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.

 

[Kenmitch]

I really like that cooling. Hot air doesn't get trapped under the card since other fan pushes it out of the case and other one sucks it through the card. This could work really nicely inside normal case.

I'd imagine a few reviewers will give the cooler a good testing out. It would be silly to just do the standard open air test benching. I'd guess as long as a person has decent airflow and some cpu cooler headroom it would turn out fine....Not sure about the gotta have a silent rig guys/gals. I guess it really depends on what the pull cools in the end.

 

[Konan]

EVGA 30 series announcements - Looks good

EVGA - Articles - EVGA GeForce® RTXᐪᔿ 30 Series

The EVGA GeForce® RTXᐪᔿ 30 Series are the absolute definition of ultimate performance.

www.evga.com

 

[Olikan]

No sub-3070 SKUs info, yet

No point, sub 3070 is way too close to consoles

 

[Glo.]

Titan RTX is 10%~15% faster than 2080Ti, in practice this puts 3090 @ 65% faster than 2080Ti. More with RT features.

Lets see if you tell the truth/are honest.

Nope. Its 15% IN BEST CASE SCENARIO, not on average.

On average its more like 8%, so up to 50% faster makes it... 55% faster than RTX 2080 Ti.

 

[IntelUser2000]

Titan RTX is 10%~15% faster than 2080Ti, in practice this puts 3090 @ 65% faster than 2080Ti. More with RT features.

Based on specs I agree. 2080 to 3080 is ~80%, then 2080 Ti to 3090 is going to be ~60%.

 

[Karnak]

Titan RTX is 10%~15% faster than 2080Ti

No, it's only like 5-8% faster.

in practice this puts 3090 @ 65% faster than 2080Ti

Since it's marketing and there are, you know, these two words "up" and "to", it's probably more like only 40-45% over a whole benchmark parcours with like 15-20 games.

 

[Tup3x]

Although not the double from 2080 to 3080 that Jensen said in the presentation. He should have said, double the performance in 2 or 3 titles. On average it seems to fluctuate between 65% and 80%.

That is still VERY impressive, please don't misunderstand me

Well, "performance" is rather vague term. Depends entirely what you are measuring. Nevertheless, the actual real word performance gains are impressive so they did do rather nice job.

 

[eek2121]

AMD will have RTX, I am not sure about a similar tech to Tensor cores and DLSS. They will have to price It accordingly or they will have a problem selling their cards. We will see. I am pretty happy with Ampere except It was not built using 7nm process.

It won’t matter if AMD has a tensor core equivalent or not as long as they price their GPUs appropriately.

RTX 2070S is faster than RX 5700 XT and 80CUs doesn't mean 2x more performance than Navi 10 because of not linear scaling. It all depends on clocks and IPC gain If any compared to RDNA1.

Except the PS5 GPU with 36CUs at “up to” 2.23Ghz with 135W power consumption beats the 2070S quite easily.

People are going to be shocked when AMD does the reveal.

Where it does matter is Perf:Watt. Which nVidia pushed on people for years as being extremely important. Its clear now that they are trying to move people away from thinking that. But it doesn't change the fact that having a GPU dump 350-400W of power into your case can cause significant cooling issues for a lot of people.

Most of the hot air is dumped out the back of the GPU. Their solution makes complete sense to me. Who knows better? Engineers at NVIDIA or armchair engineers on AT forums?

 

[IntelUser2000]

The performance gains in RT seems to be quite muted. Maybe in Cyberpunk 2077 it'll be different, but as it stands its a difference between 70% and maybe 85%.

 

[Glo.]

No point, sub 3070 is way too close to consoles

But they are still interesting.

 

[JoeRambo]

The performance gains in RT seems to be quite muted. Maybe in Cyberpunk 2077 it'll be different, but as it stands its a difference between 70% and maybe 85%.

That is the most suprising thing for me. If anything I was expecting some seriuos fraud ( read number massageing) with "up to 666% more performance" claims in one specific scene with RTX + DLSS 3.1 enabled. Yet the opposite has happened -> raw performance is there, but RTX/DLSS gains are rather muted?

 

[exquisitechar]

The performance gains in RT seems to be quite muted. Maybe in Cyberpunk 2077 it'll be different, but as it stands its a difference between 70% and maybe 85%.

Yes, I was surprised at the meager performance improvement gap in RT and non-RT games. I suppose it's not that surprising, though.

 

[eek2121]

Pretty sure some people here said 35% and insisted on it! Where are they now?

They were likely comparing vs. the 2080ti. Al so, spoiler: that NVIDIA clip from DF may have been using some form of DLSS, we will see when benchmarks drop.

 

[Kenmitch]

Engineers at NVIDIA or armchair engineers on AT forums?

AT's got some of the best armchair engineers! Why you feel the need to pick on them is beyond reprehensible! /s

 

[IntelUser2000]

@JoeRambo @exquisitechar Damn its those speculators talking about 4x improvement that misled everyone.

I'm assuming maybe its because current Ray Tracing implementations are pretty light. Which is why I'm thinking Cyberpunk 2077 will gain more. If not, then Nvidia decided balanced gains are better than going "hur dur RT cores!" or something.

 

[Stuka87]

Most of the hot air is dumped out the back of the GPU. Their solution makes complete sense to me. Who knows better? Engineers at NVIDIA or armchair engineers on AT forums?

For the record, I have worked on actual engineering projects. And pretty sure I just sat on a stool at my bench.

And while the FE version may dump a percentage of the heat out the back, the AIB cards do not. As we have already seen photos of them. And AIB cards typically consume more power than FE cards, while at the same time making up the majority of card sales.

 

[Racan]

Very very impressive, I don't remember such a large increase of gpu power gen on gen from any gpu company ever. But maybe I'm wrong.

Nvidia GeForce RTX 3080 Early Look: Ampere Architecture Performance - Hands-On!

 

[AtenRa]

They went ballistics with performance and perf/watt is only 37% higher vs Turing.

 

[Tup3x]

For the record, I have worked on actual engineering projects. And pretty sure I just sat on a stool at my bench.

And while the FE version may dump a percentage of the heat out the back, the AIB cards do not. As we have already seen photos of them. And AIB cards typically consume more power than FE cards, while at the same time making up the majority of card sales.

Most AIB cards have, quite frankly, brain dead designs. I wouldn't be surprised if NVIDIA's cooler ends up being better in real world use cases (closed case, not test bench).

 

[IntelUser2000]

RT core only improves by 70% according to Nvidia. Yup. All those speculators were wrong. 4x was bandied around so often that we took it as if it was straight from Nvidia.

Very very impressive, I don't remember such a large increase of gpu power gen on gen from any gpu company ever. But maybe I'm wrong.

Indeed you are, but not to be faulted since last time they got that was in 2006 with Geforce 8800: https://www.anandtech.com/show/2116

 

[AtenRa]

Very very impressive, I don't remember such a large increase of gpu power gen on gen from any gpu company ever. But maybe I'm wrong.

GTX1080 vs GTX980 had almost the same performance uplift.
Not to mention 1080 only increased its TDP by 15W vs 80W on the 3080 vs 2080S

 

[DooKey]

GTX1080 vs GTX980 had almost the same performance uplift.

And it wowed everyone at the time. Pascal versus Maxwell was a nice increase. Looks like Nvidia did it again.

 

[linkgoron]

I honestly don't think AMD will be able to compete with 3070 and 3080. You guys are being bit too optimistic considering AMD track record in graphics card. I first need to see a price cut on 5700XT to $300 to start things and $200 on 5600XT.

Call me optimistic, but I don't see how AMD won't be able to compete with the 3070 and 3080. The 3090 is clearly out of reach, but if the 2080ti is currently 50% faster than the 5700xt at 4k, when the 5700xt is 251mm^2, I don't see how a newer architecture at double the size won't be able to beat the 2080ti by 20%-30%.

 

[AtenRa]

And it wowed everyone at the time. Pascal versus Maxwell was a nice increase. Looks like Nvidia did it again.

Without the power efficiency though

 

[IntelUser2000]

They are claiming 1.9x perf/watt, so they just chose to push performance on the top end. I wouldn't be surprised if we see some nice high end mobile GPUs out of Ampere.