Original thread:
https://www.overclock.net/threads/official-nvidia-rtx-3090-owners-club.1753930/page-42 "BigMack70, post: 28635126, member: 263849"
The 3090 under water is still going to be 35%+ better than a water-cooled 2080 Ti.
At top end, this gen is a lot like last gen. Last gen, 1080 Ti - 2080 Ti was +30% performance for +$500. 2080 Ti - 3090 is like +35-40% performance for +$300.
In the end, this gen is only exciting for 1080 Ti (or older) owners who nope-d out of the $1200 asking price for the upgrade last gen. They got a good upgrade then with the 3080. But for anyone who bought in on the 2080 Ti, this gen feels kinda bad because 3080 is definitely not worth it over the 2080 Ti and 3090 is a bare minimum performance upgrade for another $300 on top of what was already a bad price point.
I made an error with the Metro Exodus benchmark analysis comparing 2080 Ti FTW3 under full water block @ 2100 MHz core, ~7900 MHz memory to vmanuelgm's shunted 3090 run @ 550w.
Comparing the 4K data without DLSS the 3090 is 45% faster. Seeing as how vmanuelgm's 3090 is 10% faster at 550w vs 390w we need to subtract 10% here to make the difference 35% @ the same power draw.
So it's 35% faster at the same power draw @ 4K without DLSS and 57% faster with DLSS enabled on both cards because Ampere has the equivalent of 20% more Tensor Core performance (see previous analysis).
And seeing as how the 3080 (320w) is 30% faster at 4k but only
20% faster than 2080 Ti (260W) at 1440p and seeing as how the 3090 is shaping up to be 20% faster than the 3080 at both resolutions we can safely estimate 3090 1440p performance to be 10% less.
So 25% without DLSS on both cards and ~45% with DLSS on with both cards @ 1440p.
GA102 falls off a performance-efficiency cliff beyond 400w.
The 3080 under LN2 with virtually unlimited power only beat the 2080 Ti by 1k points in Port Royal, a 2.5k point increase from factory clocks (11.5k) making for a 21.7% overclock.
The 2080 Ti does
8738 in Port Royal at factory clocks and the world record is 13090 under LN2 making for a staggering 49.8% overclock.
So much for the theory that Ampere would similarly overclock well removing thermal and power constraint.
Ampere is already 20% into it's 30% overclock.
At a point with Turing, say beyond 2150 MHz and 400w, more freq requires an inordinate amount of power and wattage. But 2150 Mhz core is a 30% overclock!
GA102 is clocked 10% under this threshold from the factory!
You MIGHT be able to squeeze another 10%, but youre going to need to run GA102 at another 47% more TDP!
It takes 160w more over 390w for just 10% return with GA102-300! Basically, 1 additional MHz beyond say 1950 MHz requires 1 watt of power!
The first 1950 MHz only requires ~390w of power! And that last 160 requires another 160w!
Basically this is the only upgrade path for someone with a 2080 Ti. The 3080 is maybe 10% faster at 1440p at the same power draw, and that's honestly pushing it.
If youre at 4K or in VR then Ampere is 10% faster than at 1440p.
I went for a walk this afternoon and thought about this situation deeply. This is such a ridiculous value proposition that it truly borders on the absurd.
A quality AIB 3090 (Strix or FTW3) would run me $1800 before taxes, or around $2k after taxes. Add a water block and back-plate from EKWB and youre looking at $2300 including tax and shipping. For a 25% gain (45% gain in DLSS titles) at 1440p and 35% gain (55% in DLSS titles) in VR / at 4K?
For $2500 this thing is going to need to be as fast as 2080 Ti SLI and transform into a sandwich maker and make me a sandwich on command.
"Introducing Transformer 3090, 2x faster than 2080 Ti, can transform into a humanoid robot and make you a sandwich = $2500!"
1080 Ti to 2080 Ti was not 30%, it was around 50% in actuality.
[URL unfurl="true"]https://www.3dmark.com/compare/spy/14026055/spy/2949106[/URL]
That's with no DLSS or RT core involvement! The 2080 Ti is @ 373w and that 1080 Ti is only at 300w though. I have another run around 16k @ 320w though, which still shows a 50% gain in this bench.
It's not simply $300 more than the $1200 2080 Ti for 35-45% more performance.
1. 2080 Ti was more than 25% faster than 1080 Ti in actuality
2. You could buy 2080 Ti used as I did, XC2 local from Craigslist for $900 making for $1050 with the water block for a solid 50% gain.
Now it's more than 2x the price for half the gain ($2300 for 25-45%, DLSS dependent, at 1440p, whereas my the 2080 Ti took my 1440p up 50% across the board in the vast majority of my demanding titles, with a few outliers, with no DLSS)
I don't know who this product is marketed to but I've had enough of this nonsense.
I know Nvidia reads this sub-forum so please don't interpret this as a diatribe directed at you, it's more for them.
They have lost touch with reality.
Also, 8K? No-one wants 8K. Ultrawide 1440p is where it's at and for that my 2080 Ti is plenty fast and will have to tide me over as we wait for Hopper, or, if Big Navi, whose 6900XT is shaping up to sit in between the 3080 and 3090 with 24GB of video memory (no engineered obsolescence here) and whose underlying architecture, RDNA 2, is shared with both next-gen consoles who will have path tracing, AI super-sampling and rapid data retrieval GPU pipeline "RTX IO" equivalent, and will most likely run said console ports natively better, oh and for $1000! In November.
I'm waiting, this thing is a joke, sad thing is, seeing as how the 3090 is only 20% faster than the 3080 best case scenario, there's no room to squeeze a 3080S or 3080 Ti in there, UNLESS, they intend to shaft the 3090 crowd with a "3080 Ti" which amounts to the same GA102-300 die with say 16GB of video memory for $1000. I mean they've basically done this before and they did just release a card that they are claiming will be faster than the 2080 Ti (conveniently left at 260w in their "comparisons") for $500 less than two years after releasing the $1200 2080 Ti.
Either way, I'm not buying this, NV have completely lost their marbles.
$1800 before taxes for 25% bump at 1440p up and over overclocked 2080 Ti @ the same power draw and with next-to-zero overclocking headroom LMAO.
No, away with you and your nonsense.
Here's that 3090 @ 375w vs 2080 Ti @ 375w
16,750 vs
20,569 Timespy GPU = 22% increase for $1700 before taxes
Metro Exodus = 25% and 45% (DLSS) at 1440p and 35% and 55% (DLSS) at 4K for $1700 before taxes
Here's 2080 Ti vs 1080 Ti =
57% increase for $1200 new (~$700-900 used)
Here's 1080 Ti vs 980 Ti =
47% increase for $700
Here's 980 Ti vs 780 Ti =
59% increase for $700
Anyhow, goodbye everyone, I don't recommend anyone buy this at this price point. I recommend waiting to see what AMD has in store for us in November.
6900XT rumors:
10% faster than 3080, 24GB video memory, under 300w (7nm TSMC) for $1000 and will most assuredly run the next gen console ports that are build on the same architecture better.
Unless youre tied to Nvidia hardware, i.e. a G-Sync panel (the whole point of the module in retrospect was to get you stuck in their ecosystem), I recommend considering supporting the competition.
Don't let the pretty cooler fool you, Nvidia is in deep kaka with both next-gen consoles running the same architecture as their only competitor in the discrete gaming GPU market, they rushed Ampere out of the door in an attempt to beat AMD to the punch and have mostly succeeded with that. But the fight isn't over. If Big Navi GPU's handle the next-gen console ports native path tracing, AI Super Sampling and rapid GPU data retrieval pipeline (I don't even know what AMD is calling this, but it's their version of RTX IO, and they were the first to invent it) is more efficient (think about it, Nvidia is going to have to render DLSS and RT on top of an engine that was built around RDNA 2 architecture, it's not going to be as efficient as simply rendering those features natively with RDNA 2 architecture, i.e. Big Navi).
Anyhow, even if you aren't keen on considering AMD, if you wait NGreedia is likely to drop whatever card they intend to shoehorn in between the 3080 and 3090 in response to a $1000 24GB card that is faster than the 3080 for the same price point.
Buying this mediocrity, and
I do mean mediocrity, right now is not very prudent.
This is what a monopoly looks like, a $1700 before taxes 80 Ti card that is the only upgrade path to the last 80 Ti card you probably bought for $1200 before taxes.
No empire lasts forever, here's to hoping AMD upsets the GPU market just as they did the CPU market.
Addendum:
If youre an Nvidia goon reading this, the final nail in my discontinuation of loyal support is not just your pricing, but your killing off of 3D Vision as of driver 425.31.
How much did that save you in the grand scheme of things? You had to pay a programmer for the extra 5 minutes required to copy the previous 3D Vision bits forward to the next driver release? Gotta get rid of that to save what, 5 minute of man-hour labor per driver update?
Ok, well I still use 3D Vision, and upgrading to Ampere means I say goodbye to that? From what we can tell thus far, 3D Vision no longer works with DX11 titles.
[URL unfurl="true"]https://www.mtbs3d.com/phpbb/viewforum.php?f=105[/URL]
So yeah, no, you DO NOT deserve $1700 before taxes for your rebadged 80 Ti card. Sorry.