guess that will depend on binning, but I could see a full fat core, it depends on Samsung.

 

[Glo.]

I expecting full GA104die will have 3584/7168SP.They keeping it for 3070TI and release it after big navi or later as refresh.Anyway it will be 20% faster(with gddr6x) than stock 3070 and will fill the gigantic hole between 3070 and 3080.Because it looks like 3080 will be 40% faster than 3070.
There is just no way ampere 104die have same SP number as turing.3070 is just cutdown die and full die have 3584/7168SP.

It doesn't. Full 104 die has 48 SM's - 3072ALUs/6144 CUDA cores.

I have seen a rumor from a few different sources indicating that AMD’s new architecture may be approaching 3 Ghz boost clocks. I have no idea on how truthful it is, though I know that 2.3 Ghz is achievable.

The rumor about 3 GHz is complete BS.

2.3 GHz is what we should see on the lower CU count GPUs. N21 will clock lower, than that.

 

[Kenmitch]

https://twitter.com/x/status/1302610796623478792

Dang internet can't make up it's mind!

 

[Mopetar]

That is incorrect , there are double the FP32 ALUs

It will be interesting to see how it works out in reality. In some ways it's similar to the approach that AMD started taking with Fury where they added more and more shaders while NVidia focused on fewer but more effective shaders.

I'll give NVidia some benefit of the doubt because they have some really good engineers that no doubt spent a lot of time grappling with the issue, but with Fury there was always the issue of keeping all of those shaders fed and busy.

If nothing else it might make the 3080 look even more attractive compared to the 3090 than it already is.

 

[Tarkin77]

sorry, wrong thread

 

[Bouowmx]

https://twitter.com/x/status/1302610796623478792

If NVIDIA can rush GA103 (60 SMs or 7680 cores after doubling) on a 7 nm process in time for Big Navi (end of 2020), is the rest of the family going to be 7 nm, or why isn't the whole family on 7 nm?

 

[Konan]

If NVIDIA can rush GA103 (60 SMs or 7680 cores after doubling) on a 7 nm process in time for Big Navi (end of 2020), is the rest of the family going to be 7 nm, or why isn't the whole family on 7 nm?

Don't think it needs to be on 7nm. Why not continue on 8nm?
I don't think we'll see a refresh on 7 to be honest, at least not for a quite a while.

 

[Karnak]

Generally speaking I don't see another chip between GA104 (48SM) and cut-down GA102 (68SM). Already too close IMO. Other thing is that I think a potential GA103 on TSMC's N7 would make GA102 look really bad especially in terms of efficiency.

 

[Gideon]

I didn't say it didn't. Neither did I claim it previously.

They increased compute cores without boosting anything else so the gain isn't as big as expected.

That's true, but it's still a major benefit. AMD's GPUs have for a long time been really compute-heavy and have thus aged quite a bit better than similar Nvidia GPUs. I also expect the 3xxx series to also age much better than 2xxx in this regard, once more shader heavy games appear in the future.

 

[JoeRambo]

Generally speaking I don't see another chip between GA104 (48SM) and cut-down GA102 (68SM). Already too close IMO. Other thing is that I think a potential GA103 on TSMC's N7 would make GA102 look really bad especially in terms of efficiency.

Remember 8800GT? It had nearly the speed of 8800GTX, but used like 66% of power. Same could happen for GA103 versus 3080. Offering say 95% of perf, but at 66% of power and 20GB of DRAM.
Would certainly make sweet card ~200W.

Of course 8800GT had advantage of jump to 65nm and i doubt SS8 => TSMC7 is that much of efficiency difference, but NV in the past did not shy away from cannibalizing its own products if it meants great product and ton of sales.

 

[Glo.]

Don't think it needs to be on 7nm. Why not continue on 8nm?
I don't think we'll see a refresh on 7 to be honest, at least not for a quite a while.

Because Nvidia needs something to compete with Navi 22.

And reusing 102 dies are too expensive.

If RTX 3080 is 68 SMs with 320 Bit GDDR6X, then 103 with 60 SMs and 256 bit GDDR6X will get you around 85% of RTX 3080 performance.

And don't forget. Already with 2944 ALUs/5888 CUDA cores RTX 3070 is 220W TDP, and full 48 SM with GDDR6X would be 250W TDP.

Don't expect 60 SM/256/320 Bit design to be any less than 300W on 8 nm process.

And that is not going to compete with 60 CU RDNA2 GPU that will use around 200-225W of power.

 

[sontin]

What? A 60CU RDNA2 will be only as fast as a 3070. nVidia doesnt need any other chips. 3070 is faster with compute, faster with raytracing and faster with DL.

Maybe AMD need something bigger. But nVidia has covered every corner.

 

[Helis4life]

EVGA GeForce RTX 3090 FTW3 teased with 2.1 GHz OC boost clock

target temp is 83 Celsius in image

https://twitter.com/x/status/1302940049751912448

Really curious what the power usage is

 

[Glo.]

What? A 60CU RDNA2 will be only as fast as a 3070. nVidia doesnt need any other chips. 3070 is faster with compute, faster with raytracing and faster with DL.

Maybe AMD need something bigger. But nVidia has covered every corner.

Nope.

 

[sontin]

Sure. A 3070 is 60%+ faster than a 5700XT. Maybe you should explain how a 60CU Navi card is beating this with the same power consumption.

 

[Zstream]

What? A 60CU RDNA2 will be only as fast as a 3070. nVidia doesnt need any other chips. 3070 is faster with compute, faster with raytracing and faster with DL.

Maybe AMD need something bigger. But nVidia has covered every corner.

That's the most ridiculous statement I've heard. The XBX GPU is equal to the 2080 super and is at 140-150W. It's not difficult to reach the 3080 perf at 275W. In fact, with the decision to go less dense, and more perf on the node, it would likely be able go to 300W while keeping decently cool with an ok cooler.

Anyways, completely off topic for this thread. The new GPU's will be nice, but the $$$ is too much. I can't see people spending this much on new hardware when the next gen consoles are coming out for less than a single GPU card. You get a flipping mobo, ram, GPU, OS etc.. for 3080 price. Good luck with that.

 

[sontin]

No, the XBSX is not equal to a 2080S. It performs around a 2080. And a console SoC is always more efficient than a stand alone GPU. To be faster than a 3070 AMD has to deliver 70% more compute performance within 220W.

 

[Zstream]

No, the XBSX is not equal to a 2080S. It performs around a 2080. And a console SoC is always more efficient than a stand alone GPU. To be faster than a 3070 AMD has to deliver 70% more compute performance within 220W.

Who told you it's equal to a 2080? It surpasses the 2080 in all aspects. And no, it doesn't need to be 220W. That's you applying stupid rules to your argument. It's baseless and silly.

 

[leoneazzurro]

Rumors say +50 CU, +19% clocks, +10-15% IPC. But as this is OT, I'd say these comparisons should be done in the RDNA2 thread or in a dedicated thread.

 

[Glo.]

Sure. A 3070 is 60%+ faster than a 5700XT. Maybe you should explain how a 60CU Navi card is beating this with the same power consumption.

Maybe you should explain first, where did you got that 3070 is 60% faster than RX 5700 XT?

60% faster than RX 5700 XT means that it is 15% faster than RTX 2080 Ti, which makes it 10-15% slower than RTX 3080. Do you genuinely believe anybody here is that stupid to buy what you are selling?

Secondly.

https://twitter.com/x/status/1302023768613761024

RTX 3070 even according to Nvidia materials in Unreal Engine 4 game is 43% faster than RX 5700 XT in 1440p.

Thats not really painting good picture for Ampere, does it?

Thirdly. Stop spreading BS about Ampere and RDNA2 based GPUs.

 

[Gideon]

RTX 3070 even according to Nvidia materials in Unreal Engine 4 game is 43% faster than RX 5700 XT in 1440p.

I wonder if they used DX11 or DX12 in the Borderlands 3 test.

According to techpowerup 2080 Ti is 49% faster than 5700 XT at 1440p DX11, but with DX12, the 2080 Ti is only 30% faster at 1440p.

In DX11 5700 XT is about equal with a vanilla 2070, in DX12 it's faster than 2070 SUPER (as 5700 XT gains 6 FPS while all Nvidia cards lose a few).

Regardless it seems that in this game 3070 doesn't quite reach 2080 Ti levels.