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.

 

[uzzi38]

I'm very sure 1700mhz is roughly where rated boost clock will land. GPU boost will go higher.

 

[reb0rn]

Can anyone speculate on bus width used for 3080 -> 3090 ?
do they need to push 512mbit on top and rest 384bit

 

[uzzi38]

Do you think AMD is going to dominate this market soon?

Hell no. AMD are still a small as hell fish in the consumer GPU market, there won't be any domination any time soon.

This is going to be their best, most competitive lineup in years, but there still won't be any domination. They'll take the lead next year when RDNA3 inevitably launches first. Still no domination. Even getting to 50% market share is one hell of an uphill battle.

 

[Karnak]

Can anyone speculate on bus width used for 3080 -> 3090 ?
do they need to push 512mbit on top and rest 384bit

It's all based on GA102 so I highly doubt they'll cut it down to 384bit all the way from 512bit.

The 3080 is most likely cut down to 320bit while the 3090/Titan/Ti (or whatever) will feature full 384bit.

 

[sontin]

I know. But i find it strange

Hell no. AMD are still a small as hell fish in the consumer GPU market, there won't be any domination any time soon.

This is going to be their best, most competitive lineup in years, but there still won't be any domination. They'll take the lead next year when RDNA3 inevitably launches first. Still no domination. Even getting to 50% market share is one hell of an uphill battle.

How do you know this? Maybe it will be their worst because unlike with Navi they wont have a two full node advantage to hide their architecture problems.

 

[Ajay]

Hell no. AMD are still a small as hell fish in the consumer GPU market, there won't be any domination any time soon.

This is going to be their best, most competitive lineup in years, but there still won't be any domination. They'll take the lead next year when RDNA3 inevitably launches first. Still no domination. Even getting to 50% market share is one hell of an uphill battle.

Yes. Seems like all the DIY sites and YouTube influencers run 2080 Ti's on their test systems. Nvidia and partners are very good at saturating the media space with halo products to give the illusion that ALL their products are superior to AMD. This is true even though AMD has very competitive GPUs in the mid to low tiers. Until AMD can gain mind share, they don't have a chance at gaining significant market-share gains.

 

[Konan]

https://twitter.com/x/status/1294552679444692992

Here he debunks the author of "some" rumors from previous pages.

Do not believe in them. Remember Blue Nugroho trashing PS5, claiming it is not RDNA2 architecture?

Fanboys exist in every side of the industry, and post stupidest BS.

some of the rumours yes, not all.
Personally, I don’t believe anything that I transferred over regarding AMD.

Couple of points/ my thoughts :
1. Everything FP 32 related was put on Baidu nearly 2 weeks before kopitekimi and KatCorgi said anything and agreed
2. TSE results were also there just under a week before they mentioned
3. The twitter guys say 3080 is only 20% better than a 2080ti and then shortly after the TSE results show, that based just off that, it throws up Conflicting posts between them

 

[Hitman928]

Hell no. AMD are still a small as hell fish in the consumer GPU market, there won't be any domination any time soon.

This is going to be their best, most competitive lineup in years, but there still won't be any domination. They'll take the lead next year when RDNA3 inevitably launches first. Still no domination. Even getting to 50% market share is one hell of an uphill battle.

The last market share update I saw had AMD at 31% for add on GPU sales. Obviously not dominant by any means, but I wouldn't consider that a small as hell player either, especially if you add in all PC GPU share where AMD and Nvidia are essentially tied.

AMD Radeon GPUs Gained Big Chunk Of Discrete Market Share Versus NVIDIA GeForce In Q4 2019 - Up From 27% To 31%

AMD Radeon GPUs grabbed a huge chunk of discrete GPU market share versus NVIDIA GPUs in Q4 2019, reports Jon Peddie Research.

wccftech.com

 

[uzzi38]

I know. But i find it strange


How do you know this? Maybe it will be their worst because unlike with Navi they wont have a two full node advantage to hide their architecture problems.

I don't know for 100% certain. I just have one hell of a lot of faith in Zen's physical optimisation team, a great breakdown and analysis of the power consumption of the Series X, and some knowledge of AMD doing what is right in terms of managing (formerly known as) RTG.

Oh and some other little bits and pieces, like MI100 CU count and clocks (spolier alert: the AdoredTV leak was accurate for once in everything short of how they presented the information, and they didn't mention power consumption) and Renoir power efficiency (sustains 1.75GHz on the iGPU at ~22W full SoC power, so including IMC and at least one CPU core active and under a light load).

I'm fairly confident AMD will have a very competitive chip on sale later this year. One with a similar shader count to top-most Ampere die and surprisingly similar clocks too. So while there's no way I can say for a fact that AMD will take the crown, I think they're in a position where they will be competing against a higher tier GPU in Nvidia's product stack than they did last gen.

 

[uzzi38]

The last market share update I saw had AMD at 31% for add on GPU sales. Obviously not dominant by any means, but I wouldn't consider that a small as hell player either, especially if you add in all PC GPU share where AMD and Nvidia are essentially tied.

AMD Radeon GPUs Gained Big Chunk Of Discrete Market Share Versus NVIDIA GeForce In Q4 2019 - Up From 27% To 31%

AMD Radeon GPUs grabbed a huge chunk of discrete GPU market share versus NVIDIA GPUs in Q4 2019, reports Jon Peddie Research.

wccftech.com

Pretty sure they dropped again in Q1 and Q2 didn't they?

 

[Hitman928]

Pretty sure they dropped again in Q1 and Q2 didn't they?

I don't know, I haven't seen any more recent numbers but I don't subscribe to the research companies either. Even if they did, I'm doubtful they would have dropped too much but if anyone has more recent numbers, I'd like to see them.

 

[uzzi38]

I don't know, I haven't seen any more recent numbers but I don't subscribe to the research companies either. Even if they did, I'm doubtful they would have dropped too much but if anyone has more recent numbers, I'd like to see them.

Don't think Q2 numbers are released yet but Q1 there was a drop to 25%

Global GPU market’s growth blunted by COVID-19 in Q1’20 reports Jon Peddie Research

PC GPU market shipments decreased -5.3% sequentially from last quarter but increased 14.9% year-to-year.

www.jonpeddie.com

 

[Hitman928]

Don't think Q2 numbers are released yet but Q1 there was a drop to 25%

Global GPU market’s growth blunted by COVID-19 in Q1’20 reports Jon Peddie Research

PC GPU market shipments decreased -5.3% sequentially from last quarter but increased 14.9% year-to-year.

www.jonpeddie.com

Ok, it took me a minute to reconcile those numbers from the 31% I linked to. The 31% share was desktop GPUs while your link is all discrete GPUs (so laptops included). In that sense AMD dropped from 27% to 25%. Even 25% I wouldn't call them a small fish, I mean, you still have 1/4 of the market. That's again not counting APUs which some unknown amount will be used for actual gaming, more often in laptops I would assume. Nvidia is still the dominant player in GPUs, but AMD also has a lot of weight numbers could change fairly rapidly of AMD were able to put out more competitive cards consistently.

 

[Konan]

Also from Baidu forums. Again I'm just copy//paste not my words. Make of what you will.
This poster said there may be errors but basically this is it.

Rumor / Speculation...
We'll see how close they come

NVIDIA Titan A
48 GB GDDR6X 5376CUDA FP32 36TF+ || 384bit 48 GB GDDR6X 192Rops
Equivalent performance is equivalent to Turing architecture 8064 level PS: 2G GDDR6X exclusive you know, the price is 2499 US dollars

NVIDIA GeForce RTX 3090
5120-5376CUDA FP32 34TF+ || 384bit 24 GB GDDR6X 192Rops
Equivalent performance is equivalent to the 7680 level PS of the Turing architecture: the price is 1999 US dollars.

NVIDIA GeForce RTX 3080
4608CUDA FP32 30TF+ || 320bit 20 GB GDDR6X 160Rops
Equivalent performance is equivalent to Turing architecture 6912 level PS: 2080Ti +30%. Price TBD

NVIDIA GeForce RTX 3070Ti
3072CUDA FP32 20TF+ || 256bit 16 GB GDDR6X 128Rops
Equivalent performance is equivalent to Turing architecture 4608 level PS: The full version of GA104 is better than 2080Ti || Equivalent cut down version TU102 by 20% or less

NVIDIA GeForce RTX 3070
2688CUDA FP32 15TF+ || 256bit 16 GB GDDR6X 128Rops
Equivalent performance is equivalent to Turing architecture 4032 level PS: basically second only to 2080Ti or the same

NVIDIA GeForce RTX 3060Ti
2304CUDA FP32 13TF+ || 192bit 12GB GDDR6X 96Rops
Equivalent performance is equivalent to Turing architecture 3456 level PS: this is OK GA104 cut version. Better than 2080 Super. Within 17%

NVIDIA GeForce RTX 3060
1920CUDA FP32 11TF+ || 192bit 12GB GDDR6X 96Rops
Equivalent performance is equivalent to Turing architecture 2880 level PS: Basically better than 1080Ti, the same as 2080 Super. Price is ~400-450 US dollars

NVIDIA GeForce RTX 3060 cut down version
1536CUDA FP32 9TF+ || 192bit 12GB GDDR6X 96Rops.
Equivalent performance is equivalent to Turing architecture 2304 level PS: basically it is 2070 but it is very cheap. Price is ~350-375 US dollars

NVIDIA GeForce RTX 3050Ti
1280CUDA 6-7TF+ || 128bit 8GB GDDR6 64Rops
Equivalent performance is equivalent to Turing architecture 1920 level PS: basically a 2060, support DLSS and RTX. To put it bluntly, it is 2060 within 200-250 US dollars

NVIDIA GeForce RTX 3050
1024CUDA 5-6TF+ || 128bit 8GB GDDR6 64Rops
Equivalent performance is equivalent to Turing architecture 1536 level PS: basically a 1660Ti price around 150-175 US dollars

Twitter leakers notes for reference

 

[CakeMonster]

The price premium for the Titan A over the 3090 seems surprisingly small.

 

[jpiniero]

If the 3080 had 30+ TF FP32 and 160 Rops I would expect it to be a lot faster than just 30%. Gonna have to say fake.

 

[Stuka87]

I REALLY doubt we will see all those Ti models at launch. If anything, those will come out as Super versions 6 months down the road.

Also doubt the 192 ROPs count. 2080Ti has 88, going to 192 seems like an utter waste. There is no way going to so many would offer any kind of benefit.

 

[IntelUser2000]

Rumor / Speculation...
We'll see how close they come

Equivalent performance is equivalent to Turing architecture 2304 level PS: basically it is 2070 but it is very cheap. Price is ~350-375 US dollars

LOL, $350 US is cheap?!?

Equivalent performance is equivalent to Turing architecture 3456 level PS: this is OK GA104 cut version. Better than 2080 Super. Within 17%

What's with quoting weirdly specific numbers? Why not add another decimal?

20-30% increase in perf/$ if that is correct. That points to yet another price increase but keeping names same to have gamers accept the new reality.

 

[Bouowmx]

2000 yuan (290 USD) for the lowest 3060. And under 3000 yuan (guess: 2500 yuan/360 USD?) for the next higher 3060. Nice: RIP me.
Disclaimer: I have no idea how China hardware prices work (ex. if they have VAT included like in Europe), or if they can be directly converted to USD.

 

[Bouowmx]

What's with quoting weirdly specific numbers? Why not add another decimal?

The weirdly specific numbers comes from the fact that the writer basically did 1 Ampere core = 1.5 Turing cores

 

[Konan]

What's with quoting weirdly specific numbers? Why not add another decimal?

20-30% increase in perf/$ if that is correct. That points to yet another price increase but keeping names same to have gamers accept the new reality.

I just copy and pasted what was posted From a translation.

 

[IntelUser2000]

I just copy and pasted what was posted From a translation.

I know, I'm not blaming you. It's like Intel saying 18% for Sunny Cove and AMD saying 52% for Zen. Really? Did everybody forget ranges exist and can be used?

Looks like Nvidia hasn't learned from Turing if the prices are true, but who knows?

The gap between RTX 3080 and 3090 seems way too large. The RTX 3080 performs 30% better than RTX 2080 Ti, but the RTX 3090 that's barely 10% better than RTX 3080 is $1999? So the RTX 3080 is $1299 for a zero perf/$ gain? Or are they going to insert a RTX 3080 Ti that's 5% faster but costs $250 more?

 

[reb0rn]

With those prices NV can take those cards and push them where sun do not shine.....
also TDP is insane...... just no tnx, good bye PC gaming let it die already

 

[GodisanAtheist]

I know, I'm not blaming you. It's like Intel saying 18% for Sunny Cove and AMD saying 52% for Zen. Really? Did everybody forget ranges exist and can be used?

Looks like Nvidia hasn't learned from Turing if the prices are true, but who knows?

The gap between RTX 3080 and 3090 seems way too large. The RTX 3080 performs 30% better than RTX 2080 Ti, but the RTX 3090 that's barely 10% better than RTX 3080 is $1999? So the RTX 3080 is $1299 for a zero perf/$ gain? Or are they going to insert a RTX 3080 Ti that's 5% faster but costs $250 more?

- I think it's safe to say that leaker is just some guy spitballin based on updated info.

That ROP count is absolutely outrageous unless NV is redefining what a ROP is or they plan on pushing 16k displays or something (VR maybe?)

 

[FaaR]

That ROP count is absolutely outrageous unless NV is redefining what a ROP is or they plan on pushing 16k displays or something (VR maybe?)

They're gonna need way more than a mere 1TB/s memory BW to effectively use 192 ROPs... Sounds like a fanboy making up his personal wet dream numbers to me.