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.

 

[Heartbreaker]

If they want a more capable GPU with more RAM, they won't have to pay more. They can buy RDNA2.

So instead of a GPU that might someday slowdown a little bit because of VRAM, they will start with one that's slower right away.

 

[Kuiva maa]

It is not about getting slower by the lack of VRAM. What personally prevents me from even considering the 3080 10GB model is that I will need to reduce detail from the get go and I find this out of the question. The way I see it lack of VRAM is serious because you either lose IQ or you face stutter.

 

[Mopetar]

Alternatively, people ignore both overpriced 3080s and get a 16GB AMD card that's 5% slower for $200-$400 less. If Nvidia charges $900 for the 3080 20GB ($1000+ realistically) and keeps the 10GB variant at $700, then AMD will absolutely eat their lunch. Nvidia can't afford to be Intel right now (old Intel, that is). They have to compete or they will LOSE.

AMD offering a card with 95% of 3080 performance for $500 isn't going to happen. Not after the way the 5700 was priced and how they've been pricing their CPUs. AMD only gives deep discounts when it's the only way to get people to buy the product. When it's good they charge for it.

Also, I should think that you of all people should know better than to start making dubious performance claims regarding possible AMD products after an NVidia launch.

 

[mohit9206]

It is not about getting slower by the lack of VRAM. What personally prevents me from even considering the 3080 10GB model is that I will need to reduce detail from the get go and I find this out of the question. The way I see it lack of VRAM is serious because you either lose IQ or you face stutter.

Then buy AMD. People were happily buying $1200 2080Ti up until now. Amd will give you 16gb for $600. Buy that. Although you will miss out on DLSS and good quality ray tracing.

 

[deathBOB]

Then buy AMD. People were happily buying $1200 2080Ti up until now. Amd will give you 16gb for $600. Buy that. Although you will miss out on DLSS and good quality ray tracing.

Is a 16gb card with a 500ish mm 7nm GPU for $600 possible? The 3080 is bigger, but I imagine samsung 8nm is cheaper than TSMC 7nm.

 

[amenx]

It is not about getting slower by the lack of VRAM. What personally prevents me from even considering the 3080 10GB model is that I will need to reduce detail from the get go and I find this out of the question. The way I see it lack of VRAM is serious because you either lose IQ or you face stutter.

Nvidia is fanatical about competition. They would never allow themselves to be upstaged by AMD over something like vram gimping their performance. It would be too big of a disaster for their star of the show, the 10gb 3080, to be seen as inadequate vs AMDs 16gb cards. Nvidia KNOW that the 10gb will be perfectly fine over the next 3 years or so, even at highest settings vs the competitions highest settings. The only reason they are bringing in 20gb versions is marketing for the doubters who may be swayed by the competitions 16gb. Nvidia knows game development inside and out and If there was any chance the 10gb card would fail to achieve vs the competition due to vram inadequacy, they would not have released it as a 10gb card.

 

[Ajay]

Is a 16gb card with a 500ish mm 7nm GPU for $600 possible? The 3080 is bigger, but I imagine Samsung 8nm is cheaper than TSMC 7nm.

I would think not, if the performance is there. Other than that, AMD and their AIB partners are the only ones who know.

 

[TitusTroy]

so which is the best 3080 card to purchase from a performance/capacitor standpoint?

 

[Tup3x]

so which is the best 3080 card to purchase from a performance/capacitor standpoint?

Asus TUF is the best all things considered in my opinion.

 

[Grooveriding]

so which is the best 3080 card to purchase from a performance/capacitor standpoint?

If you don't care about cost? I would say Asus Strix, Gigabyte Aorus are the current top tier cards. Performance wise, it's probably largely a lottery, there is not much headroom on these cards from what I have read and seen with overclock results. I'm fairly sure those two cards use chips that have been binned to some degree as better than most though.

Otherwise I would look at aesthetics, acoustics and warranty.

 

[JustMe21]

With the impending release of 20 GB 3080s and 16 GB 3070s and now the rumor of Nvidia moving the 3000 series to TSMC's 7 nm, people ordering 3000 series cards right now are going to become even more upset since the TSMC versions will either run cooler and generate less heat or have higher clock speeds.

NVIDIA、超微、蘋果訂單爆發　台積電三度上修資本支出機會大

面臨疫情與華為禁制令危機，台積電2020年迄今營運表現依舊穩健，近期更出乎預期再上調7奈米與5奈米月產能，顯見產能提前滿載盛況。不僅如此，供應鏈傳出，台積電三度上修資本支出機會相當大，估計增幅約5%左右，再創歷史新高。

www.digitimes.com.tw

If the highest end RDNA2 GPU does perform similar to the 3080, Nvidia won't be able to put too much of a price premium on a 20 GB 3080 as people will opt to go for a 16 GB RDNA2 card instead of a 10 GB 3080. So, for the Nvidia consumer's sake, let's hope the high end RDNA2 does compete with the 3080.

 

[Heartbreaker]

With the impending release of 20 GB 3080s and 16 GB 3070s and now the rumor of Nvidia moving the 3000 series to TSMC's 7 nm, people ordering 3000 series cards right now are going to become even more upset since the TSMC versions will either run cooler and generate less heat or have higher clock speeds.

If the highest end RDNA2 GPU does perform similar to the 3080, Nvidia won't be able to put too much of a price premium on a 20 GB 3080 as people will opt to go for a 16 GB RDNA2 card instead of a 10 GB 3080. So, for the Nvidia consumer's sake, let's hope the high end RDNA2 does compete with the 3080.

Dream on. Double GDDR6x will cost at least $200. No real logic to them switching to 7nm. It costs tens of millions of dollars to move a design to another Fab.

I remember similar talk about Turing. "Don't buy 12nm Turing, Turing 7nm is coming in 6-12 months...".

 

[Stuka87]

With the impending release of 20 GB 3080s and 16 GB 3070s and now the rumor of Nvidia moving the 3000 series to TSMC's 7 nm, people ordering 3000 series cards right now are going to become even more upset since the TSMC versions will either run cooler and generate less heat or have higher clock speeds.

NVIDIA、超微、蘋果訂單爆發　台積電三度上修資本支出機會大

面臨疫情與華為禁制令危機，台積電2020年迄今營運表現依舊穩健，近期更出乎預期再上調7奈米與5奈米月產能，顯見產能提前滿載盛況。不僅如此，供應鏈傳出，台積電三度上修資本支出機會相當大，估計增幅約5%左右，再創歷史新高。

www.digitimes.com.tw

If the highest end RDNA2 GPU does perform similar to the 3080, Nvidia won't be able to put too much of a price premium on a 20 GB 3080 as people will opt to go for a 16 GB RDNA2 card instead of a 10 GB 3080. So, for the Nvidia consumer's sake, let's hope the high end RDNA2 does compete with the 3080.

There won't be 7nm 3000 series cards. The article clearly states nVidia will start using them in 2021, and going by how long it takes for GPU's to tape out, it won't be any time soon. So it should be expected that these cards will be sold as 4000 series. Even if it is only a direct port of the Ampere GPUs. Very similar to the GTX 480 being respun as the GTX 580.

 

[TitusTroy]

If you don't care about cost? I would say Asus Strix, Gigabyte Aorus are the current top tier cards. Performance wise, it's probably largely a lottery, there is not much headroom on these cards from what I have read and seen with overclock results. I'm fairly sure those two cards use chips that have been binned to some degree as better than most though.

Otherwise I would look at aesthetics, acoustics and warranty.

the Strix and Aorus are definitely on my list along with the MSI RTX 3080 Gaming X Trio and EVGA RTX 3080 FTW3 Ultra Gaming...anyone know the capacitor layout of the Gigabyte Aorus?...do they use a mix of MLCC and SP-Caps?

 

[Stuka87]

the Strix and Aorus are definitely on my list along with the MSI RTX 3080 Gaming X Trio and EVGA RTX 3080 FTW3 Ultra Gaming...anyone know the capacitor layout of the Gigabyte Aorus?...do they use a mix of MLCC and SP-Caps?

The issues with cards crashing was resolved by nVidia releasing a driver update. So the capacitor layout should not matter.

 

[TitusTroy]

The issues with cards crashing was resolved by nVidia releasing a driver update. So the capacitor layout should not matter.

even with the driver fix I still thought that it was best to use a mix of MLCC and SP?

 

[Hitman928]

even with the driver fix I still thought that it was best to use a mix of MLCC and SP?

It may get you slightly higher frequencies, maybe. Probably not enough to make any real performance difference but if you want every last percentage point possible, then it might be worth it.

 

[Bouowmx]

even with the driver fix I still thought that it was best to use a mix of MLCC and SP?

The running hypothesis was that a design using all MLCC was best, as in the ASUS TUF.
Ever since the driver update, almost all discussion has been resolved. So, I assume the design doesn't matter much anymore.

 

[Kuiva maa]

Nvidia is fanatical about competition. They would never allow themselves to be upstaged by AMD over something like vram gimping their performance. It would be too big of a disaster for their star of the show, the 10gb 3080, to be seen as inadequate vs AMDs 16gb cards. Nvidia KNOW that the 10gb will be perfectly fine over the next 3 years or so, even at highest settings vs the competitions highest settings. The only reason they are bringing in 20gb versions is marketing for the doubters who may be swayed by the competitions 16gb. Nvidia knows game development inside and out and If there was any chance the 10gb card would fail to achieve vs the competition due to vram inadequacy, they would not have released it as a 10gb card.

Nvidia is more fanatical of their margins. Also their halo card is the 3090 so they are covered there. 10GB isn't enough right now for a flagship, let's call a spade a spade. There is a model with twice the VRAM on its way, we all know that.

 

[Det0x]

https://twitter.com/x/status/1315638349038866433

Meanwhile, the 4K/Ultra specs have taken a major hit, upping the CPU requirements from Intel Core i7 9700K to i9 9900K and the GPU requirements from NVIDIA's Turing-based GeForce RTX 2080Ti to the brand new, almost-top-of-the-line Ampere-based GeForce RTX 3080 graphics card. Additionally, Ubisoft recommends setting DLSS to 'Performance Mode', which means Watch Dogs Legion will be effectively upscaled from 1080p to 2160p.

 

[PhoBoChai]

Then buy AMD. People were happily buying $1200 2080Ti up until now. Amd will give you 16gb for $600. Buy that. Although you will miss out on DLSS and good quality ray tracing.

"Good quality ray tracing".. it will be funny if it turns out RDNA2 RT acceleration is even better than RTX Ampere.

 

[GodisanAtheist]

And here we go...

Probably just rampant speculation and rumor-mongering but if true this would be yet another nail in the RTX3xxx series coffin. Maybe a mid gen refresh (Super Cards on TSMC 7nm) or NV does the whole new line-up thing to get away from a tarnished line name (RTX 4xxx series).

NVIDIA Reportedly Moving Ampere to 7 nm TSMC in 2021

A report straight from DigiTimes claims that NVIDIA is looking to upgrade their Ampere consumer GPUs from Samsung's 8 nm to TSMC's 7 nm. According to the source, the volume of this transition should be "very large", but most likely wouldn't reflect the entirety of Ampere's consumer-facing...

www.techpowerup.com

 

[sze5003]

And here we go...

Probably just rampant speculation and rumor-mongering but if true this would be yet another nail in the RTX3xxx series coffin. Maybe a mid gen refresh (Super Cards on TSMC 7nm) or NV does the whole new line-up thing to get away from a tarnished line name (RTX 4xxx series).

NVIDIA Reportedly Moving Ampere to 7 nm TSMC in 2021

A report straight from DigiTimes claims that NVIDIA is looking to upgrade their Ampere consumer GPUs from Samsung's 8 nm to TSMC's 7 nm. According to the source, the volume of this transition should be "very large", but most likely wouldn't reflect the entirety of Ampere's consumer-facing...

www.techpowerup.com

Maybe it would be a super refresh of the 30 series. Boy that sucks if true, still hanging on to my 1080ti for now. Wonder if this would still mean that they plan on releasing 20gb 3080's in December.

 

[Bouowmx]

TechPowerUp's opinion injected into this news report is pretty top-tier cancer.

A fluent Chinese reader would be helpful. I read the original source with machine translation, and the phrase used was "It has also been reported", which may be a restatement of previous reports, and not an affirmation.

 

[ozzy702]

And here we go...

Probably just rampant speculation and rumor-mongering but if true this would be yet another nail in the RTX3xxx series coffin. Maybe a mid gen refresh (Super Cards on TSMC 7nm) or NV does the whole new line-up thing to get away from a tarnished line name (RTX 4xxx series).

NVIDIA Reportedly Moving Ampere to 7 nm TSMC in 2021

A report straight from DigiTimes claims that NVIDIA is looking to upgrade their Ampere consumer GPUs from Samsung's 8 nm to TSMC's 7 nm. According to the source, the volume of this transition should be "very large", but most likely wouldn't reflect the entirety of Ampere's consumer-facing...

www.techpowerup.com

I wonder what kind of performance uplift TSMC's 7nm would bring. 10-15% better clocks with lower power consumption? Assuming 20GB versions, I'd absolutely be willing to hold off for that, I wonder how far out we're talking though.