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.

 

[IntelUser2000]

I discarded the push-pull as the blades in both fans appear angled to push air into the card

Either its a 3D-printed joke as Krteq says or they'll come out with a press release saying its the most revolutionary HSF design ever but with regular aftermarket cards outperforming it in both noise and temperatures.

 

[jpiniero]

I looked back at kopite's tweets. He seems to be fairly accurate. His leaks for SUPER series are correct.

If he says its Samsung 10nm(ahem... 8nm) then it could be correct.

The increase in SMs are a bit disappointing though. 10nm still brings density gains so maybe the power efficiency gains aren't as much? Though relatively small details can change things. 14% more SMs, plus a 2GHz Boost, plus 10-15% more perf/clock means 40-50% faster.

I'm sure nVidia wanted to get the die size down from the 754 mm2 that TU102 is. 600 mm2 would be 20% smaller.

 

[IntelUser2000]

I'm sure nVidia wanted to get the die size down from the 754 mm2 that TU102 is. 600 mm2 would be 20% smaller.

I still think if they didn't have RT or Tensor cores they could have increased SMs by quite a bit.

You can't push such new standards without being a dominant force in the market. It also puts the leading company further in the driver seat because they have started the whole thing.

 

[reb0rn]

That small PCB large die would make PCB very complex with gddr6 or easy to make with HBM2 as it save PCB space

 

[Glo.]

So, if Nvidia uses 102 die for 3080 SKU, then wouldn't it be logical for them to move SKUs and Dies this way:
RTX 3070 - 104 die, RTX 3060 Ti - 104 die, RTX 3060 - 106 die, RTX 3050 Ti - 106 die, RTX 3050 - 107 die.

Makes sense?

 

[uzzi38]

I looked back at kopite's tweets. He seems to be fairly accurate. His leaks for SUPER series are correct.

If he says its Samsung 10nm(ahem... 8nm) then it could be correct.

The increase in SMs are a bit disappointing though. 10nm still brings density gains so maybe the power efficiency gains aren't as much? Though relatively small details can change things. 14% more SMs, plus a 2GHz Boost, plus 10-15% more perf/clock means 40-50% faster.

Yeah, Kopite has a solid track record. Leaked the SUPER series the day after they were teased (a few months before launch) and GA100 specs a year early with A100 specs 3 months prior to launch.

Only Sharkbay and Momomo I'd say have as solid a track record at the moment for any of the three major companies. Not including the benchmark and driver sniffers

 

[mopardude87]

My body is still ready for Ampere, come on with it already Nvidia. Stop being such a tease I assume its still a October/Nov/December release? I hear it should in theory be 60% faster then my 1080ti at least . I hope so.

 

[amenx]

.. particularly as the guy is a strong AMD fan...

He did appear that way in the earlier days of his channel. But aware of rising criticism and perception of that which may affect his cred, he then appeared to be more 'balanced' and even very critical of AMD. Occasionally he does come out with solid stuff. He was the first guy to accurately leak Nvidias switch from GTX to RTX name scheme for Turing, so he does have industry contacts that may occasionally deliver.

 

[MrTeal]

I would guess the rear flow-through fan and heatsink would be heatpiped to the central area with the GPU on it. Sapphire does that with their triple fan cards, though they mount all the fans on the same side like sane people.

 

[DiogoDX]

https://twitter.com/x/status/1269239881512873984

Also, considering Kopite says it might not be fake, I would not disregard this "rumor" as fake or joke .

P.S. I wonder why Nvidia is releasing 3080 SKU based on 102 die... what could it possibly be the reason behind it...

Last time was with the GTX780. And was also the last time AMD had a good high end lineup with the 290X and 290.

 

[MrTeal]

Last time was with the GTX780. And was also the last time AMD had a good high end lineup with the 290X and 290.

True, but that was also a little different. The 680 was GK104 in 2012 against Tahiti, and GK110 was straight to datacenters and supercomputers. They didn't have a new GPU in 2013 to respond to Hawaii so they gave a really cut down version of the big die as a 780.

You could say the 480 was the last time they launched a new gen with a big die 480, but that's not really the same. The x80/x800 then was the biggest and best, everyone now know that there will be a Ti/Titan above it.

 

[Stuka87]

Man, I don't know about that card. I am looking at my case here, and on my X570 board, that backside fan would be pulling hot air from the ram. I am just not sure what would be gained from having it blow through air in the opposite direction.

But the blow through designs are cool, the Fury cards certainly benefited from it. But they had HBM, how would they get a GDDR6 board small enough to do it?

 

[DrMrLordX]

@Stuka87

The air around your RAM shouldn't be that hot.

 

[Tup3x]

Man, I don't know about that card. I am looking at my case here, and on my X570 board, that backside fan would be pulling hot air from the ram. I am just not sure what would be gained from having it blow through air in the opposite direction.

But the blow through designs are cool, the Fury cards certainly benefited from it. But they had HBM, how would they get a GDDR6 board small enough to do it?

It would logical to push the hot air up... Doesn't seem optimal at least in standard vertical PC cases. Hot air exhausts are in opposite direction and the hot air would rise back to card. It would end up recycling the hot air.

 

[amenx]

I'll bet this is it. Its way too original to be made up by the usual leakers.

close up:
https://i.redd.it/kkt8or0fod351.png

I believe the cover is comprised of fin slats all over.

 

[FaaR]

I am just not sure what would be gained from having it blow through air in the opposite direction.

Seems an odd design choice, maybe made simply for sake of infamy. As cards are typically mounted in PC towers, the rear side fan pushes air down into the chassis, whereas most PC ventilation schemes are set up to evacuate air at the top of the chassis. Seems inefficient to have it this way, especially as the top side fan near the bracket will undoubtedly ingest hot air from the rear side fan...

Maybe there's some secret Nvidia sauce I'm missing here, but damned if I know what it might be.

But the blow through designs are cool, the Fury cards certainly benefited from it. But they had HBM, how would they get a GDDR6 board small enough to do it?

If this card uses the same kind of compact VRMs seen on the GA100 mezzanine boards there would be plenty space I'd think.

 

[GodisanAtheist]

Hopefully this does not mean performance numbers are not where NV wants them to be, and so we're going to get a lot of gimmicky crap to try and sell the card.

I doubt it, NV is nothing if not proficient at execution, but you never know.

 

[DisEnchantment]

I looked back at kopite's tweets. He seems to be fairly accurate. His leaks for SUPER series are correct.

If he says its Samsung 10nm(ahem... 8nm) then it could be correct.

The increase in SMs are a bit disappointing though. 10nm still brings density gains so maybe the power efficiency gains aren't as much? Though relatively small details can change things. 14% more SMs, plus a 2GHz Boost, plus 10-15% more perf/clock means 40-50% faster.

RTX3090(?), 5248, GA102-300, 21Gbps GDDR6X

Not really sure about GDDR6X, there is no such JEDEC spec.
Most likely G6 16-18 Gbps, factory overclocked probably.

 

[.vodka]

I'll bet this is it. Its way too original to be made up by the usual leakers.
I believe the cover is comprised of fin slats all over.

What the F is that. That's the only thought I get looking at these pictures and posts a page or two back.

What's up with that array of fins in the middle not being actively cooled blocking everything? The fan configuration makes no sense. There's no wind tunnel to direct air from one fan to the other as the wireframe drawing one page back seems to imply. Even if this is an HBM card with a small PCB like the Fury X, the heatsink surface area is lacking compared to the usual blower or three fan open air heatsink.

How is this more efficient than a big heatsink and three fans as usual? If you're gonna heat up the case, then at least do it the right way I guess.

Either they've gone insane and pulled another FX5800 level cooler out their rear end, or this is some alien cooling tech that's gonna leave us all looking for our jaws on the floor. No middle ground here.

 

[amenx]

What the F is that. That's the only thought I get looking at these pictures and posts a page or two back.

What's up with that array of fins in the middle not being actively cooled blocking everything? The fan configuration makes no sense. There's no wind tunnel to direct air from one fan to the other as the wireframe drawing one page back seems to imply. Even if this is an HBM card with a small PCB like the Fury X, the heatsink surface area is lacking compared to the usual blower or three fan open air heatsink.

How is this more efficient than a big heatsink and three fans as usual? If you're gonna heat up the case, then at least do it the right way I guess.

Either they've gone insane and pulled another FX5800 level cooler out their rear end, or this is some alien cooling tech that's gonna leave us all looking for our jaws on the floor. No middle ground here.

I think it makes a lot of sense. Traditional ref coolers exhausted all the hot air from one end. Meaning that hot air was passing over the fins on its exit path. With the new design the hot air is expelled directly through the fins over a much larger area thus keeping the heat pipes cooler. Yes the hot air is being dumped into the case, bu in decent cases with good airflow I dont see that as much of a problem.

 

[Stuka87]

I think it makes a lot of sense. Traditional ref coolers exhausted all the hot air from one end. Meaning that hot air was passing over the fins on its exit path. With the new design the hot air is expelled directly through the fins over a much larger area thus keeping the heat pipes cooler. Yes the hot air is being dumped into the case, bu in decent cases with good airflow I dont see that as much of a problem.

There have been other cards with pass through coolers (Fury cards). The very strange part on this one is the opposite air flow direction.

 

[darkswordsman17]

There have been other cards with pass through coolers (Fury cards). The very strange part on this one is the opposite air flow direction.

Those were quite different. They weren't pushing air over a heatsink, they just had openings on the backside that enabled the blower fan to pull air from both the back and front of the card. Nvidia had some of those (600 series?) as well I believe.

This seems to have actual heatsink fans extended out and the fan is blowing air over them. Its a weird design for sure. Its also weird that it'd be sucking air down through the card, since many cases have fans on the bottom blowing air upwards. It'll be interesting to see how it performs. Perhaps its simply that they know many enthusiasts pass on reference designs if possible, so they want a memorable looking design such that its more of a marketing thing when most people won't suffer the drawbacks of the design. Plus its probably not any worse than single blower designs.

Which did anyone notice the fans themselves? Perhaps they're already like that on current cards, but if you look, the fans are like tubes (meaning the blades are connected by a ring on the edges). I'd think that'd be a bad design for the one nearest the outlet.

 

[Glo.]

https://twitter.com/x/status/1269909686674808832

And here, Kopite says that power draw will acttually be between 300 and 375W.

Considering the die configurations of 102 chip, I would say that he means that RTX 3080 is going to use 300W of power, and RTX 3090 will(or something with full die) will use 375W of power.

 

[AtenRa]

Mechanically the design is ok, I believe there is a Vapour Chamber or dual chambers beneath each fan with the heat fins going all the way up to the height of the fan.
Other wise I dont see any problems with the fan configuration.

 

[Stuka87]

Those were quite different. They weren't pushing air over a heatsink, they just had openings on the backside that enabled the blower fan to pull air from both the back and front of the card. Nvidia had some of those (600 series?) as well I believe.

This seems to have actual heatsink fans extended out and the fan is blowing air over them. Its a weird design for sure. Its also weird that it'd be sucking air down through the card, since many cases have fans on the bottom blowing air upwards. It'll be interesting to see how it performs. Perhaps its simply that they know many enthusiasts pass on reference designs if possible, so they want a memorable looking design such that its more of a marketing thing when most people won't suffer the drawbacks of the design. Plus its probably not any worse than single blower designs.

Which did anyone notice the fans themselves? Perhaps they're already like that on current cards, but if you look, the fans are like tubes (meaning the blades are connected by a ring on the edges). I'd think that'd be a bad design for the one nearest the outlet.

You are thinking of AMD's blower design. I am talking the after market cards. Such as this Sapphire which has the third fan blow clear through the heat sink. It proved to be very effective. But was only possible because the card used HBM, so the board was super tiny for a 300W card.