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.

 

[Glo.]

The thing I hate most about adored is his clear bias towards AMD which is most abundantly clear with his failure to call out AMD when they do something anti consumer but then accuse Nvidia of raising prices.

No clearer example of this is calling out GK104 for raising consumer prices, without calling Tahiti and Pitcairns first. These cards much like Turing kept price to performance vs last gen flat and actually had a negative price to performance vs last gen.

Remember 28nm was not really an expensive node as this was prior to finfet and AMD charged $350 for a 213mm2 die and $550 for a 351mm2 die. This was without Nvidia setting prices first(so you can't use that excuse). It was AMD initial 28nm pricing which set off the price escalation charging big money for small dies.

More thing he fails to call out is when AMD does a silent spec change like using a cut down variant of a chip, while keeping the name the same with no memory difference or pricing difference which means only AMD reaps the benefit. Yet he will call out something like the 3gb variant of the 1060, which atleast provides some cost savings and an easier way parts across.

AMD Silently Cuts Down Core Count on Radeon RX 560 Graphics Card - Now Shipping With 896 Cores Instead of 1024 SPs

AMD has silently introduced a new variant of the Polaris 11 based Radeon RX 560 which cuts down the core count.

wccftech.com

I would say one of the worst way AMD hurts even knowledgeable consumers is using viral and guerrilla marketing channels to get people to wait months to years for their products with fake leaks(fake release dates, extreme performance leaks) like they did with Fiji and Vega to stall potential buyers. And when these products come to light with similar price to performance as existing nvidia cards that have been on the shelf for months or over a year, they drum out the guerrilla marketing team to add rumors like Vega create a loss and can't be priced lower, nvidia is evil.

The funny thing is Adored finally caught on to this last point and semi terminated his relationship with AMD as he realized he was being used as a Guerilla marketing machine for AMD. He was pissed off he wasn't getting paid and while other youtubers were. It was in the video below.

https://www.reddit.com/r/realAMD/comments/cgsyhd

This video has since been made private or deleted and I suspect AMD has contacted him and decided to give him what he wanted.

Would you hate him equally if he would have equal, in your view, Bias towards AMD, and bash on AMD for everything and praise Nvidia for everything?

 

[Konan]

Kopite's latest....

https://twitter.com/x/status/1267293095873220609

Interesting as GDDR6x has not been announced yet.

 

[jpiniero]

If you compare Ti to Ti, that's only 14% more shaders. 1080 Ti -> 2080 Ti was 21%. Have to think it is SS8 then.

 

[AtenRa]

First images

NVIDIA GeForce RTX 3080 pictured? - VideoCardz.com

NVIDIA GeForce RTX 3080 pictured? Alleged pictures of the upcoming GeForce RTX 3080 graphics cards have been leaked on Chiphell. The models photographed by a leaker are clearly prototypes or engineering samples likely to be one of the many suggested designs for the final choice. The design does...

videocardz.com

 

[Krteq]

That's a great joke/fake

All those fins and fan have no purpose at all... there is no airflow allowed, fan blades are melted to a frame etc.

Just some joker with 3D printer

 

[Glo.]

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...

 

[uzzi38]

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...

If anything it lines up with what he said months ago. The PCB is an "irregular" shape.

Actually seems possible that consumers could get a similar shaped PCB to A100.

 

[maddie]

If anything it lines up with what he said months ago. The PCB is an "irregular" shape.

Actually seems possible that consumers could get a similar shaped PCB to A100.

HBM?

That's a great joke/fake

All those fins and fan have no purpose at all... there is no airflow allowed, fan blades are melted to a frame etc.

Just some joker with 3D printer

Push-push on opposite sides? WHY????

Only HBM can fit such a PCB.

 

[Glo.]

If anything it lines up with what he said months ago. The PCB is an "irregular" shape.

Actually seems possible that consumers could get a similar shaped PCB to A100.

Now, we have to wait for...

Double 8 pin connectors on the back of the GPU.

Those GPUs will use quite a lot of power.

 

[uzzi38]

HBM?

Push-push on opposite sides? WHY????

Only HBM can fit such a PCB.

The PCB size should be fine for GDDR6, I don't see an issue there. It's about ITX GPU size I think

 

[maddie]

The PCB size should be fine for GDDR6, I don't see an issue there. It's about ITX GPU size I think

The fans and outlets are on opposite sides. Both fans appear to pull air into the card body and exhaust directly through to the opposite side. A PCB extending fully to the support bracket will block one fan. Reconcile this.

 

[uzzi38]

The fans and outlets are on opposite sides. Both fans appear to pull air into the card body and exhaust directly through to the opposite side. A PCB extending fully to the support bracket will block one fan. Reconcile this.

I'm not sure I follow.



This image is the back of the GPU. The PCB is on the left. The fan is on the right.



This is the front of th GPU. The fan is now directly above the PCB next to the PCIe bracket... wait a second, that's a cutout for two power connector's on the bottom right! How in the.... that makes NO sense

 

[IntelUser2000]

@maddie

Man I'm tired of all these leaks from the CPU and GPU side. I'd like to see some of them released just to see which leakers are lying and which are genuine.

Just some joker with 3D printer

@uzzi38 You sure that's the back of the GPU? The PCI Express connector seems not flush with the backplate if that's indeed the back. There seems to be a cm or so distance between the connector and the front of the picture indicating its the front of the card.

Can you tell me why the PCB can't extend the length of the bracket?

Welcome - Canada Computers & Electronics

Canadacomputers.com offers the best prices on Computers, Computer Parts, Laptops, Hard Drives, PC Hardware & Accessories with fast shipping and top-rated customer service.

www.canadacomputers.com

Clearly the picture above shows the PCB can run full length of the HSF bracket.

 

[jpiniero]

Now, we have to wait for...

Double 8 pin connectors on the back of the GPU.

Those GPUs will use quite a lot of power.

The 2080 Ti FE is dual 8-pin, so it shouldn't be a surprise they used double pins. I could see them just using dual 8-pin on the GA102 models just to save time.

The 2080 Super FE is 8+6 btw.

 

[IntelUser2000]

The 2080 Ti FE is dual 8-pin, so it shouldn't be a surprise they used double pins. I could see them just using dual 8-pin on the GA102 models just to save time.

I know there are RX 580's with 8+8 configurations. It's not a guarantee that it'll use that much power.

 

[uzzi38]

@maddie

Man I'm tired of all these leaks from the CPU and GPU side. I'd like to see some of them released just to see which leakers are lying and which are genuine.



@uzzi38 You sure that's the back of the GPU? The PCI Express connector seems not flush with the backplate if that's indeed the back. There seems to be a cm or so distance between the connector and the front/back/whatever indicating its the front of the card.

Can you tell me why the PCB can't extend the length of the bracket?

Welcome - Canada Computers & Electronics

Canadacomputers.com offers the best prices on Computers, Computer Parts, Laptops, Hard Drives, PC Hardware & Accessories with fast shipping and top-rated customer service.

www.canadacomputers.com

Clearly the picture above shows the PCB can run full length of the HSF bracket.

Here's a mockup of what's shown here. Yes it's a disaster for ITX systems.



Only thing is the power connectors should be in the top left hand corner by the PCB.

No, maybe I'm overstimating the size of the gap. Could be the spot for the SLI bridge.

 

[IntelUser2000]

Damnit, I see it now.

That's very interesting.

wait a second, that's a cutout for two power connector's on the bottom right! How in the.... that makes NO sense

No reason they can't route the connectors using wires to that side.

 

[uzzi38]

@uzzi38 Let me clarify.

View attachment 22395

It may be a matter of different perspective, but it seems to me there's a cm or so of a gap, which tells me its the front of the card, not back. Otherwise the connector would be pretty much be flush with the shroud.

Hm, it does look like the top. Ugh, I dunno then.

 

[IntelUser2000]

Hm, it does look like the top. Ugh, I dunno then.

I edited my post. Looking at the picture from videocardz without the yellow PCB edit does change the perspective. It does look like its the rear.

So it seems with all the hardware conventions gone, manufacturers are going to hide the info until the day of the launch.

@Krteq By the way, I have a reference RX 480. The fins are nothing more than fluff. It's actually plastic and the real heatsink is hidden beneath it and it doesn't look like it can handle lot of power, hence the card sounding like a jet engine despite being 150W.

 

[maddie]

@maddie

Man I'm tired of all these leaks from the CPU and GPU side. I'd like to see some of them released just to see which leakers are lying and which are genuine.



@uzzi38 You sure that's the back of the GPU? The PCI Express connector seems not flush with the backplate if that's indeed the back. There seems to be a cm or so distance between the connector and the front of the picture indicating its the front of the card.

Can you tell me why the PCB can't extend the length of the bracket?

Welcome - Canada Computers & Electronics

Canadacomputers.com offers the best prices on Computers, Computer Parts, Laptops, Hard Drives, PC Hardware & Accessories with fast shipping and top-rated customer service.

www.canadacomputers.com

Clearly the picture above shows the PCB can run full length of the HSF bracket.

I was thinking of the fans blowing straight through. Apparently this is wrong. Fan closest to bracket exhausts through bracket. Other straight through?

 

[IntelUser2000]

I was thinking of the fans blowing straight through. Apparently this is wrong. Fan closest to bracket exhausts through bracket. Other straight through?

I don't understand it either.



Like what the hell is that? The rear fan will vent straight out of the front fins. If that was enclosed at least the airflow could work in a push-pull configuration.

 

[Tup3x]

That's certainly looks interesting... PCB and cooler design.

 

[IntelUser2000]

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.

 

[Mopetar]

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

Could be a lot of reasons. The first is that if they’re making a really big die they want to have a part they can sell heavily binned chips from. It would be interesting if the 104 die could technically have been used if it came out perfect because it means that as the process matures they could eventually transition, especially if they do something like the SUPER refreshes again.

They probably also don’t want to sleep on AMD having something more competitive for a change. Using the biggest die for an xx80 gives them more flexibility. If they need more performance to better position the product they can likely elect to leave more hardware enabled at the expense of a few more chips that can’t make the bin. If they’re already using a whole 104 die the only thing they can do is increase clock speeds.

 

[maddie]

I don't understand it either.

View attachment 22396

Like what the hell is that? The rear fan will vent straight out of the front fins. If that was enclosed at least the airflow could work in a push-pull configuration.

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