JohnnyFootball
GerAlt-Right. Ciriously.
MSRP.....um yeah nope.
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
NVIDIA GeForce RTX 2000 Series card reviews thread
- Thread starter GoldenEye98
- Start date
n0razi
Member
Gigabyte 2080 available right now for under $800, about a $80 difference... not what I would consider "saving a lot".
https://www.newegg.com/GraphicsCardsPromoStore/EventSaleStore/ID-2041438
Last edited:
JohnnyFootball
GerAlt-Right. Ciriously.
Then get it if you have to have something right now.
The 2080 series is just not enough of a leap for me over my 1070 to be worth it.
The 2080 Ti is great, but costs wayyyyyyyyyyy too much.
JohnnyFootball
GerAlt-Right. Ciriously.
People lie so that they can say they're in the "Im a cool guy with cool gear" club. Not to mention you completely lied about the dates of when you got the card.
Again, the card wasn't selling for that price on that date. No record of it.
strange headache
Banned
"Just buy it"
PhoenixTank
Member
Agreed - heck of deal. Benefit of the doubt given - I've snapped up some deals that have been too short lived for price tracking websites. Not that deep though - that price would be more into 1080 territory.
Um either way, can we clear the dude of the charge of buying a 1080Ti at inflated prices and contributing to the upsetting pricing this gen? Either you believe he doesn't have one, or that he got a steal of a deal.
SpaceNoodles007
Member
WOW, two of them! Let us know your thoughts once you've assembled it. How do you plan on cooling them?
Ulysses 31
Member
Stock cooler is good enough, I'm don't plan to overclock.
lukilladog
Member
I think that the dlss provided comparisons might be considerably misleading image quality wise, because if you feed the AI a fixed frame by frame demo it should be relatively simple to recreate that quality when you ask it to do its "magic" over the same demo, working over the same frames showing the same scenery and perspective. But actual games will be challenging the AI with hundreds of thousands of frames that has never seen before so the result of the educated guess cannot be that good. It should be great for in-game cinematics which I bet nvidia will be using (besides fixed benchmarks), but for actual gameplay I doubt it, definitely not a replacement for high quality anti aliasing.
Last edited:
Tesseract
Banned
and yet i got it for that price, on that date. funny how that works.
i don't give a shit about the cool guy with cool gear club, that sounds stupid as fuck.
will not derail further, suffice it say i must've lucked out.
Last edited:
BenBuja
Neo Member
If you watch Digitalfoundrys analysis of DLSS they show the combat sequence of the FFXV benchmark which varies every time and the quality still holds up.
manfestival
Member
I remember when I used to think it was insane to spend more than $300 for a mid range GPU.... now that barely buys you low end.... in this case it won't buy you anything of interest.
lukilladog
Member
Video is not a good way to do IQ comparisons, you shouldn´t expect it to go all pixelated or something, even chequerboarding can be hard to spot on video.
Ps.- I´m not saying that it cannot be good in games, I´m just pointing that the way it has been presented is a really optimal best case scenario and can be a bit misleading.
Last edited:
SpaceNoodles007
Member
dude, just ignore the people who don't believe you. it doesnt benefit you in any way to prove them wrong. you know what you got and the money you paid for it. in the end, you got a great video card. that's it.
BenBuja
Neo Member
True, but their impressions seem pretty positive, but I guess we'll see soon enough how well it holds up during actual gameplay.
Omega Supreme Holopsicon
Banned
DLSS like solutions are likely to get better over time(Many DNN solutions have seen constant improvements in quality). More stable, cleaner and sharper images will likely be possible in the future. Look at the brain, the human eye has a very small high resolution full color area of vision, the number of saccades per second do not seem high enough to sample the entire scene at a fast enough rate, most probably we're dealing with high quality reconstruction for most of the visual field(Hence why foveated rendering works), suggesting it is possible to do ridiculous quality with enough training and the right computations.
Last edited:
neckthrough
Neo Member
Exactly. Unless you're steeped in this field, it takes a while to internalize how DL works and how creepily effective it can be for certain problems. The network isn't memorizing specific training pairs and trying to re-create them. Neither is it upsampling (as in checkerboarding) in the traditional sense.
The specific DL model typically used for such problems is called an "autoencoder". The neural network uses the training data to create an internal model of what constitutes the structure of an image - over time it automatically "learns" about lines, curves, textures and gradients, how to infer these structures from a sampled set of pixels, and how to re-sample those structures into a different resolution.
Of course the quality of the model will depend on the size and nature of the training data. So a model trained on Super Mario Odyssey is unlikely to work very well on Uncharted. And yes, a model re-run on its training data is likely to show overly-optimistic results. But the approach is likely to be far more robust than most human-crafted interpolation techniques.
vpance
Member
I watched their FFXV demo in 4K. DLSS was blurrier but exhibited less jaggies especially in transparencies. Depends what you prefer, I guess.
Omega Supreme Holopsicon
Banned
Put mouse onscreen and put it offscreen to see difference. On screen is the much much much blurrier TAA, off screen is the much sharper DLSS if I'm not mistaken.
Last edited:
nemiroff
Gold Member
I'm so confused right now.. Not only is it the exact opposite of what I actually see in the video (and f.ex. the screenshots above), but it's also not what the guys (DF) in the video is saying.
vpance
Member
Go to 1:51 in the DF vid. Less jaggies but blurrier.
Those 2 screens aren't at the exact same time. Looks like the focus is slightly different, and bloom too.
Last edited:
Omega Supreme Holopsicon
Banned
Those screens are at most like a second apart, it is the same scene and point in the demo, no reason for massive changes to occur in that scene for no reason. Though could happen, but I would be surprised.
I mean look at the wheel close up, and do the screen switch, it moves like a single inch.
Last edited:
Omega Supreme Holopsicon
Banned
True but there are other shots arbitrarily close in time that show similar discrepancies in favor of DLSS over TAA
DLSS vs TAA on transparencies reddit
EDIT: One comment suggested it may be DLSS lacks depth of field which may explain why it is sharper in multiple areas if true.
Last edited:
DemonCleaner
Member
what people need to understand, is that these cards are significantly more expensive to make than pascal. here you have the respective die sizes.
RTX 2080 Ti | RTX 2080 | GTX 1080 Ti | GTX 1080
754mm2 | 545mm2 | 471mm2 | 314mm2
because of the additional surface area, nvidia get just about half as much dies from one wafer. that means pure GPU production cost doubled. then there are the cards themselves. they have much higher power requirements than before. that makes VRMs and cooling lot more expensive. then you have years of research that went into the RT stuff, DLSS etc. that has to be recouped.
for nvidia to retain the same gross margin per card, they pretty much had to raise the prices to that level. as the only accurate leak by before that launch suggested, Jensen thought about eating some of the additional cost (after all their cash reserve would have easily allowed that), to get RTX under the people. realizing that there won't be a counter move by amd, i guess he decided a few months back that that won't be necessary.
i find it kinda ironical that people that defended every shit nvidia (if i remember correctly you were one of them) pulled the last few years, are now complaining about being bend over. maybe should have bought some Radeon cards back then, ay?
ymoc
Member
Oh boo hooooo.
Their gross margins have been going up for the last 9 years and more than doubled compared to 2009.
That wafer size is reaaaally cutting into their profits right.
DemonCleaner
Member
can't upset those sharholders!
pawel86ck
Banned
I have tested many games and downsampling frequently results in DOF reduction because you are scaling from much higher resolution. If you choose extreme downsampling resolution you can even remove DOF totally. For examply try fable PC version. Play it in your native resolution first, and I bet you will see ugly DOF. Then use gedosato downsampling tool and play it in much higher resolution. Result is no DOF at all if you choose high enough resolution
. Because DLSS shows DOF reduction also, it's a proof it indeed can somehow restore missing details and this thing works better than expected. Of course DLSS didnt restored all details perfectly because otherwise you would see no DOF at all. So sharpness is good enough to rival TAA and even surpass it, but still true downsampling would give better results, because it would remove DOF totally. But because DLSS x1 already looks so good even despite upscaling from lower resolution, then DLSS x2 (it starts it's magic from native resolution) will easily produce the best picture quality and AA reduction ever possible for free unlike MSAA and SSAA methods.
Last edited:
ISee
Member
I thought about this for a while now....
The manufacturing cost for a 28nm wafer is about $3500 per 300mm.
The manufacturing cost for a 16nm wafer is about $4800 per 300mm.
TSMC manufacturing process for 16/12nm has a defect rate of 0.06 percent and is improving. That rate was at 0.1 percent when the process was brand new. TSMC does not differentiate between 16 and 12 nm here. That's because the current "12nm" process is the old 16nm process, but they were able to increase density instead of dimensions. It is like 12nm, but manufactured at 16nm.
If you take a 23.45mm x 23.45mm (=~250mm^2) die size, factor in that a 300mm wafer costs ~$4800 and a defect rate of 0.06% we get ~70 good GPU dies out of manufacturing. Without binning and other steps, which will increase the amount of good dies by a bit.
$4800 : 70 = $68.6 per TU104 (rtx 2080)
GP104 (GTX 1080) uses a very similar process to the TU104
17.72mm x 17.72mm, ~$4800 for a wafer, 0.06 defect rate =
$32.7 per GP104 (GTX 1080).
Defect rate was probably a bit higher when gp104 production rate started, so cost was probably more around $34 or so.
In the end, yes TU104 production costs ~$34 more per die then for the GP104. In fact it is about twice as expansive. If we play devils advocate, the price increase is justified. Factor in GDDR6, better FE cooling, a better pcb design, 80 additional tiny screws and it all comes down to a justifiable +$100 MSRP over last generation.
GTX 480 = 530mm^2 ($500 MSRP, 250W)
GTX 580 = 520mm^2 ($500 MSRP, 250W, + 25% more performance over last gen)
GTX 680 = 294mm^2 ($500 MSRP, 200W, + 25% more performance over last gen)
GTX 780 = 561mm^2 ($650 MSRP, 250W, +25% more performance over last gen)
GTX 980 = 398mm^2 ($550 MSRP, 165W, + 40% more performance over last gen)
GTX 1080 = 314mm^2 ($700 MSRP, 180W, + 60% more performance over last gen)
RTX 2080 = 550mm^2 ($800 MSRP, 225W, +35% more performance over last gen)
Nvidias price structure was pretty flat for a long time, independent of die size and production nod. But the performance difference between AMD and Nvidia wasn't huge at that time either. Till the 900 series arrived. It was a bit cheaper then the 700 series and significantly stronger. At this point AMD started to fall behind. Then came pascal and hit like a truck, a relative small price increase for ~60% more performance. The RTX 480/580 wasn't even able to compete against the 1070. It took AMD a year to release a product that reached 1080 performance levels, but the 1080Ti stood undefeated.
All the time nvidia was able to gather market share, they even learned that people are willing to pay high prices for top tier hardware when they released the first Titan card back in the 600 series days. The mining craze emphasized that knowledge, add in the fact that there is no competition to their new RTX cards (both from a technological and pure performance level point of view) and you get to a point where they are able to ask as much as they want. And why shouldn't they: RTX cards are selling pretty fast, and the best competition is their cheaper, 1080Ti. It's a win - win for them. It's a loose for every budget orientated consumer though. I strongly believe that we are paying a premium here.
People will of course disagree with me, rightfully so. I can't 100% proof that, becaue I have no idea how much nvidia is making per card and how much amd is making per card. But there is way more to this then just production and development cost imo.
pawel86ck
Banned
25-68% (depending on the game), around 30% on average. But the thing is, that 30% is very misleading because if someone is using 1080ti, that person will most likely dont care about games that already runs great, but rather care about the most demanding ones, that run slower on 1080ti (performance around 40-50fps). In games like that performance is much higher than 30%.
If my math is correct, result is 64% better, GTX 1080 results are doubled, and I have found more games with similar jump.
Wolfenstein 2, results from the most demanding location (new orleans), 68% better results.
So 25% up to 68% range, and that's still only half of GPU being benchmarked. Very soon games will use entire turing chip and what then? Of course performance will be even better.
-near 100% performance improvement thanks to DLSSx1, and DLSSx2 (unlike DLSSx1 it starts it's magic from native resolution, so it will be very high quality AA for free, people will no longer need to use extremely demanding MSAA/SSAA and waste performance for AA alone)
-additional 20% thanks to veriable rate shading (and I'm not even mentioning other new shading features so probably performance increase can be even higher than 20%)
-potential additional porformance in HDR (some HDR games can run 20% slower on pascal)
-6x RTX performance 9-11 fps on 1080ti, 45-65fps on 2080ti in star wars RTX tech demo
So here we have 25-68% + 100% + 20% + 20%, and + 6x in RTX games, that's very impressive performance jump. Few months from now it will be very interesting to see pascal vs turing performance in new games when AAA games will start using all turing features.
Last edited:
thelastword
Banned
First off, FFXV is an Nvidia sponsored title, that demo is also the worst benchmark to have surfaced in-gaming for the last ten years. Having excessive drawcalls and rendered hairworks (when not in proximity of the player) in a large open-game, so many issues, which they said they've fixed but yet so many issues still exist....So much so, that Gamersnexus (who discovered all of that in the first place and exposed it) said he is not using this joke of a benchmark to do any testing whatsoever......
That infiltrator demo is missing many details at distance with DLSS, and there's an instance of lots of shimmering around a light source compared to the more stable AA'd image of TAA in that video. This whole launch is full of holes, an entire conference where they mentioned RTX (raytracing), so everybody thought RTX was raytracing, till we saw their description of it includes DLSS as an RTX feature too. Even that low end RTX feature (DLSS) is not live at launch.....10 years in the making and still not ready for prime time........So when some journalists say DLSS looks better with no games to test, this is simply those doing PR for Nvidia at this point....How can you test what does not exist?... Hell, even Jayztwocents (an NV guy) says he is not touching any of the NV demos to do benchmarks since it's not interactive and available to the public....and it makes sense......These features should be tested when they show up in real live games, not before....
That infiltrator demo is missing many details at distance with DLSS, and there's an instance of lots of shimmering around a light source compared to the more stable AA'd image of TAA in that video. This whole launch is full of holes, an entire conference where they mentioned RTX (raytracing), so everybody thought RTX was raytracing, till we saw their description of it includes DLSS as an RTX feature too. Even that low end RTX feature (DLSS) is not live at launch.....10 years in the making and still not ready for prime time........So when some journalists say DLSS looks better with no games to test, this is simply those doing PR for Nvidia at this point....How can you test what does not exist?... Hell, even Jayztwocents (an NV guy) says he is not touching any of the NV demos to do benchmarks since it's not interactive and available to the public....and it makes sense......These features should be tested when they show up in real live games, not before....
DemonCleaner
Member
yeah i pretty much did the same calculation as you. just set the defect rate to 0,05 instead and you get 48 good dies/wafer for 2080ti and 89 for the 1080ti. 74 for the 2080 and 151 for the 1080. are your sure about those wafer cost though? i know the source for that figure and other sources are thin spread. but i read somewhere else like 6000$ for a 14nmFF wafer. maybe your figure doesn't include handling and TSMCs mark-up.
please don't get me wrong. i never wanted to dispute that. but nvidia really began screwing people over when they started to sell midrange GPUs (seen from a cost perspective) for high end prices, like with the 680 and 1080 and not just with this launch now. now with turing they just don't want to give up the margins per card that they worked up to the generations before.
Last edited:
ISee
Member
Estimating wafer prices is pretty hard. They are continuously increasing. No idea how far off from reality I am tbh. It very well could be $6000 for 14nm, at this point in time.
Last edited:
SpaceNoodles007
Member
LucidFeuer
Banned
Long story short: dumb couch potatos with money to burn will buy it anyway, so why not inflate the price for minor performance boost?
thelastword
Banned
SLI? according to Nvidia the more you buy the more you save AMIRITE?
I wonder how Tom feels now, I mean how can you miss...
"Life is short. How many months or years do you want to wait to enjoy a new experience? You can sit around twiddling your thumbs and hoping that an RTX 2080 gets cheaper, or you can enter the world of ray-tracing and high-speed, 4K gaming today and never look back. When you die and your whole life flashes before your eyes, how much of it do you want to not have ray tracing?"
https://www.tomshardware.com/news/nvidia-rtx-gpus-worth-the-money,37689.html
Looking at this whole launch and prior ones.....In every gen, you had the base x80's beating the prior gen's TI's by a wide margin. The GTX 980 beat a 780ti, a GTX 1080 destroyed a 980ti...but now.....a 2080 vs a 1080ti (is a no show).....but if we spec up...not that we need to...
780 ti -> 980ti =30% improvement, 7% price decrease = better value for money spent
980 ti -> 1080ti =44% improvement, 8% price decrease = better value for money spent
1080 ti -> 2080ti =28% improvement, 72% price increase = much lower value for money spent....
Also, why is no one mentioning that all the FE cards being reviewed are OC'd cards out the gate......Since when is that a normal thing at launch (to bolster numbers obviously, much closer to their vague chart numbers prior to launch), which brings me to the question, so how do these cards fare at standard clocks? 2080 vs 1080ti??
If there is so little uplift from an entirely new architecture at such a huge markup, on simple, yet good ol' fashioned rasterized games. I shudder to see performance with RTX ON...Nvidia has been very foxy about all of this for a reason, and we can already see it now with rasterized games performance from last gen to now. The next phase which I think will disappoint even more, is with RTX ON (relative to resolution and perf)....
If Raytracing is taxing a 2080ti as can be seen even in their selected demos, then 2080 and 2070 RTX performance will be absolutely bottom barrel......I think NV is just trying to get as many persons on the hype train as possible at $800.00 and 1200.00 before said folk realize what they've really bought into at such prices......In other words, buy RTX before it's officially released and implemented in any game. Talk up 4x Quadro server performance in one turing card, talk about DLSS before we even see it in a playable game. Preorder before you see actual real world results from reviewers on said newgen features. As Tom said....
"When you die and your whole life flashes before your eyes, how much of it do you want to not have ray tracing?"
I wonder how Tom feels now, I mean how can you miss...
"Life is short. How many months or years do you want to wait to enjoy a new experience? You can sit around twiddling your thumbs and hoping that an RTX 2080 gets cheaper, or you can enter the world of ray-tracing and high-speed, 4K gaming today and never look back. When you die and your whole life flashes before your eyes, how much of it do you want to not have ray tracing?"
https://www.tomshardware.com/news/nvidia-rtx-gpus-worth-the-money,37689.html
Looking at this whole launch and prior ones.....In every gen, you had the base x80's beating the prior gen's TI's by a wide margin. The GTX 980 beat a 780ti, a GTX 1080 destroyed a 980ti...but now.....a 2080 vs a 1080ti (is a no show).....but if we spec up...not that we need to...
780 ti -> 980ti =30% improvement, 7% price decrease = better value for money spent
980 ti -> 1080ti =44% improvement, 8% price decrease = better value for money spent
1080 ti -> 2080ti =28% improvement, 72% price increase = much lower value for money spent....
Also, why is no one mentioning that all the FE cards being reviewed are OC'd cards out the gate......Since when is that a normal thing at launch (to bolster numbers obviously, much closer to their vague chart numbers prior to launch), which brings me to the question, so how do these cards fare at standard clocks? 2080 vs 1080ti??
If there is so little uplift from an entirely new architecture at such a huge markup, on simple, yet good ol' fashioned rasterized games. I shudder to see performance with RTX ON...Nvidia has been very foxy about all of this for a reason, and we can already see it now with rasterized games performance from last gen to now. The next phase which I think will disappoint even more, is with RTX ON (relative to resolution and perf)....
If Raytracing is taxing a 2080ti as can be seen even in their selected demos, then 2080 and 2070 RTX performance will be absolutely bottom barrel......I think NV is just trying to get as many persons on the hype train as possible at $800.00 and 1200.00 before said folk realize what they've really bought into at such prices......In other words, buy RTX before it's officially released and implemented in any game. Talk up 4x Quadro server performance in one turing card, talk about DLSS before we even see it in a playable game. Preorder before you see actual real world results from reviewers on said newgen features. As Tom said....
"When you die and your whole life flashes before your eyes, how much of it do you want to not have ray tracing?"
Last edited:
lukilladog
Member
Bang for buck vs console is at its lowest ever, I always liked that you could buy a $200 card and enjoy 50-60fps at high settings and resolutions 50% higher than consoles. But now they think that $1200 for 40fps at the same resolution of 5 year old consoles is somehow acceptable because tech has to advance, pff.
longdi
Banned
Any site that benchmark the OC RTX cards on continuous run? With Pascal, a simple liquid cooler solution helps maintain the peak boost clocks over those heavy customed fans.
But looks like Nvidia FE cards are now impossible to slap on a AIO cooler.
My 1080Ti FE works with EVGA hybrid cooler like a charm!
Unscrew part of the face plate, remove the tiny vapor chamber and slap AIO on.
The stock fan and memory heatsink plate remains intact, exhausting the heat outwards, piss on any AIB solution.
Even the Geforce GTX illumination still works.
Very easy mod, very pleased with the results.
Its a beauty n a beast.
But looks like Nvidia FE cards are now impossible to slap on a AIO cooler.
My 1080Ti FE works with EVGA hybrid cooler like a charm!
Unscrew part of the face plate, remove the tiny vapor chamber and slap AIO on.
The stock fan and memory heatsink plate remains intact, exhausting the heat outwards, piss on any AIB solution.
Even the Geforce GTX illumination still works.
Very easy mod, very pleased with the results.
Its a beauty n a beast.
Last edited:
longdi
Banned
Intel has been for years, accused of selling laptop designed chips as desktop chips by unlocking the TDP.
You know those i3/i5/i7 with integrated gpu wasting space? You know Intel skimping on thermal paste?
Their schit finally caught up with them as Intel will suffer for the short-mid term.
Imo the 2 RTX cards launching now, are designed for servers and render farms, too much bulk that gaming dont need.
I see Nvidia will cut down on the Tensor cores as we get enough power to render 4K full pixels by next year.
Tbh the Volta GPU with more CUDA cores, also do not show much improvement over Pascal in today's games. I wonder if traditional engine is running into limits, and it make sense for Nvidia to push/seed RTX.
You know those i3/i5/i7 with integrated gpu wasting space? You know Intel skimping on thermal paste?
Their schit finally caught up with them as Intel will suffer for the short-mid term.
Imo the 2 RTX cards launching now, are designed for servers and render farms, too much bulk that gaming dont need.
I see Nvidia will cut down on the Tensor cores as we get enough power to render 4K full pixels by next year.
Tbh the Volta GPU with more CUDA cores, also do not show much improvement over Pascal in today's games. I wonder if traditional engine is running into limits, and it make sense for Nvidia to push/seed RTX.
Last edited:
