"Krzych04650, post: 28637180, member: 469415"
Sold out completely, even here where good models are like 3x average monthly salary [image]data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7[/image] Not getting any for few months it seems.
Performance isn't that great though, even more favorable reviews like TPU have 45% average gain vs 2080 Ti before considering zero OC potential of Ampere, so real world average OC vs OC is likely less than 35%. In practice this is very enough to make you able to hit performance target instead of being far from it, performance target is kind of a 0-1 situation especially in most demanding scenarios, so all of these percentages are not telling full story, but still one would hope for more than another 35% gen on gen leap.
Try 20-25%.
Do the math, compare 2080 Ti at the same power draw (the OC Strix on the chart) to both the 3080 and 3090.
Here, let me help you:
RDR2 4K3090 FE Stock: 92 FPS, 3080 FE OC: 90 FPS, 2080 Ti OC: 72 FPS, Percentage change: 27% and 25%
RDR2 1440p3090 FE Stock: 134, 3080 FE OC: 131, 2080 Ti OC: 118, Percentage change: 14% and 11%
Rainbow Six Siege 4K3090 FE Stock: 203, 3080 FE OC: 184, 2080 Ti OC: 161, Percentage change: 26% and 14%
Rainbow Six Siege 1440p3090 FE stock: 358, 3080 FE OC: 337, 2080 Ti OC: 299, Percentage Change: 20% and 13%
SOTTR 4K3090 FE Stock: 104, 3080 FE OC: 94, 2080 Ti OC: 82, Percentage Change: 27% and 15%
SOTTR 1440p3090 FE Stock: 169, 3080 FE OC: 158, 2080 Ti OC: 138, Percentage Change: 22% and 14%
Now, while we are at it, lets have a look at overclocked 2080 Ti vs overclocked 1080 Ti performance for comparison, including synthetics. Because I'm tired and because there isn't a meaningful percentage change between 1080 Ti and 2080 Ti at 4K I will only look at 1440p below. Youre more than welcome to look at 4K on your own.
RDR2 1440p2080 Ti OC: 118, 1080 Ti OC: 74, Percentage Change: 59%
Rainbow Six Siege 1440p2080 Ti OC: 299, 1080 Ti OC: 198, Percentage Change: 51%
SOTTR 1440p2080 Ti OC: 138, 1080 Ti OC: 95, Percentage Change: 45%
Before going forward I feel the need to place emphasis that, as I stated repeatedly, GA-102 is right at the very edge of the performance-efficiency curve and any more power does not translate linearly into performance gain. We see this with someone here commenting that Strix @ 480w yields 3% improvement over FE @ 390w and the fact that GA102-200 has a whopping 7% overclock and all of the figures above are comparing overclocked 2080 Ti to overclocked 3080 and that 20% at 1440p and 30% at 4K average has shrunk down to 11-14% at 1440p and 14-25% at 4K. I will include both non-overclocked and overclocked performance gains between architectures in the following synthetics. We will be looking at both Timespy and Port Royal, to be fair to the meaningful RT gains of the new architecture with exception of the 980 Ti to 1080 Ti comparison where Firestrike is used instead:
3DMark Timespy GPU
3090 FE: 20,144 (390w) Overclocked: (no Timespy GPU data here, but it seems 3090 FE overclocked is good for another 5%)
3080 FE: 17,832 (320w)
2080 Ti FE: 13,610 (260w)
2080 Ti FE OC on air at same power draw as 3080: 15,500 (320w)
1080 Ti FE: 9521 (260w) (earlier benches have this card at 8660, they may have re-benched this with newer drivers)
Percentage change normalizing for power output:
1080 Ti FE to 2080 Ti FE: 43% (with newer bench, 57% with original bench of 8660)
2080 Ti FE to 3080 FE @ 320w: 15%
2080 Ti FE to 3090 FE @ comparable power draw (373w vs 390w): 20% and 25% (3090 overclocked)
GTX 1080 Ti vs 980 Ti @ 300w (3DMark site is slow right now, but it's 20k vs 30k, or 50%)
Port Royal: 2080 Ti @ 373w = 10,500, 3080 @ 320w = 11,450, 3090 @ 390w = 13,500, percentage change: 9% and 30%
The 3090 shows a meaningful gain in RT, but imagine this metric with a 628mm2 7nm TSMC die, or about 30-40% higher than this for the same amount of money. But don't worry, NGreedia saved $26 per yield going with Samsung 8nm EUV and no, the queue wasn't full and wasn't forecasted to be full (current 7nm TSMC capacity is at 30%) because crApple was switching their demand to 5nm and freeing up that much capacity.
This is a good one, I think HyperMatrix made this statement:
"The 3080 is unrivaled in efficiency, we have never seen a gain of 25% over the outgoing 80 Ti card at the same power draw".
GTX 1080 = 23% faster than 980 Ti @ 36% less power draw.
GTX 1080 Ti = 50% faster than 980 Ti @ the same power draw.
AND THEY ONLY DROPPED THE NODE SIZE 43%.
12nm FFT to 8nm EUV = 50%.
And we are celebrating this mediocrity?
Additionally, not only is this mediocre performance uplift considering the node halved but imagine the performance uplift between a 425mm2 3080 (same performance as now) and 628mm2 3090 on 7nm TSMC.
10% and most of you will buy this crap, telling NV "hey, that's great, keep making garbage and slap whatever price-tag on it, as long as it's 10% faster than the next fastest card I will buy it at any price because I have more money than sense and don't understand that I've created this very situation by buying your overpriced Titan Xp and your overpriced 2080 Ti at full price and then some"
It's just disgusting honestly.
And please save your "you can't afford it so youre complaining nonsense". I could buy this if I wanted to, but with the water block bringing the total to $1850 after taxes (FE + Bykski block) I would probably have to cancel my HP Reverb G2 pre-order, and for what, 25% gain at 1440p? Maybe 45% with DLSS on?
This is garbage.
I will wait to see what AMD has in store.
Ampere is basically Nvidia's Comet Lake moment where the value proposition become absurd. Like Comet Lake, Ampere is an expensive inferior node and needs an incredible amount of power for marginal gains over whatever came before it. Meanwhile AMD is taking over as the CPU industry leader if they haven't done so already.
$26 per yield for this mediocrity. That's all NV is saving having gone with 8nm EUV over 7nm TSMC and they knew that with slick marketing, a dumbed down consumer-base and a growing niche "boutique" market demographic (the .01%, the only demographic that can buy this crap) that they could make whatever and sell it.
They literally made you whatever and youre buying it and cheering all along the way.
If AMD release a card (full 80 CU unit Navi 21) that runs faster than the 3080 with 16GB of video memory @ $1000 NV is done.
With any console port, which 90% of PC games are, NV is going to have implement it's exclusive technologies on top of an engine that has been heavily optimized around RDNA 2, the architecture of Big Navi. The 3080 may still be faster on paper, in 3DMark Timespy and Port Royal, or it may be slower, but watch Big Navi run the console games, I don't know, 25% faster because the console based path tracing, AI super-sampling and "RTX IO" will be done natively and therefore with less driver and API involvement.
See this is how it works, when you keep buying whatever a manufacturer makes you invariably end up in a situation where they sit on their laurels and don't push the envelop because they know that they have a captive market who will buy whatever is shoveled their way.
See: Intel
Jim from AdoredTV is right. This is Nvidia's Dumbest Decision because now AMD is going to do to Nvidia what they did to Intel. We are going to see a HUGE upset in the discrete GPU industry this year.
And guess what?
See Intel? They are getting off the 14nm node.
They have to.
Nvidia?
They will be forced to develop Hopper early.
I don't bear any allegiance to any brand or flag. As a combat veteran thrown under the bus by my country long ago I learned that lesson the hard way.
Stop being a fan boy.
Don't buy this garbage.
Compel NV to produce a quality product.
Remember, NV is saving $26 per yield with 8nm EUV.
Imagine the 3090 with a 628mm2 7nm TSMC die that is 30-40% faster than the 425mm2 die (that is as fast as the current GA-102-200 628mm2 8nm EUV die).
Same price.
40% faster, easily.
Now THAT is card worthy of a moniker denoting dual GPU, THAT would be a 3090.