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.

 

[Bouowmx]

ZOTAC GeForce RTX 3090, GeForce RTX 3080, GeForce RTX 3070 Custom Graphics Cards Pictured

The first pictures of ZOTAC's GeForce RTX 30 graphics cards which include the GeForce RTX 3090, RTX 3080, and the RTX 3070 have leaked out.

wccftech.com

The ZOTAC TRINITY: 3 slots, 2x8-pin (?), 19+1 inductors, very closely spaced memory. Finally, no more VirtualLink (for those liquid cooling and want a single-slot aesthetic).

 

[Konan]

Looking at the number sequences makes me believe it's fake.

Looks about 18TF with those 3080 numbers

Ahhh the memory and GPU clock are crazy high

 

[Glo.]

Looking at the number sequences makes me believe it's fake.

Its not fake. Zotac leaked quite a lot of RTX 3000 series stuff, already.

HOWEVER. If people will want to see the truth there, and not just confirm their own Biases, they will see GPU base clock: 1234 MHz, and Boost clock at 4321 MHz .

So this is just an image of an app, without any real indication of 3000 series specs.

 

[DiogoDX]

350W and 320W LOL

Since is Nvidia everything will be OK. Reviews will only talk about performance like the GTX480.

 

[Kenmitch]

350W and 320W LOL

Since is Nvidia everything will be OK. Reviews will only talk about performance like the GTX480.

Bumpgate 2.0 video?

 

[Kenmitch]

HOWEVER. If people will want to see the truth there, and not just confirm their own Biases, they will see GPU base clock: 1234 MHz, and Boost clock at 4321 MHz .

Wow 4321MHz boost! You think there will be any OC room left? AMD is doomed for sure! I saw it on the internet so it's got to be true! And only 350w's to top it off. /s

 

[CakeMonster]

There seems to be at least one 100% confirmed screwup here, the memory size. They just couldn't get to a bus width that supported the right amount of memory for any of the models, so suddenly twice the amount has to be made available. And neither the original or double amount seem to be a good size for this generation, with the original being slightly too little for each price point/model, while the double being obviously way too much.

 

[Krteq]

Looking at the number sequences makes me believe it's fake.

Return of pre-Fermi separated clock domains?

//nvm, just some placeholders/dummy numbers - 1234, 4321, 2345 etc.

 

[Saylick]

Wow 4321MHz boost! You think there will be any OC room left? AMD is doomed for sure! I saw it on the internet so it's got to be true! And only 350w's to top it off. /s

I'm getting GTX1080 launch flashbacks here where JHH asked to show the clocks.

JHH: "Just for kicks, let's look at the clocks. It's up there. Can you guys see it? WoooOoOooah, I know. It's insane. Four thousand megahertz! And it's on air and only 350W! Only 350W. It's ridiculous, just absolutely irresponsible amounts of power."
Audience: *cringy woops and hollers*

 

[Stuka87]

There seems to be at least one 100% confirmed screwup here, the memory size. They just couldn't get to a bus width that supported the right amount of memory for any of the models, so suddenly twice the amount has to be made available. And neither the original or double amount seem to be a good size for this generation, with the original being slightly too little for each price point/model, while the double being obviously way too much.

With GDDR6X they are limited to 8Gb (1GB) modules. But they also needed a wide bus for bandwidth. Which ended up being 320bit, which means you need 10 modules, which then means you get 10GB or 20GB.

EDIT for clarity: To get 20GB, it looks like they are using 20 modules (with 10 on the backside per leaks). This is done with a bus split resulting in 16bits going to each module, instead of 32bits. This retains the overall bandwidth, but does technically lower the bandwidth per module. I suppose there are some specific use cases where this could hurt performance, but generally would not be an issue.

 

[GodisanAtheist]

Man, the days leading up to a GPU announcement. Only happens every couple of years.

In this thread you can see enthusiasts in their natural environment mainlining adrenaline and riding this transient high while they can.

This thread is a circus and I love it.

I don't know how I'm going to survive the next few days, I have the shakes already

 

[Fanatical Meat]

Man, the days leading up to a GPU announcement. Only happens every couple of years.

In this thread you can see enthusiasts in their natural environment mainlining adrenaline and riding this transient high while they can.

This thread is a circus and I love it.

I don't know how I'm going to survive the next few days, I have the shakes already

Yeah I have been completely disappointed with graphics cards since 2012ish. Nothing revolutionary except on the uber high end and costs have gone up. $1k plus for the top end is insane.

 

[Stuka87]

Man, the days leading up to a GPU announcement. Only happens every couple of years.

In this thread you can see enthusiasts in their natural environment mainlining adrenaline and riding this transient high while they can.

This thread is a circus and I love it.

I don't know how I'm going to survive the next few days, I have the shakes already

I always like seeing what rumors were true, and which were not. And of course, the reactions of certain people trying to spin rumors

 

[nurturedhate]

I always like seeing what rumors were true, and which were not. And of course, the reactions of certain people trying to spin rumors

Like watching daytime soap operas for the comedy.

 

[DDH]

I really hope AMD have something impressive to showcase that at *least* offers buyers considering the 3090 a choice.

All i want is Cyberpunk 2077, 3440x1440, ultra with all the ray tracing, touching 90fps. Is that too much to ask?

Sent from my SM-N975F using Tapatalk

 

[Stuka87]

I really hope AMD have something impressive to showcase that at *least* offers buyers considering the 3090 a choice.

All i want is Cyberpunk 2077, 3440x1440, ultra with all the ray tracing, touching 90fps. Is that too much to ask?

Sent from my SM-N975F using Tapatalk

Going off what CDPR has said, most likely yes. Maybe top tier cards will do it. But I would have a rough time paying $1500 on a GPU for one game.

 

[DDH]

Going off what CDPR has said, most likely yes. Maybe top tier cards will do it. But I would have a rough time paying $1500 on a GPU for one game.

One game to start with anyway. But if it lasts 3 years and delivers top performance for that time......


Sent from my SM-N975F using Tapatalk

 

[Fanatical Meat]

Going off what CDPR has said, most likely yes. Maybe top tier cards will do it. But I would have a rough time paying $1500 on a GPU for one game.

Haha was just going to say that will cost about $1800.00

 

[Glo.]

The good news with this generation is that we will get decent sub 199$(bottom lurker tier, for those companies, I guess) GPUs.

Be it in the form of RTX 3050/3050 Ti, RX 5600 XT being price cut to this level in order to compete with 199-250$ Ampere GPUs, or with RDNA2 based GPU with this price level.

Listen companies. I will be happy with RTX 2060/RX 5600 XT performance for this price. This would be FINALLY after all of those years, decent performance/dollar uplift.

 

[tviceman]

Even so, we got way more from 980Ti to RTX2080Ti with only 20% more power consumption than we are getting from 2080Ti to RTX3090 with 40% increase of power.

Not only that, but if RTX3090 is 60% faster vs RTX2080Ti in Raster performance but with a 40% increase of power, then they only managed a 15% increase of perf/watt from Turing to Ampere ????

with a full node ???? wtf

Exactly. This makes no sense.

 

[tviceman]

duplicate

 

[linkgoron]

I really hope AMD have something impressive to showcase that at *least* offers buyers considering the 3090 a choice.

All i want is Cyberpunk 2077, 3440x1440, ultra with all the ray tracing, touching 90fps. Is that too much to ask?

Sent from my SM-N975F using Tapatalk

I'm skeptical that AMD will have something competitive with the 3090 (I hope I'm wrong). However, I assume that AMD will deliver something stronger than the 3080. IMO, there are only two reasons for Nvidia to release both a 3080 and 3090 at the same time. Either the 3080 is rather weak, or AMD is competitive. I expect some decent performance uplift from the 3080, so I can only assume that AMD will actually have something better than the 3080 once they release RDNA2.

 

[jpiniero]

I'm skeptical that AMD will have something competitive with the 3090 (I hope I'm wrong). However, I assume that AMD will deliver something stronger than the 3080. IMO, there are only two reasons for Nvidia to release both a 3080 and 3090 at the same time.

May not be time to play around. I am kind of expecting the entire lineup to be released by the end of the year. Part of it is wanting to get new cards out in front of the new consoles; esp with the whole lineup having RT, part of it is the idea that they might refresh relatively quickly.

 

[lobz]

Yeah I have been completely disappointed with graphics cards since 2012ish. Nothing revolutionary except on the uber high end and costs have gone up. $1k plus for the top end is insane.

You know, AMD really screwed up with 14nm, Polaris was OK at most, and only because of its price. Pascal's clockspeeds and low power consumption on the other hand were a brutal surprise for everyone. No wonder the prices were insane immediately. I mean, after the $329 970 which was really good and insanely popular, the 1070 was a whopping $500... I still bought it because it was still the best in its price category. All in all, Pascal was a great generational jump. Compared to that, Turing is a dud, I don't care what anyone says. All that's coming from a 2070S owner

 

[Jaskalas]

The 12 pin is not bad, looks like PSU makers are onboard already.
The other question is what will Ampere use all that power for.

Gotta trace them rays and cook some eggs.