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.

 

[Elfear]

Ok, first off "hot" just means inadequate cooling. It literally means nothing else. You can have a 10W chip run 'super hot'. The whole reason the RX 3090 has a gigantic cooler is so that it won't run hot while consuming 350W+.

Claiming Microsoft broke FTC law for a 'wow' factor is rather ludicrous.

And finally, you last posted in 2017 in posts attacking AMD and making claims that we now know were wrong, why come back now?

Alt account for Happy M?

 

[Maverick177]

Ok, first off "hot" just means inadequate cooling. It literally means nothing else. You can have a 10W chip run 'super hot'. The whole reason the RX 3090 has a gigantic cooler is so that it won't run hot while consuming 350W+.

Claiming Microsoft broke FTC law for a 'wow' factor is rather ludicrous.

And finally, you last posted in 2017 in posts attacking AMD and making claims that we now know were wrong, why come back now?

Who would want to use inadequate for early samples ? I never said Ampere didnt run hot, it does but that comes with the features and performance one might expect. RDNA2 on the other hand is still pretty much Raja product with MS/Sony added features.

Last time I was here I claimed Vega was a turd, how is that wrong ? In fact I owned a modded Vega 56 for 2 years for $200, best purchase ever. How is that even possible?

Back to the topic, GA102 is lapping Navi 2x as in tradition even on inferior node. This speaks volume about R&D, a simple fact that many posters here just won't adhere to every launch ever. AMD might be able to compete on equal footing with RDNA3, I might come back in summer 2022 to spread some more "fuds'.

 

[Kenmitch]

Alt account for <redacted>?

lol....Just remember don't type it 3 times.

 

[Maverick177]

lol....Just remember don't type it 3 times.

I can only assume this Happy M is a vastly superior being to everyone here.

 

[USER8000]

Alt account for Happy M?

Or rollo!

 

[maddie]

Who would want to use inadequate for early samples ? I never said Ampere didnt run hot, it does but that comes with the features and performance one might expect. RDNA2 on the other hand is still pretty much Raja product with MS/Sony added features.

Last time I was here I claimed Vega was a turd, how is that wrong ? In fact I owned a modded Vega 56 for 2 years for $200, best purchase ever. How is that even possible?

Back to the topic, GA102 is lapping Navi 2x as in tradition even on inferior node. This speaks volume about R&D, a simple fact that many posters here just won't adhere to every launch ever. AMD might be able to compete on equal footing with RDNA3, I might come back in summer 2022 to spread some more "fuds'.

You love turds? Wow.

 

[Glo.]

Oh lord.

KittyCorgi's 106 die and 107 die configs are actually correct O_O

So then: 36 SM SKU might be based on 104 die, instead of 106, and 24 SM die might be based on 106 die.

So then: 36 SM SKU is RTX 3060 Ti, based on 104 die(at least, it might have 256 bit bus...), and 24 SM SKU: RTX 3050 Ti, with potentially 6 GB's on 192 bit bus.

I very much like this idea.

 

[Glo.]

P.S. Don't feed the troll. Ignore him, and move on, guys.

 

[blckgrffn]

Or rollo!

That just blew a mental circuit breaker. I hadn’t thought of that guy in a really, really long time...

 

[Maverick177]

You love turds? Wow.

For $200 and that includes 2 Noctua black out fans and Morpheus cooler, I got a card than is quiet even with full load @ 65C. A deal of a lifetime if you ask me.

P.S. Don't feed the troll. Ignore him, and move on, guys.

I'll be back in summer 2022 for RDNA 2 vs Hopper. Right now Ampere as usual will crush RDNA 2 again, near and far future ( in tech world that is ). Any hopes left for RTG will be gone on September 28th.

 

[Maverick177]

P.S. Don't feed the troll. Ignore him, and move on, guys.

 

[USER8000]

That just blew a mental circuit breaker. I hadn’t thought of that guy in a really, really long time...

Yeah,he was also on uk forums also. Before fermi he was saying ati was rubbish,hd5870 would be ended by fermi,people would regret not waiting for it,etc. Nvidia making big hype before a launch,is saying something is amiss. They didn't make big hype before turing.

 

[Saylick]

Kimi confirmed it. The GA 104 core count too recently. Oh and at 220W

GA104 is 220W at 95% of the performance of a 2080 Ti? The TDP of a 2080 Ti is 260W, so that doesn't seem like all that big of a perf/W improvement...

 

[Konan]

GA104 is 220W at 95% of the performance of a 2080 Ti? The TDP of a 2080 Ti is 260W, so that doesn't seem like all that big of a perf/W improvement...

We'll if it is similar at 4K (diff between a 2070 and 2080Ti) then 070 (Turing) to 070 (Ampere) generational difference could be seeing something like over 50% increase performance for 18% more power against predecessor equivalent SKU. We'll have to wait and see.. since that twitter poster said the GA104 will have 8GB there has been tons of fresh rumors that it will be now 16GB so that may make a difference....

 

[Gideon]

https://twitter.com/x/status/1289148083574411264

Compare Ampere and Turing, you will find that the improvement of TS is much smaller than TSE. I don't know why now.
...
That's why I insist on double-fp32.

 

[A///]

Ok, first off "hot" just means inadequate cooling. It literally means nothing else. You can have a 10W chip run 'super hot'. The whole reason the RX 3090 has a gigantic cooler is so that it won't run hot while consuming 350W+.

Claiming Microsoft broke FTC law for a 'wow' factor is rather ludicrous.

And finally, you last posted in 2017 in posts attacking AMD and making claims that we now know were wrong, why come back now?

My guess is he came across the trio of tools 'n fools on youtube posting nonsense.

 

[DXDiag]

No, but you have to decide who you believe. What happened to the Timespy extreme leaks? Huh?

fresh today:

https://twitter.com/x/status/1298505452888551424

The 3090 is ~56% faster than the 2080Ti.

The 3090 is ~18% faster than the 3080.

The 3080 is ~33% faster than the 2080Ti.

The 3080 is ~70% faster than the 2080.


But hey, someone will insist it's only 35% because reasons ..

 

[DXDiag]

https://twitter.com/x/status/1289148083574411264

Regular TimeSpy is 1440p only, it's CPU limited at this resolution, especially with more powerful GPUs like the new Ampere lineup. People who think the 3090 is going to be much faster than 2080Ti @1440p or 1080p, are not thinking straight, our current CPUs don't allow for much scaling at these resolutions.

 

[PJVol]

it's CPU limited at this resolution

What signs should it be obvious from ?

 

[amenx]

I think the real performance gains will be in 4k. Even the 2080ti was CPU limited at 1080p in some games.

 

[DXDiag]

What signs should it be obvious from ?

limited performance uplift, you can see that in any game @1080p right now, the 2080Ti is not that much faster than a 2080 or 1080Ti, some games exhibit that at 1440p too. At 4K or with RT the 2080Ti flexes it's muscles with a 40 lead over 1080Ti.

 

[AtenRa]

The 3090 is ~56% faster than the 2080Ti.

The 3090 is ~18% faster than the 3080.

The 3080 is ~33% faster than the 2080Ti.

The 3080 is ~70% faster than the 2080.

But hey, someone will insist it's only 35% because reasons ..

If TA104 with 3074 cores is true, i dont know how you come up with a 70% higher Raster performance over RTX2080 (2944 cores)

 

[Gideon]

limited performance uplift, you can see that in any game @1080p right now, the 2080Ti is not that much faster than a 2080 or 1080Ti, some games exhibit that at 1440p too. At 4K or with RT the 2080Ti flexes it's muscles with a 40 lead over 1080Ti.

As far as time-spy goes 2080 Ti vs 2080 the gap between benchmarks is quite minor:

2080Ti vs 2080:
Time Spy Extreme (GPU): 7092 / 5151 = 38% faster
Time Spy (GPU): 14611 / 11121 = 31% faster

Time Spy Extreme (Overall): 6576 / 4888 = 36% faster
Time Spy (Overall): 13484 / 10444 = 29% faster

 

[uzzi38]

As far as time-spy goes 2080 Ti vs 2080 the gap between benchmarks is quite minor:

2080Ti vs 2080:
Time Spy Extreme (GPU): 7092 / 5151 = 38% faster
Time Spy (GPU): 14611 / 11121 = 31% faster

Time Spy Extreme (Overall): 6576 / 4888 = 36% faster
Time Spy (Overall): 13484 / 10444 = 29% faster

That actually shows better scaling than actual games, even at 4K.

Original 2080Ti review:

NVIDIA GeForce RTX 2080 Ti Founders Edition 11 GB Review

NVIDIA debuted its Turing graphics architecture today, straightaway with the flagship RTX 2080 Ti. This card packs the promise of real-time ray tracing at 4K UHD, besides huge gains in performance. NVIDIA also put out its best cooler design since TITAN, commanding a very high price for some very...

www.techpowerup.com

2080 Super review for more recent testing (with more driver patches applied):

NVIDIA GeForce RTX 2080 Super Founders Edition Review

NVIDIA swings to the rescue of its $700 price point by "Super"-charging the TU104 silicon and giving it the fastest memory solution ever. In our testing, the RTX 2080 Super achieves a 16% lead over the RTX 2070 Super, making it a solid choice for 4K gaming that doesn't break the bank.

www.techpowerup.com

You can check other reviews too. More reviews will put the 2080Ti closer to 30% over the 2080 instead of 40% over the 2080 at 4K.

 

[Gideon]

That actually shows better scaling than actual games, even at 4K.

You can check other reviews too. More reviews will put the 2080Ti closer to 30% over the 2080 instead of 40% over the 2080 at 4K.

Yes, my point was to address this claim:

Regular TimeSpy is 1440p only, it's CPU limited at this resolution.