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.

 

[jpiniero]

Shrink should be more than 50% actually. The question is more what do they spend the budget on.

 

[alcoholbob]

I don't know about the specs, but I do know Nvidia must be very careful with their prices this time around.

Ps5 and Xbox Scarlett are coming in one year. Ps5 is rumored to cost 500$. If it can do 4k/60fps in a convincing way, people will be fleeing by the millions if you need 350$ just for equal graphics for the pc.

If it's as much as they claim, it's pretty much the PS3 launch all over. Consoles will be considered an equal if not the more premium platform due to value & larger community for first 2 years due to Sony/MSN selling bleeding edge hardware at a loss. Between 2006-2008 PC ecosystem definitely felt like a backwater to the consoles until the launch of Intel Nehalem CPUs and the 40nm GPU node in 2009.

 

[Golgatha]

Summary of an ol' Twitter user's now-deleted writings. Kopite7Kimi joined Twitter in 2010; and this year, has reported on NVIDIA Turing SUPER very early. 3DCenter (German), English.

Concerning HPC, GA100 has 6144 (6*1024) -bit HBM, and GA101 has half as much (3072). HPC will use TSMC 7 nm +.

GA100 allegedly has 8192 cores (8 GPC, 8 TPC per GPC, 2 SM per TPC, 64 cores per SM).

Concerning GeForce, it will use Samsung 7 nm.

More stuff in the links.


My thoughts:
8192 won't fit in the 1-2-3 pattern as used in Kepler*, Maxwell, Pascal, and Turing: -06-level has 1x (the baseline) cores, -04-level has 2x, and -00/02-level has 3x.
I'm going to guess 1-2-4?

- GeForce RTX 3060 (A106): 2048 cores
- 3070 (A104): 3072
- 3080 (A104): 4096
- Mind the gap
- 3080 Ti (A102): 7936
- TITAN (A102): 8192

So basically more of the same. RAM should be 8GB, 12GB, and 16GB for the mid-range. 3080 Ti will be a strange 15GB variant and the Titan will be 16GB or 32GB.

 

[Golgatha]

Reason I said that was the last jump 980 Ti to 1080 was almost 100% the result of clockspeed gains. That gain was once in a lifetime for nvidia and has never happened before from a nodeshrink. I dont believe a 300mm2 die can pull off that much of an increase over a 2080ti which is a 754mm2 die.

If they pull it off I suspect they go large right away on 7nm, i.e. around 400mm2, which wont bode well for the price ($800 gtx 3080?)

Clockspeed and memory compression which increased efficiency or IPC if you will I believe. They got two huge boosts and weren't able to replicate that with the RTX 20x0 line.

 

[CakeMonster]

I would be very surprised if nVidia comes out with new cards so soon. It took us how many years to go from 10x0 to 20x0? I don't seen that dramatically shortening.

NV has released a new card with 20-30% performance increase every year the last 10+ years now, what makes you think it will not be possible now?

 

[maddie]

NV has released a new card with 20-30% performance increase every year the last 10+ years now, what makes you think it will not be possible now?

Every year or every generation?

 

[CakeMonster]

Every year. Most consumers don't care about the chip revision or methods use to get that increase, but NV has had something to increase performance in that range basically every year.

 

[alcoholbob]

Every year. Most consumers don't care about the chip revision or methods use to get that increase, but NV has had something to increase performance in that range basically every year.

But absolute performance peak is only going up every 2 years, so it's really more like 16-17% a year.

 

[mopardude87]

So basically more of the same. RAM should be 8GB, 12GB, and 16GB for the mid-range. 3080 Ti will be a strange 15GB variant and the Titan will be 16GB or 32GB.

I wonder if we will really need more then 8gb vram soon. I picked up a 1080 ti to hold me over till Ampere is FULLY released or close to. Besides broken COD ports consuming insane large amounts of vram, something like Pacific dlc on BF5 looks amazing yet only consumes a little over 4gb at 2k maxed out minus RT of course. My guess next gen consoles will push vram usage and warrant it but i got a feeling Nvidia could have a refresh maybe with more vram? Maybe a refresh for everything not a **80 Ti variant or Titan.

 

[jpiniero]

NVIDIA 下代 7 奈米製程產品，黃仁勳：台積電仍會取得大多數訂單

南韓媒體曾報導，南韓 NVIDIA 高層表示，公司新一代 GPU 繪圖晶片將轉單給南韓三星，並採用內含極紫外光刻技術（EUV）的 7 奈米製程技術打造，捨去原本長期合作對象台積電。NVIDIA 創辦人黃仁勳日前接受媒體訪問時澄清，未來還是會將大多數 7 奈米製程產品訂單交由台積電生產，三星只會獲得少...

finance.technews.tw

Maybe confirmation that Ampere is mostly being fabbed at TSMC, presumably 7+. Likely this is because Samsung's yields are not great and they had to go back to TSMC. Hence the delays.

 

[GodisanAtheist]

NVIDIA 下代 7 奈米製程產品，黃仁勳：台積電仍會取得大多數訂單

南韓媒體曾報導，南韓 NVIDIA 高層表示，公司新一代 GPU 繪圖晶片將轉單給南韓三星，並採用內含極紫外光刻技術（EUV）的 7 奈米製程技術打造，捨去原本長期合作對象台積電。NVIDIA 創辦人黃仁勳日前接受媒體訪問時澄清，未來還是會將大多數 7 奈米製程產品訂單交由台積電生產，三星只會獲得少...

finance.technews.tw

Maybe confirmation that Ampere is mostly being fabbed at TSMC, presumably 7+. Likely this is because Samsung's yields are not great and they had to go back to TSMC. Hence the delays.

-Ouch, those big 12nm dies are still going to be big 7nm dies. Samsung might still get the high volume mid-low market products, but given the demand for 7nm I wonder how much inventory NV will be able to build up for the Ampere line.

At least AMD has the excuse of having Zem Chiplets to manufacture (and being the second option in a market with two options as far as GPUs are concerned).

Guess shortages and high prices will be the norm for a while.

 

[Ajay]

Guess shortages and high prices will be the norm for a while.

Yup, good luck NV on getting a good wafer allocation from TSMC. Also, loss of a premier semiconductor company for Samsung - seems like they can’t get their act together. Not good for consumers.

 

[Bouowmx]

Are there specific issues with Samsung 7 nm, or is it on EUV tool availability/reliability?

 

[jpiniero]

Are there specific issues with Samsung 7 nm, or is it on EUV tool availability/reliability?

Says yields are bad.

 

[RetroZombie]

NVIDIA 下代 7 奈米製程產品，黃仁勳：台積電仍會取得大多數訂單

南韓媒體曾報導，南韓 NVIDIA 高層表示，公司新一代 GPU 繪圖晶片將轉單給南韓三星，並採用內含極紫外光刻技術（EUV）的 7 奈米製程技術打造，捨去原本長期合作對象台積電。NVIDIA 創辦人黃仁勳日前接受媒體訪問時澄清，未來還是會將大多數 7 奈米製程產品訂單交由台積電生產，三星只會獲得少...

finance.technews.tw

Maybe confirmation that Ampere is mostly being fabbed at TSMC, presumably 7+. Likely this is because Samsung's yields are not great and they had to go back to TSMC. Hence the delays.

Why did they stupidly leave tsmc in the first place?
What big chip dies did samsung manufactured in the past?

 

[Dribble]

Why did they stupidly leave tsmc in the first place?
What big chip dies did samsung manufactured in the past?

Well they couldn't have done the current 16xx and 2xxx at TSMC - being as they are huge chips, Nvidia need a lot of them (a lot more then AMD) and TSMC are very busy. There current solution has meant they are not supply limited which has been a huge advantage (e.g. probably why AMD are only just starting to release mid range cards).

 

[jpiniero]

Why did they stupidly leave tsmc in the first place?
What big chip dies did samsung manufactured in the past?

Samsung offered them a good deal and the node is slightly more dense than TSMC 7. Plus they had EUV started up earlier.

 

[Bouowmx]

Tom's Hardware: starting the hype train?

Nvidia Ampere Purportedly 50% Faster Than Turing At Half The Power Consumption

Apparently Ampere is more than ready for Big Navi.

www.tomshardware.com

Original: http://www.taipeitimes.com/News/biz/archives/2020/01/02/2003728557
Note: This is based on a "client note" from a investment consulting group.

 

[Ajay]

Tom's Hardware: starting the hype train?

Nvidia Ampere Purportedly 50% Faster Than Turing At Half The Power Consumption

Apparently Ampere is more than ready for Big Navi.

www.tomshardware.com

Original: http://www.taipeitimes.com/News/biz/archives/2020/01/02/2003728557
Note: This is based on a "client note" from a investment consulting group.

Please. 3 times the perf per watt of Turing?! Ludicrous.

 

[IntelUser2000]

Please. 3 times the perf per watt of Turing?! Ludicrous.

Same trick manufacturers claim for their process.

"20% faster transistors AND 30% lower power"

Really means

20% faster transistors at same power AND configure them to be 30% lower power at same performance.

 

[EXCellR8]

...and 100% more expensive.

(is the going joke)

 

[happy medium]

NVIDIA Will Be Announcing 7nm Ampere GPUs At GTC 2020 In March

NVIDIA Will Be Announcing 7nm Ampere GPUs At GTC 2020 In March

So I received a tip-off from a source that we have not used before and we will be tagging this as a rumor for now. While the source is new, it has proven to be right at least once in the past and we consider this worth publishing at this time. Our source states that NVIDIA is going to be...

wccftech.com

looks like pro cards in March and consumer cards in June/July.

 

[Glo.]

Ampere is HPC chip, that replaces GV100.

Don't expect anything consumer from Nvidia before at least late 2020, and realistically - 2021.

 

[SteveGrabowski]

Ampere is HPC chip, that replaces GV100.

Don't expect anything consumer from Nvidia before at least late 2020, and realistically - 2021.

God that would be disappointing to see another year of Turing. But I have been expecting nothing but disappointment on the gpu front the last couple of years and so far it has delivered there, so it would make sense.

 

[Ajay]

Ampere is HPC chip, that replaces GV100.

Don't expect anything consumer from Nvidia before at least late 2020, and realistically - 2021.

Makes sense that the GV100 would would be updated first. I imagine that's why NV is using TSMC - they need a fab used to pushing to the reticle limit. Maybe consumer cards are being produced on Samsung. Curious that another code name hasn’t emerged. Maybe, for a change, NV is using the same code name.