Happy new years all, I had an awesome break, comments are great so far. For performance it will be interesting to see if Blackwell can provide the gains. Sorry made a mistake on the power supply for the 5090 vs 4090, it's as I said and not shown, should be 1000W req for the 5090 and 850W for the 4090.
Love the fe 5090 love the down size of the card even though I normaly do custom water that said it will be intresting to see what water blocks will come out for the fe
20-30% increased performance at each tier for 20-30% increased power draw at each tier, looks more like an overclock than an actual performance increase + more a.i. crap. But hey maybe i'm wrong and the prices won't be absolutely scalped into oblivion. But i doubt it😅.
5090 and 5070 do have 20-30% increase power but the 5080 has 12.5% and the 5070ti is only 5% so at least for these it may not be the whole story. But until 3rd party testing, we won't know
@techyescity it's just the truth, sorry. Mesured by nvidia. This 5090 is just a 4090 Ti super. The node is barely different 4N to 4NP. They can't give us a gap of performance. Wait the 6090 and the 2N to see a major gap like 50% at least vs 5090. Btw i hope for who gonna buy this , this 5090 can reach 3ghz as the 4090 cause mine do it ez (an FE watercooled) and for example on speedway with my poor 5800x i do 11326 pts (world record with 5800x and 4090) And the 5090 ~13500 pts..(with a better cpu and ddr5 for sure)
21,000 cuda is no joke though. 512 bit memory bus and 32gb Gddr7. The $400 upcharge is Justified! The 5090 is the card to get really. Too bad it'll sell out within Seconds and be scalped for a heavy premium immediately. 😢
Especially when all the GPUs use more power. You're looking a 0-10% better efficiency. Not a performance uplift per core.. better hope that your tier GPU gets a decently bit more cores.
@@arenzricodexd4409 from the leaks we've heard.. " 5080 = 4090 in raster and 5090 is 1.8x raster and over 2x ray tracing. " 🥴 Same node and Nvidias own slide say about 35% in 2 games WITH RT! So raster MUST be no more than 25%. 30% will be 20 game average including several heavy RT wins.
Great to see you calm and relaxed, the vacation was awesome for your Mojo. I am sure it will not be long before you jump back on the content train once again. Always waiting for more Sir!
16GB even without Neural Rendering should be enough for 4K for a long time. Probably more time than 12GB for 1440p. To be honest, if you want to game at 4K but buying a 16GB GPU for $1000 bothers you, then you can get a 5070Ti for $250 less, or if you don't care about DLSS4, a 4070TiSuper for even less when the 5000 series arrive.
What I'm worried about, are people buying a new, relatively expensive graphics card which only has *12GB of VRAM.* RTX 5070, for example. Some/few recent AAA Games, with certain settings are already a problem. At least with 4K + RT + DLSS etc. with Ultra/Very High settings. But.. as long as 12GB is enough, RTX 5070 will be just fine. In a similar fashion as current RTX 4070 regular/Super/Ti (12GB) has been. 16GB of VRAM (like RTX 5070 Ti) should be enough for any Game of near future. And most others after that, at least when lowering settings. Sure, 16K resolution with a specific niche DIY Custom Ultra HC -scenario might require 24GB or even 48GB already today. Eventually, 16GB absolutely will be too little. Just as 24GB and 32GB will be. Still, if I had a choice, I'd never choose Radeon 7900 XTX (24GB) over RTX 5070 Ti (16GB) for Gaming or Blender (OptiX).
I don't like that the focus goes to fake frames instead of raw performance, but to be honest the time has come to me. My 3070 Ti will be substituted by 5070 Ti
Upscaling is no different, 25 to 75% of the pixels are fake but its what it takes to game at 4K, I game at 1440p and rarely use DLSS, when I do , it looks fantastic, I have no problem with fake frames
@@Reeper-x6xwrong. Upscaling works different from multi frame generation. Some pixels in upscaling are indeed fake, but all the frames are generated by game’s engine while 75% of the fake frames in MFG are not going thru game’s engine hence increasing input lag significantly. The AI predictive model can’t accurately predict all the player’s inputs :)
@@Reeper-x6x completetly incorrect, upscaling just renders the game at a low resolution, but those are real frames. Fake frames from MFG are NOT real frames at all, they're just guesses the gpu does with AI.
@ 1080p is 2,073,600 pixels, upscaled to 4K you must generate / fake 6,220,800 pixels 3/4 of the frame Upscale to1440p at 3,686,400 pixels is less demanding Even so, you must increase the pixels by 1.7777 times or nearly double to create each frame, MFG is not evil, its just another way of getting there
It's all features and software. I buy hardware for actual hardware improvement. It's looking like 4090 is maybe the endgame of that. Similar to 1080ti prior to RTX.
RTX 5090 vs. RTX 4090 * Internal memory bandwidth +78 % - GDDR7 21Gbps: +33 % - 512-bit bus: +33 % * VRAM +33 % * PCIe 5.0 x16 (bandwidth +92 %) * CUDA cores +33 % * TGP 575 W, yet still *2-slot card* (sadly, FE only) * Improved decoders (2 vs. 1) * Improved encoders (3 vs. 2) * "AI stuff" may or may not have use for gamers/consumers. In the High End, AMD and Intel are very, very far behind. After the (probable) Super-refreshes, almost all TOP 20 Gaming cards will be Nvidia (when including the previous generations, too). Which is not exactly optimal situation, as far as pricing goes. Hopefully Radeon *7090 XT* (16GB), *7090* (16GB) and future Intels *B7xx* etc. are both decent cards and competitively priced. In the lower/mid tier. High End is now purely Nvidias realm.
Hopefully with DLSS 5 we will be able to completely change the code of the game, so for example while playing doom eternal I can change the genre to "dating simulator" and doomguy can actually go out with the demons instead of ripping them appart.
15:00 "2ms from mouse click to the final image on monitor" are you sure about that? x)) they can somewhat emulate mouse movement (scene + crosshair") before new frames are even ready, but they can't emulate mouse click. yes, wired mouse 1ms and 500fps rtx+monitor = 2ms, but will have nothing to do with nvidia tech.
i find it extremly funny talking about justification..i will get one because i can and because it makes me happy. that’s all justification i need. there are far more expensive hobbies than high end gaming, its a 2000$ not a brand new mercedes for gods sake 😂
its like do you go with the FE version being a couple hundred cheaper but also extremely smaller vs AIB that are extremely larger that you would think might beagle to dissipate heat quicker but obviously its an engineering marvel to have a 2 slot 5090 so im sure it can also cool it.
@deadpool790 According to some it's supposedly has the most expensive heatsink on the market so if we subtract that from the equation and add a non reference model which shouldn't cost that much more than the fe model should be either at MSRP or a few hundred dollars more based on the 4090 aib models which were available at $1599 at launch and cost $150 more. for models like the Suprim liquid 4090
There is a 24gb 5080ti “leaked”. Too game at 4k even with DLSS not sure standard 4080 is giving enough VRAM headroom, especially with all these features. I might wait to see what happens there.
3:10 In all fairness, bud. There's no "perhaps" about it. You aren't going to get a massive increase, as they did quietly slip in some "actual" render performance of 4090 vs 5090, and it was roughly 25%.
Yeah buddy. But 40 series was already huge on raster, beter than AMD in every class. And thats what all those watts DO. That part is already perfected but if you think you are going to get playable RT performance out of raster boost youre crazy.
I mean, 32GB of GDDR7 with 5000 more cuda cores, new tensor cores and better RT cores... the 5090 is nowhere near necessary to enjoy the last AAA games, is a workstation card at this point. So if 4090 was $1600, 5090 being $400 extra is not *that* crazy. I am more concerned about the 5080 being almost *half* 5090, costing the same as the 4080S, so we're not having much increase in price/performance in this generation outside of the 90 class. The same goes for the 5070Ti being basically a 4070TiS, 5070 a 4070Ti and so on.
@@KnightofAges And not including the added cost the premium board partners add on to their models either. Usually the base MSRP FE models are available in fairly small numbers as compared to overall demand and are out of stock much more often than they are available for the highest tiered cards. Cost will likely run more along the $2200-$2300 USD + taxes for the majority of 5090 models.
First off the 5090 versus the 4090 there's a 21% increase in cuda core count with a recommended power supply increase from 850 to 1000 watts. Pure rasterization power will be right around that 21 percent increase (give or take) without DLSS 4 mult generation as I'm sure 3rd party testing will confirm. Other than the founders edition cards the other cards will be anywhere from $2500 to $4000 on release day for the 5090.
Generated frames are made only with the GPU, not a mix of GPU and CPU like the "real" ones. So if your CPU can handle 120 fps, then will be enough to go up to 480Hz. One pro of FG that little people consider is that FG is a way to go over CPU limits. I had an i7 8700 (no K) + RTX 2060, so going at 120fps (40 base and "x3" in lossless scaling or a little less with FSR3) in games like Space Marine 2 shouldn't be possible with such CPU. So like RTX 4000, Zen 4/5 or Alder Lake and Beyond should be enough, especially a Ryzen 7 or high tier i/ultra 5 (the higher, the better obviously).
Seeing how high the prices are for GPUS are getting i'm getting MORE keen on APUS waiting on Strix Halo varients to fully mature and get better before I jump ship to a MINI PC with an APU.
Many thanks for your nice advice! 5070 ti will be the hammer for gaming 5080 will be the most selling 5090 will be more or less for the TH-camr and professional channels Please try to add one very important notice Which model will have LED light ! Many thanks ❤
@@jdmnissan 5070 is comparable to 4070 Super, which is why Nvidia deliberately didn't mention it and instead focused on 4070 for comparison. 5070 is basically 4070 Super with Multi Frame Gen capability. Previous Gen will be getting DLSS 4 as well, only difference is Multi Frame Gen.
Reflex 2 is a feature Nvidia took from VR gaming called Asynchronous Spacewarp, it basically takes the input just before the frame is rendered and they shift the frame accordingly, sometimes chopping part of the frame off. However Nvidia added AI onto it, by letting the A.I. filling in the chopped off portions of the frame. This should reduce the overall latency of Multi-Frame Generation to feel very responsive despite having poor latency. Maybe from 57 ms to feel like a 35-36 ms. Which would be pretty decent add on for all games regardless if you turn on MFG at all.
The gpu steve did the video on was in fact blow through. However a major difference also is the unreleased had the pcb on the side not the center. I'm sure that will effect the blow through effect as well.
The reflex2 thing is strange but not really. I had an idea back then similar to this, but my idea is for frame generation and it is actually more involved than just using current frame + mouse movement (it is more integrated into the engine) because without it, it is basically a technique called asynchronous reprojection (it is being used in VR space). Current frame generation works by delaying the current real frame from being displayed in order to calculate and display the fake frame. Async reprojection in VR space works very much like reflex2 and more since not only it is being used to reduce latency, but also for generating frames. The biggest difference between reflex2 and the current implementation in VR is this inpainting thing where obviously it will need AI to fill in the blanks while in VR it usually introduce defect like edge warping + the fact that in VR it is also being used to generate frame. My idea is more about this kind of frame generation being integrated into the game engine directly thus instead of trying to render high quality frame every time, you render high quality frame and then you render in between frame in a more crude manner and combine it with the previous rendered frame data. It is kinda like async reprojection but instead using only the previous rendered frame data, you use that + current frame data to fill the holes. This will not reduce the input latency like reflex2 or async reprojection, but it will give you more frames without extra latency (basically 60fps with this idea is similar to a real 60fps) and hopefully better quality than reflex2. But I assume if AI is good enough to paint those holes then might as well drop the game engine integration (since it is more costly because of the rendering cost) and go straight to async reprojection combined with AI inpainting.
Guys i need big help right here im buying a new gpu should i get RTX 4060 or RX 6700 XT bith with Ryzen 5 5600 X for gaming Anyone knows whats better for me please reply
Interesting that the TDP for the 5090 has gone up but the recommended PSU has dropped to 850W. I seem to recall 40 series cards had an issue with power spiking so maybe they've gotten that under control with this gen.
Happy new year ! Hey what’s your thoughts on the clock speeds of the 4090 vs 5090. Will the lower clocks be worse for raw performance when your playing certain games that’s actually do take advantage of higher clocks. Specifically call of duty titles. I don’t like frame gen, I always leave it disabled. I’m really struggling to wrap my head around how we get less latency also with generating frames.
Only 4k gamers are going to see 25-30% increase with 5090, 1440p gamers are going to get disappointed it seems like, 10% at best at most GPU bottlenecked titles.
I'm more interested in the Founders Edition cooler designs than the actual cards themselves. Will be very interesting to see how Nvidia cool almost 600w with just a 2 slot design.
LMAO!! My 14900ks has been amazing with my 4090 aorus master. They max out 900 watts both overclocked on a benchmark. The 5090 with a 14900ks oc won't pass 1000 watts. ✌️🇺🇲
Honestly, if they want to ai generate frames, so be it, as long as it still looks amazing and has low latency. If they can accomplish that, whatever, if not its trash. My only other gripe is that the only good looking card is the 5090, no 384 bit card, amd that gap to the 5080 is absurd. Im still looking to maybe upgrade this generation, but thats a maybe until i see benchmarks
For 5080, Core count and tdp both increased around %10, clocks are not too much of an uplift too. So they didn't bring any efficiency this series. GDDR7 will do the heavy lifting I guess, we need some benchmarks with underclocked vram vs 4080.
Seems like this series is mostly a software upgrade. Raster improvement is minimal. These fake frames will be great on slower and story driven games but racing and competitive it won't feel any better. Also framegen doesn't help VR
Even single story games is doubtful, there are people that are more sensitive to input lag compared to other, so Nvidia pushing generated frame as main sellingpoint is weird, to those people, buying 5070 won't be like buying 4090 if they don't plan on using framegen.
@tomthomas3499 the imput lag will be the same as the base frame. So a base of 60 will still be good even if it scales up to 144 for example it's not like frame gen but into tvs that makes input lag worse.
@@ivan-j4l9x The problem is what you see and what you feel is different and hence why it feel odd to them, Daniel Owen said framegen should be treated as motion smoothness not performance improvement and i agree with him.
Lovelace is made on the N4 while Blackwell is made on the improved N4P silicon. Below is from TSMCs website. The 4nm (N4) process technology is an enhanced version of 5nm (N5) technology with density improvement. It started volume production in 2022. To further enhance the N5 family’s performance and power, TSMC introduced N4P and N4X, targeting next wave 5nm products. N4P offers an 11% performance boost compared to N5 and entered risk production in July 2022. N4X is the first in the ‘X’ lineage of TSMC’s extreme performance semiconductor technologies, with 6% speed gain over N4P at moderate leakage trade-off, and volume production is planned in H2’2024. The N4, N4P, and N4X nodes are design rules-compatible with 5nm technology for easy design migration.
They are still intermittent frames. People are mad because they think they pull next 3 frames from somewhere. That picture in nvidia slide is misleading also. They render 2 and AI calculate the transition in between. Correct me if wrong. (I have been trying understand this)
Does seem odd. The specs for the 4090/5090 (clock speeds down), 4080/5080 (clock speeds up), 4070/5070 (clock speeds up) seem odd too. Not sure if that is nvidia, or these tables.
Just wanted to share a thought I had: If the flow through design of the 5090 cooler is integral to keeping temperatures in check, then the Lian Li x Dan A4 H2O is may not be a very suitable SFF case to put it in considering how the vertical mount will cause the card to blow all the heat to the back of the motherboard and PSU. You could flip the PSU fan side to face the GPU to allow for better flow through for the GPU but that still only provides some flow through to one of the two flow through sides. I could be completely wrong about the temps being bad in a case like the A4 H2O though.
That 2ms is not end to end latency. It's just the system latency. You still have the output and peripheral latency to contend with. The end to end latency is probably pretty close to the 5ms you've been able to get as well. Maybe 1ms faster.
W video love your content man, I’m hoping for gaming the rtx 5080 is as fast a 4090 but for thing like video editing and stuff like that which I do as well (at 17) lowkey pretty fun. The dual encode and decode will be really nice plus the extra vram and other improvements vs the current rtx 2080 ti I have now
Only reason I’m upgrading is cause I’ve started to really enjoy single player games like cyberpunk, Alan wake 2, black myth wukong etc and I game at 4k and have been using amds 7900 xtx I bought it on its release day, it’s been great so far only issue is ray tracing, and upscaling methods. Nvidia clearly has taken the cake on that and seeing the 5000 series tech I think it will be a good upgrade cause I really want to get into some 4k path tracing, going for a FE 5080 cause 2000+ dollars for a gpu is just insane to me, regardless if it’s the best in the world
@ my brothers got a 4090 and path tracing always BLOWS my mind, and the fact the 5000 series isint a rasterization upgrade its upgrades are all for ray tracing, path tracing and improving dlss, ghosting, and fracturing, so even if I got only a 4090 I’d be impressed but I know for a fact a 5080 will outperform a 4090 in ray tracing let alone a 7900 xtx I can’t even turn regular ray tracing on 😭 so I’m very excited
Once the benchmarks start coming out with REAL performance without upscaling frame generation etc. I'll be choosing between a 5070Ti or 4080 Super depending on price/performance. No way I can justify a US$1000 plus card I'm afraid.
Happy New Year Prices in Singapore is going to be hell as well, and it's probably the same old thing as with Australia. Higher than currency converted MSRP and longer supply chains. At least in Singapore if I remember correctly the cards go through an importer + distributor before hitting the shelves and everyone wants profit so yeah, we're cooked
$4000 AUD for a GPU.... just to play a computer game....have you looked at the current crop of games? what computer game justifies this expenditure? Madness.
For real, I decided to just get a cheap 2070super I found online for my main rig (125euro), it's basically as fast as a 4060, which isn't a powerhouse at all, but I stick to 1080p, and it's honestly been doing fine so far. The Vram is not great but for 125 euro I can deal with that.
looking at it this way, perhaps gta 6 if it comes to pc, and some future games justify the investment but in general, things are looking pretty bad, yes
Sadly, for most Gamers, even RTX 5070 Ti is too expensive. *If one day in the future...* AMD or Intel will be able to compete in the High End, prices might come down a bit. Until then, Nvidia can freely choose any numbers it sees optimal (for Nvidia). If both a decent product and smartly priced, Radeon *9070 XT* (16GB) _might bring some competition_ to the middle class (RT: ~RTX 4070 Ti? Rasterization: ~7900 XT?). Perhaps even future models from Intel (Arc Bxxx), at least in the lower class.
PSU "requirements" are about how much the PSU needs to be able to deal with, GPU TDP is about how much the GPU can draw sustained. They are not the same thing. For one the electrical specifications for PCI devices allow for short bursts far higher than the average. So this would suggest that the 50series simply has smaller power-swings and never spikes that much higher than its normal use. A GTX 1080 had 250W TDP, yet the powerspikes could go well into the 400W region.
I would have loved if they gave us the full 25k CUDA cores GB202 die for the 5090. It could have been another 3090 to 4090 uplift (that was around 65% on average, more in some RT workloads). I will still buy a 5090. Already built my 9800x3D pc in November. And already bought my 32" 4k 240hz QD-OLED in August. If the 6090 is a massive uplift, I will just upgrade my 5090 to the 6090 when that one gets released. For now I will be fine with my 5090.
It is just really misleading marketing on Nvidias side. They did not lie like some haters want to claim as they did state the differences directly on the slides, but it just makes those graphs all rather useless and misleading. Comparisons should be like-for-like and things like new features like multi-frame-generation should then be mentioned and shown separately.
I think one segment of the users that are really looking forward to the raw power of the new 5090 is going to be those that are heavy into high resolution VR gaming like flight sims and harder to run racing sims as they may finally actually have a GPU that will not struggle to deliver a smooth ample playable FPS and be able to run the higher graphical settings at the same time. Granted this will be coming at an expensive price point! Other than the 5090 though as far as raw performance minus frame generation and using DLSS the 80 series and lower tiered cards seem to be getting weaker as compared to the prior generation offerings of earlier series. In the past the new series card of one tier would generally match the performance of the next higher tiered card from the previous generation. No way is the new 5080 anywhere close to matching the prior 4090 offering which when comparing to past series offerings is a step backwards. So really what Nvidia has done has made the 90 series cards sort of straddle the fence of being a top gaming GPU and/or a slightly cut down professional GPU. Then the 80 series cards are the top tier really designed as a gaming GPU. If you go back to the 30 series and the original launch of the 90 tiered cards even back then in the presentation Jensen said that the 80 series was the top "Gaming" GPU of the series although the 90 series could also be used for gaming. But that separation and distinction was made for the 90 and 80 tiered cards even back then.
I think the more appropriate thing to say is that AI needs to be used in different ways. Like focusing much more on making normal DLSS give you more image quality and even more performance (as an example, I don't know if it's possible to give you more real performance with AI). I agree frame gen using AI is not the way
Thanks for sharing your thoughts Bryan. I am in the market for a new card and like the proposition of the refresh 5080 (Super or Ti variant). Saying this because I believe it will have 24GB VRAM and not cost me 4K AUD. I just don't think 16GB is viable going forward. Thoughts?
I'm very interested in the mechanical design. I hope that it's something other manufacturers copy because it solves the final "dead zone" in the ATX spec where air recirculates. By the way, only the 5090 and 5080 have the new cooler design, the 5070 doesn't get it.
@ is it? Well I am Little bit dissapointed. The biggest upgrade in 50 series is the cooler! That is actually cool new invention! Now 5070 is even more boring…
Happy new years all, I had an awesome break, comments are great so far. For performance it will be interesting to see if Blackwell can provide the gains.
Sorry made a mistake on the power supply for the 5090 vs 4090, it's as I said and not shown, should be 1000W req for the 5090 and 850W for the 4090.
Happy 600000!
Jacket has antialiasing
Love the fe 5090 love the down size of the card even though I normaly do custom water that said it will be intresting to see what water blocks will come out for the fe
Which one do you think will be the best 4K 60 fps High Quality card? I don't care about massive framerates, but I want pretty.
why is the psu required for the 5090 lower than that of the 4090?
Won’t stop the UE5 stutters
underrated comment lol.
@@techyescity Nvidia is enabling this brute forced approach to game development shyte show!
Don't forget you get it stutters just as much on EU4 games :D
I'm playing UE5 games without stutters :) Not at default but there is a route to get smooth gameplay
@@KLA22323 Please help a brother out, what do? or point me to a source I can read for myself. Thanks :)
Hype is never justified, it is hype.
You must not know Lil' Jon
here's my excitement..............MEH.
I feel bad giving you a thumbs up because it was at 69 before I clicked. Hype responsibly people. High expectations = disappointment 90% of the time.
@@ilbro7874 YAEAAAAHH!! HUH!!! AYEEE!! GET DOWN!!! YEAAAAAAAAHH!!
At 1:55, in the chart it says that a 1000 watt psu is required for 4090 and an 850 watt psu is required for 5090 when it's the other way around.
I noticed that too
Stream units are also incorrect for BOTH cards!
@@nichronos No, that's correct.
you can probably get away with an 850 watt psu on the 5090. several sff 4090 builds only use a 750 watt psu
20-30% increased performance at each tier for 20-30% increased power draw at each tier, looks more like an overclock than an actual performance increase + more a.i. crap.
But hey maybe i'm wrong and the prices won't be absolutely scalped into oblivion. But i doubt it😅.
5090 and 5070 do have 20-30% increase power but the 5080 has 12.5% and the 5070ti is only 5% so at least for these it may not be the whole story. But until 3rd party testing, we won't know
The 5080 is only 15%.
The 5090 will absolutely be scalped & will be done so by all major retailers except Best Buy. Ur delusional to think ANY different.
you can't call a 30% of performance with RT a huge gain when the price rise up by 400$ (25%)
Do you think it will be really only 30% though? I was hearing 40-50%~
@techyescity it's just the truth, sorry.
Mesured by nvidia.
This 5090 is just a 4090 Ti super.
The node is barely different 4N to 4NP.
They can't give us a gap of performance.
Wait the 6090 and the 2N to see a major gap like 50% at least vs 5090.
Btw i hope for who gonna buy this , this 5090 can reach 3ghz as the 4090 cause mine do it ez (an FE watercooled) and for example on speedway with my poor 5800x i do 11326 pts (world record with 5800x and 4090)
And the 5090 ~13500 pts..(with a better cpu and ddr5 for sure)
That is a huge gain, in price! Pay up, piggy!! 😂
@@BBWahoo i'm just fine with my 4090, i'll prefer get an oled 240hz and a better cpu on these 2 years.
I'm waiting UDNA from AMD
Vote with ur wallet.
Spend $2,000 to get Fake Frames? I don't think so.
😂
On support game 😂
21,000 cuda is no joke though. 512 bit memory bus and 32gb Gddr7. The $400 upcharge is Justified! The 5090 is the card to get really. Too bad it'll sell out within Seconds and be scalped for a heavy premium immediately. 😢
@@perfectgaming788 There is anti scalper policies in play so let them try!
Buy AMD and play at 30 fps 🤣
Sounds like the 5090 is slightly better than the 4090 but for more money.
GB202 die size is massive.
Who would've thought!
Wow didn't study maths and physics. The hardware is double everything and the raggers come up with "slightly better" LOL
Raw performance on 5090 is 40% better than 4090, which is pretty decent. But the other cards suck.
@@rdmz135its terible like shokingly bad
Hyped by the company, not hyped by fans 😂
with nvidia and the fanboys not understanding what is going on, it is both
Never ever become a fanboy of any brand or company. You will be milked and disappointed in time to come.
I wouldn't call 30% huge gains.
Especially when all the GPUs use more power. You're looking a 0-10% better efficiency. Not a performance uplift per core.. better hope that your tier GPU gets a decently bit more cores.
That's quite a gain on the very same node. There are times when new architecture use new node but the jump is only 30% gain.
@@arenzricodexd4409 from the leaks we've heard.. " 5080 = 4090 in raster and 5090 is 1.8x raster and over 2x ray tracing. " 🥴 Same node and Nvidias own slide say about 35% in 2 games WITH RT! So raster MUST be no more than 25%. 30% will be 20 game average including several heavy RT wins.
27.5% was rounded up to 30% 🤦♂️
@@christophermullins7163 30% more perf for 30% more power and 30% more price.
Great to see you calm and relaxed, the vacation was awesome for your Mojo. I am sure it will not be long before you jump back on the content train once again. Always waiting for more Sir!
The aud pricing will be most likely double the usd pricing. Conversion is 1.63x aud then you have at least 20% markup on top.
The prices will be scary here in Australia 😢
Happy New Year
My take on the jacket was because it looked snake skinned to mirror the year of the wood snake with the coming Chinese New Year.
You’d expect a 1000$ card to last for a long time, but with only 16GB of RAM it will be limited by its VRAM pretty soon.
Rtx neural rendering compresses textures and 16gb vram might not be a problem if nvidia pushes this onto developers
Ram compression apparently means it won't
16GB even without Neural Rendering should be enough for 4K for a long time. Probably more time than 12GB for 1440p.
To be honest, if you want to game at 4K but buying a 16GB GPU for $1000 bothers you, then you can get a 5070Ti for $250 less, or if you don't care about DLSS4, a 4070TiSuper for even less when the 5000 series arrive.
What I'm worried about, are people buying a new, relatively expensive graphics card which only has *12GB of VRAM.* RTX 5070, for example.
Some/few recent AAA Games, with certain settings are already a problem. At least with 4K + RT + DLSS etc. with Ultra/Very High settings.
But.. as long as 12GB is enough, RTX 5070 will be just fine. In a similar fashion as current RTX 4070 regular/Super/Ti (12GB) has been.
16GB of VRAM (like RTX 5070 Ti) should be enough for any Game of near future. And most others after that, at least when lowering settings.
Sure, 16K resolution with a specific niche DIY Custom Ultra HC -scenario might require 24GB or even 48GB already today.
Eventually, 16GB absolutely will be too little. Just as 24GB and 32GB will be.
Still, if I had a choice, I'd never choose Radeon 7900 XTX (24GB) over RTX 5070 Ti (16GB) for Gaming or Blender (OptiX).
Which games are limited by 12g of vram?
I don't like that the focus goes to fake frames instead of raw performance, but to be honest the time has come to me. My 3070 Ti will be substituted by 5070 Ti
Ayee, 6700xt going to a 5070 ti/5080 (or possibly 9070xt)
Enjoy
Upscaling is no different, 25 to 75% of the pixels are fake but its what it takes to game at 4K, I game at 1440p and rarely use DLSS, when I do , it looks fantastic, I have no problem with fake frames
@@Reeper-x6xwrong. Upscaling works different from multi frame generation. Some pixels in upscaling are indeed fake, but all the frames are generated by game’s engine while 75% of the fake frames in MFG are not going thru game’s engine hence increasing input lag significantly. The AI predictive model can’t accurately predict all the player’s inputs :)
@@Reeper-x6x completetly incorrect, upscaling just renders the game at a low resolution, but those are real frames. Fake frames from MFG are NOT real frames at all, they're just guesses the gpu does with AI.
@ 1080p is 2,073,600 pixels, upscaled to 4K you must generate / fake 6,220,800 pixels 3/4 of the frame Upscale to1440p at 3,686,400 pixels is less demanding Even so, you must increase the pixels by 1.7777 times or nearly double to create each frame, MFG is not evil, its just another way of getting there
Happy new year duder
duder
It's all features and software. I buy hardware for actual hardware improvement.
It's looking like 4090 is maybe the endgame of that. Similar to 1080ti prior to RTX.
RTX 5090 vs. RTX 4090
* Internal memory bandwidth +78 %
- GDDR7 21Gbps: +33 %
- 512-bit bus: +33 %
* VRAM +33 %
* PCIe 5.0 x16 (bandwidth +92 %)
* CUDA cores +33 %
* TGP 575 W, yet still *2-slot card* (sadly, FE only)
* Improved decoders (2 vs. 1)
* Improved encoders (3 vs. 2)
* "AI stuff" may or may not have use for gamers/consumers.
In the High End, AMD and Intel are very, very far behind. After the (probable) Super-refreshes, almost all TOP 20 Gaming cards will be Nvidia (when including the previous generations, too).
Which is not exactly optimal situation, as far as pricing goes. Hopefully Radeon *7090 XT* (16GB), *7090* (16GB) and future Intels *B7xx* etc. are both decent cards and competitively priced. In the lower/mid tier. High End is now purely Nvidias realm.
Hopefully with DLSS 5 we will be able to completely change the code of the game, so for example while playing doom eternal I can change the genre to "dating simulator" and doomguy can actually go out with the demons instead of ripping them appart.
Great video. Very informative, mostly to the point, hardly any fluff or bs. Well done.
15:00 "2ms from mouse click to the final image on monitor" are you sure about that? x))
they can somewhat emulate mouse movement (scene + crosshair") before new frames are even ready, but they can't emulate mouse click.
yes, wired mouse 1ms and 500fps rtx+monitor = 2ms, but will have nothing to do with nvidia tech.
yeah that claim is total bullshit, it's currently looking at 60ms+ with MFG enabled.
If I don't see Jensen pull out a can of multipurpose spray, then they are not justified purchases.
Well I think he may have already done such and sprayed the jacket!
@@techyescityYea but the jacket won't fit into my PCI Express socket. I tried a different one and it just split the socket in half.
@@MarcoGPUtuber haha 😂 😂
i find it extremly funny talking about justification..i will get one because i can and because it makes me happy. that’s all justification i need. there are far more expensive hobbies than high end gaming, its a 2000$ not a brand new mercedes for gods sake 😂
@@medovk Yeah PCs are MUCH cheaper than cars! Though still have to not give companies blind reign.
The 5090 is gonna need a case that has a lot of air volume move through it in order to dump that heat out of the case
its like do you go with the FE version being a couple hundred cheaper but also extremely smaller vs AIB that are extremely larger that you would think might beagle to dissipate heat quicker but obviously its an engineering marvel to have a 2 slot 5090 so im sure it can also cool it.
@deadpool790 According to some it's supposedly has the most expensive heatsink on the market so if we subtract that from the equation and add a non reference model which shouldn't cost that much more than the fe model should be either at MSRP or a few hundred dollars more based on the 4090 aib models which were available at $1599 at launch and cost $150 more. for models like the Suprim liquid 4090
Hey Captain Obvious, if you can afford a 5090, whats the odds you'll realize this?
Same was said about my 4090. It's always running cool. The 5090 will be the same. ✌️🇺🇲
Coolers in the top of the case to move air out also from the top!
There is a 24gb 5080ti “leaked”. Too game at 4k even with DLSS not sure standard 4080 is giving enough VRAM headroom, especially with all these features. I might wait to see what happens there.
3:10 In all fairness, bud. There's no "perhaps" about it. You aren't going to get a massive increase, as they did quietly slip in some "actual" render performance of 4090 vs 5090, and it was roughly 25%.
Yeah buddy. But 40 series was already huge on raster, beter than AMD in every class. And thats what all those watts DO. That part is already perfected but if you think you are going to get playable RT performance out of raster boost youre crazy.
Agree with u Bryan, the 5080 & 5070ti are the sweet spots. Let hope there are cards available at MSRP
People should look up Jim Jones and the Jonestown massacre. Don't drink the Kool-aid. $2000 for a GPU? GTFO. $1600 was absurd enough.
If Jensen tells me it's time to buy, I buy. The more you buy the more you save!
2000 dollars before taxes...
@@maxview99damn straight, im buying 5.
I mean, 32GB of GDDR7 with 5000 more cuda cores, new tensor cores and better RT cores... the 5090 is nowhere near necessary to enjoy the last AAA games, is a workstation card at this point. So if 4090 was $1600, 5090 being $400 extra is not *that* crazy.
I am more concerned about the 5080 being almost *half* 5090, costing the same as the 4080S, so we're not having much increase in price/performance in this generation outside of the 90 class. The same goes for the 5070Ti being basically a 4070TiS, 5070 a 4070Ti and so on.
@@KnightofAges And not including the added cost the premium board partners add on to their models either.
Usually the base MSRP FE models are available in fairly small numbers as compared to overall demand and are out of stock much more often than they are available for the highest tiered cards.
Cost will likely run more along the $2200-$2300 USD + taxes for the majority of 5090 models.
First off the 5090 versus the 4090 there's a 21% increase in cuda core count with a recommended power supply increase from 850 to 1000 watts. Pure rasterization power will be right around that 21 percent increase (give or take) without DLSS 4 mult generation as I'm sure 3rd party testing will confirm. Other than the founders edition cards the other cards will be anywhere from $2500 to $4000 on release day for the 5090.
2 questions. What games that are worth playing even need a 50 series. 2. What cpu can even keep up without melting down?
We were still GPU bottlenecked with RTX 4090 so those same CPUs will just work fine.
None presently. There wont be any graphically intensive games that are pushing new technical bells and whistles until probably GTA6 or Witcher 4
Generated frames are made only with the GPU, not a mix of GPU and CPU like the "real" ones. So if your CPU can handle 120 fps, then will be enough to go up to 480Hz. One pro of FG that little people consider is that FG is a way to go over CPU limits. I had an i7 8700 (no K) + RTX 2060, so going at 120fps (40 base and "x3" in lossless scaling or a little less with FSR3) in games like Space Marine 2 shouldn't be possible with such CPU. So like RTX 4000, Zen 4/5 or Alder Lake and Beyond should be enough, especially a Ryzen 7 or high tier i/ultra 5 (the higher, the better obviously).
Yo bryan, with all these new gpus sucking up 400watts, that is like a third of my house allowed electricity. Can you do a watt/$ or watt/fps video?
Yeah definitely will take a look at that one for you on these high end suckers.
What types of discounts will the 4070 to 4090 series once released?
No discounts… retailers just increase 50 series gpu prices to compensate…
$4,000AUD for a graphics card is hilarious. Nvidia have no respect for their consumers, and who can blame them?
Seeing how high the prices are for GPUS are getting i'm getting MORE keen on APUS waiting on Strix Halo varients to fully mature and get better before I jump ship to a MINI PC with an APU.
What respect? Did consumer ever give respect to dozens of gpu maker that used to compete in the market?
@@Epic3032those Strix halo will get stomp by AMD own gpu if they ever come to mini pc.
@@arenzricodexd4409 I could see MINI PC become popular as Handheld PC is getting these days. especially with high powered APUs coming in the future
They should respect you by cancelling 5090, so the most you can spend is $2,000AUD.
The 4090 undervolted quite well while maintaining good performance. I'm curious to see how the 5090 does in this regard.
at the long end.... 6 - 7 years I upgrade. On the I am eager end of the scale, 4 years. I have a 3090 and guess what that means... 5090 ASTRAL baby.
What song does he have playing in the background it’s really soft
Many thanks for your nice advice!
5070 ti will be the hammer for gaming
5080 will be the most selling
5090 will be more or less for the TH-camr and professional channels
Please try to add one very important notice
Which model will have LED light !
Many thanks ❤
I don't care about fake frames. But I'm interested in the raw performance of the 5090
It's a measley 28% better so dont get your hopes up too much.
@parkersgarage4216 28% is a great jump
Dude, you HAVE to look at the 4070 SUPER, relative to the 5070. The stackup is... kinda weird to say the least lol 😏
wym?
@@jdmnissan 5070 is comparable to 4070 Super, which is why Nvidia deliberately didn't mention it and instead focused on 4070 for comparison. 5070 is basically 4070 Super with Multi Frame Gen capability. Previous Gen will be getting DLSS 4 as well, only difference is Multi Frame Gen.
@@Berserkismthis "only difference" is gonna be HUGE difference
Reflex 2 is a feature Nvidia took from VR gaming called Asynchronous Spacewarp, it basically takes the input just before the frame is rendered and they shift the frame accordingly, sometimes chopping part of the frame off. However Nvidia added AI onto it, by letting the A.I. filling in the chopped off portions of the frame. This should reduce the overall latency of Multi-Frame Generation to feel very responsive despite having poor latency. Maybe from 57 ms to feel like a 35-36 ms. Which would be pretty decent add on for all games regardless if you turn on MFG at all.
The gpu steve did the video on was in fact blow through. However a major difference also is the unreleased had the pcb on the side not the center. I'm sure that will effect the blow through effect as well.
The reflex2 thing is strange but not really. I had an idea back then similar to this, but my idea is for frame generation and it is actually more involved than just using current frame + mouse movement (it is more integrated into the engine) because without it, it is basically a technique called asynchronous reprojection (it is being used in VR space). Current frame generation works by delaying the current real frame from being displayed in order to calculate and display the fake frame. Async reprojection in VR space works very much like reflex2 and more since not only it is being used to reduce latency, but also for generating frames. The biggest difference between reflex2 and the current implementation in VR is this inpainting thing where obviously it will need AI to fill in the blanks while in VR it usually introduce defect like edge warping + the fact that in VR it is also being used to generate frame. My idea is more about this kind of frame generation being integrated into the game engine directly thus instead of trying to render high quality frame every time, you render high quality frame and then you render in between frame in a more crude manner and combine it with the previous rendered frame data. It is kinda like async reprojection but instead using only the previous rendered frame data, you use that + current frame data to fill the holes. This will not reduce the input latency like reflex2 or async reprojection, but it will give you more frames without extra latency (basically 60fps with this idea is similar to a real 60fps) and hopefully better quality than reflex2. But I assume if AI is good enough to paint those holes then might as well drop the game engine integration (since it is more costly because of the rendering cost) and go straight to async reprojection combined with AI inpainting.
Gonna dig out my old 2060 cause its = performance to a 4070ti with lossless scaling maybe it can raytrace now
Lossless scaling is awesome and they just released a new version.
How is it awesome? You can definitely notice the input latency
@@evilleader1991Latest version improved latency
It isn't huge. And it doesn't seems that the architecture does much for gaming. Gains are mainly direct compute, AI and ray tracing
Why didn't you compare it to the 40 Super series? Pretty disappointing specs, I'm not impressed.
Guys i need big help right here im buying a new gpu should i get RTX 4060 or RX 6700 XT bith with Ryzen 5 5600 X for gaming
Anyone knows whats better for me please reply
The best value card for gamers should be at a lower price point and not a higher price point, that's for enthusiasts.
I don't know what is TH-cam is on, but i just got an intro ad FOR AN AD for this video 💀
Interesting that the TDP for the 5090 has gone up but the recommended PSU has dropped to 850W. I seem to recall 40 series cards had an issue with power spiking so maybe they've gotten that under control with this gen.
He has got them the wrong way round.
@@virtual-adam Right, thanks for that 👍
I finally had some success bringing a GPU back to life by baking it. In this case it was an HD 6970M from an iMac 2011 27". Miss those videos.
As a only Radeon GPU buyer since years it is sad to see Nvidia killing the iconic xx70 series 😞
Great video. When can we expect the 5090/5080 benchmark?
After Jan 30
Happy new year ! Hey what’s your thoughts on the clock speeds of the 4090 vs 5090. Will the lower clocks be worse for raw performance when your playing certain games that’s actually do take advantage of higher clocks. Specifically call of duty titles. I don’t like frame gen, I always leave it disabled. I’m really struggling to wrap my head around how we get less latency also with generating frames.
Only 4k gamers are going to see 25-30% increase with 5090, 1440p gamers are going to get disappointed it seems like, 10% at best at most GPU bottlenecked titles.
I'm more interested in the Founders Edition cooler designs than the actual cards themselves. Will be very interesting to see how Nvidia cool almost 600w with just a 2 slot design.
The 1/4 performance gain with 1/4 price mark-up will make 4090 stay in value for sure.
I already have the cash for the 5090 i would love the FE but doubt i will get my hands on it
We have to wait for a super refresh, like, every single time to hope for more vram i guess.
So far, only two(?) of those refreshes have given more VRAM:
4070 Ti (12GB) --> 4070 Ti Super (16GB) _+33 %_
2060 (6GB) --> 2060 Super (8GB) _+33 %_
Even more curious will be the 50 series VS the 40 Super series. Because their specs are a lot more comparable.
Whats up with PSU requirements? 1000w for 4090 (450watts) vs 850w for 5090 (575watts)?
They were probably able to minimize transient spikes and they’re assuming that you’re not running a 14900K with the GPU. That’s my guess at least. :D
@@Simon_Denmark no one is running such intel cpu anymore, a total failure of generation from intel with 400 watts to self destruct the cpu
LMAO!! My 14900ks has been amazing with my 4090 aorus master. They max out 900 watts both overclocked on a benchmark. The 5090 with a 14900ks oc won't pass 1000 watts. ✌️🇺🇲
@@bm373 LMAO!! ✌️🇺🇸
@@Simon_Denmark ✌️🇺🇲
Honestly, if they want to ai generate frames, so be it, as long as it still looks amazing and has low latency. If they can accomplish that, whatever, if not its trash. My only other gripe is that the only good looking card is the 5090, no 384 bit card, amd that gap to the 5080 is absurd. Im still looking to maybe upgrade this generation, but thats a maybe until i see benchmarks
For 5080, Core count and tdp both increased around %10, clocks are not too much of an uplift too. So they didn't bring any efficiency this series. GDDR7 will do the heavy lifting I guess, we need some benchmarks with underclocked vram vs 4080.
increased density but increased leakage i.e more transistor in same space leak more.
Seems like this series is mostly a software upgrade. Raster improvement is minimal. These fake frames will be great on slower and story driven games but racing and competitive it won't feel any better. Also framegen doesn't help VR
Even single story games is doubtful, there are people that are more sensitive to input lag compared to other, so Nvidia pushing generated frame as main sellingpoint is weird, to those people, buying 5070 won't be like buying 4090 if they don't plan on using framegen.
@tomthomas3499 the imput lag will be the same as the base frame. So a base of 60 will still be good even if it scales up to 144 for example it's not like frame gen but into tvs that makes input lag worse.
@@ivan-j4l9x The problem is what you see and what you feel is different and hence why it feel odd to them, Daniel Owen said framegen should be treated as motion smoothness not performance improvement and i agree with him.
no that is still hardware. The intermitten frames are generated not done by software on the shaders.
@ABaumstumpf obviously some hardware is involved. My coment was an exaggeration and more related to the input latency that is not improved.
when will we see testing benchmarks?
be on the 29th most likely.
Lovelace is made on the N4 while Blackwell is made on the improved N4P silicon. Below is from TSMCs website.
The 4nm (N4) process technology is an enhanced version of 5nm (N5) technology with density improvement. It started volume production in 2022.
To further enhance the N5 family’s performance and power, TSMC introduced N4P and N4X, targeting next wave 5nm products. N4P offers an 11% performance boost compared to N5 and entered risk production in July 2022. N4X is the first in the ‘X’ lineage of TSMC’s extreme performance semiconductor technologies, with 6% speed gain over N4P at moderate leakage trade-off, and volume production is planned in H2’2024.
The N4, N4P, and N4X nodes are design rules-compatible with 5nm technology for easy design migration.
They are still intermittent frames. People are mad because they think they pull next 3 frames from somewhere. That picture in nvidia slide is misleading also. They render 2 and AI calculate the transition in between. Correct me if wrong. (I have been trying understand this)
I think the PSU required is mixed up between the 4090 and 5090 on your spec sheets
Does seem odd. The specs for the 4090/5090 (clock speeds down), 4080/5080 (clock speeds up), 4070/5070 (clock speeds up) seem odd too. Not sure if that is nvidia, or these tables.
@@Skobeloff... Clock speeds correct
@@Adam-hn2iw Fair enough, assumed it was a nvidia thing.
Just wanted to share a thought I had: If the flow through design of the 5090 cooler is integral to keeping temperatures in check, then the Lian Li x Dan A4 H2O is may not be a very suitable SFF case to put it in considering how the vertical mount will cause the card to blow all the heat to the back of the motherboard and PSU. You could flip the PSU fan side to face the GPU to allow for better flow through for the GPU but that still only provides some flow through to one of the two flow through sides. I could be completely wrong about the temps being bad in a case like the A4 H2O though.
I will definitely just FAFO with it! you will get to see the results none the less :P
@techyescity I love FAFO 🤣
I'm totally sold for the RTX 5070Ti as its well balanced for the jack of all trades IMO. Gonna get me one of those MSI variants for sure
I run two A4-H2O's + Strix Z690-i + Strix LC240 + Corsair SF750; 12700K + 64GB 6000CL26 + 3070OC, 14900K + 96GB 6400CL28 + 3090OC. They make for pretty sweet sleeper builds.
I wonder if they'll have surprise pre-orders start a little early...I also wonder if I'll be able to get one before a scalper does...
That 2ms is not end to end latency. It's just the system latency. You still have the output and peripheral latency to contend with. The end to end latency is probably pretty close to the 5ms you've been able to get as well. Maybe 1ms faster.
W video love your content man, I’m hoping for gaming the rtx 5080 is as fast a 4090 but for thing like video editing and stuff like that which I do as well (at 17) lowkey pretty fun. The dual encode and decode will be really nice plus the extra vram and other improvements vs the current rtx 2080 ti I have now
Happy new year. Cards im happy about 5070ti and 5090 they are value for money
I'm excited for reflex 2 as well. And interested to see the raw performance uplift on Rtx 5070ti and RX 9070xt. I don't really care for upscaling.
Only reason I’m upgrading is cause I’ve started to really enjoy single player games like cyberpunk, Alan wake 2, black myth wukong etc and I game at 4k and have been using amds 7900 xtx I bought it on its release day, it’s been great so far only issue is ray tracing, and upscaling methods. Nvidia clearly has taken the cake on that and seeing the 5000 series tech I think it will be a good upgrade cause I really want to get into some 4k path tracing, going for a FE 5080 cause 2000+ dollars for a gpu is just insane to me, regardless if it’s the best in the world
Prepare for disappointment. Path Tracing is not what you think it is, it's slow, inaccurate, blurry and performance killing, just not worth it.
@ my brothers got a 4090 and path tracing always BLOWS my mind, and the fact the 5000 series isint a rasterization upgrade its upgrades are all for ray tracing, path tracing and improving dlss, ghosting, and fracturing, so even if I got only a 4090 I’d be impressed but I know for a fact a 5080 will outperform a 4090 in ray tracing let alone a 7900 xtx I can’t even turn regular ray tracing on 😭 so I’m very excited
Want more info on the 5080 ti
Once the benchmarks start coming out with REAL performance without upscaling frame generation etc.
I'll be choosing between a 5070Ti or 4080 Super depending on price/performance.
No way I can justify a US$1000 plus card I'm afraid.
Your thoughts on Unitek cables? I find them pretty good for the price
I am holding out for the 9070 xt. FSR 4.0 looks really good.
Hi @techyescity will my 650W PSU support RTX5070ti? im pairing it with r7 7700 and 240 AIO cooler
Happy New Year
Prices in Singapore is going to be hell as well, and it's probably the same old thing as with Australia.
Higher than currency converted MSRP and longer supply chains. At least in Singapore if I remember correctly the cards go through an importer + distributor before hitting the shelves and everyone wants profit so yeah, we're cooked
Remember when the 80 series cards launched around $500 usd.....those were the days
For the 3 years happy with 3080ti
$4000 AUD for a GPU.... just to play a computer game....have you looked at the current crop of games? what computer game justifies this expenditure? Madness.
People will still buy it
For real, I decided to just get a cheap 2070super I found online for my main rig (125euro), it's basically as fast as a 4060, which isn't a powerhouse at all, but I stick to 1080p, and it's honestly been doing fine so far. The Vram is not great but for 125 euro I can deal with that.
Jenson says the 5090 is for people who have 10k usd rigs 😂
looking at it this way, perhaps gta 6 if it comes to pc, and some future games justify the investment
but in general, things are looking pretty bad, yes
Sadly, for most Gamers, even RTX 5070 Ti is too expensive. *If one day in the future...* AMD or Intel will be able to compete in the High End, prices might come down a bit.
Until then, Nvidia can freely choose any numbers it sees optimal (for Nvidia).
If both a decent product and smartly priced, Radeon *9070 XT* (16GB) _might bring some competition_ to the middle class (RT: ~RTX 4070 Ti? Rasterization: ~7900 XT?).
Perhaps even future models from Intel (Arc Bxxx), at least in the lower class.
How does the wattage go up but the psu requirement go down?
PSU "requirements" are about how much the PSU needs to be able to deal with, GPU TDP is about how much the GPU can draw sustained. They are not the same thing.
For one the electrical specifications for PCI devices allow for short bursts far higher than the average. So this would suggest that the 50series simply has smaller power-swings and never spikes that much higher than its normal use.
A GTX 1080 had 250W TDP, yet the powerspikes could go well into the 400W region.
the table is wrong, is 1000w for the new card, 850w for the 4090
I like the 4070 ti. Cos of extreme 1440p on pc, and entry level 4k on a TV style system.
I think it is something that the rest of us mortals cannot buy, at least on my side of the pond, Spain.
I would have loved if they gave us the full 25k CUDA cores GB202 die for the 5090. It could have been another 3090 to 4090 uplift (that was around 65% on average, more in some RT workloads).
I will still buy a 5090. Already built my 9800x3D pc in November. And already bought my 32" 4k 240hz QD-OLED in August.
If the 6090 is a massive uplift, I will just upgrade my 5090 to the 6090 when that one gets released. For now I will be fine with my 5090.
So what performance hit in gaming will there be if I am running pcie 4.0 instead of pcie 5.0.
Normally only 2% or so. It's not that bad.
Should be almost none really even with PCIe 2.0 If it is a full x16 slot
Basically none…
I'm excited about the 5090. It will be a significant boost over my 3090Ti.
It is just really misleading marketing on Nvidias side. They did not lie like some haters want to claim as they did state the differences directly on the slides, but it just makes those graphs all rather useless and misleading.
Comparisons should be like-for-like and things like new features like multi-frame-generation should then be mentioned and shown separately.
How does the 4070 ti super compare?
5070 is slover than 4070ti super and 5070ti is faster. So it drops between those two gpus.
I wonder how the laptop variants will be
bad, as always
Sorry, but what is envidierr?
if you have anything below the 20 series it would be a great upgrade.
I think one segment of the users that are really looking forward to the raw power of the new 5090 is going to be those that are heavy into high resolution VR gaming like flight sims and harder to run racing sims as they may finally actually have a GPU that will not struggle to deliver a smooth ample playable FPS and be able to run the higher graphical settings at the same time.
Granted this will be coming at an expensive price point!
Other than the 5090 though as far as raw performance minus frame generation and using DLSS the 80 series and lower tiered cards seem to be getting weaker as compared to the prior generation offerings of earlier series.
In the past the new series card of one tier would generally match the performance of the next higher tiered card from the previous generation.
No way is the new 5080 anywhere close to matching the prior 4090 offering which when comparing to past series offerings is a step backwards.
So really what Nvidia has done has made the 90 series cards sort of straddle the fence of being a top gaming GPU and/or a slightly cut down professional GPU.
Then the 80 series cards are the top tier really designed as a gaming GPU.
If you go back to the 30 series and the original launch of the 90 tiered cards even back then in the presentation Jensen said that the 80 series was the top "Gaming" GPU of the series although the 90 series could also be used for gaming.
But that separation and distinction was made for the 90 and 80 tiered cards even back then.
AI needs to disappear.
i dont think so
Ai should not be considered as help improve something that they could work on performance instead using ai to hide the real performance
@@Dark.Syndicatea random idiot has emerged!
They include AI cause they dont want to sell to gamers but they also want to mitigate research and development costs with gamers
I think the more appropriate thing to say is that AI needs to be used in different ways. Like focusing much more on making normal DLSS give you more image quality and even more performance (as an example, I don't know if it's possible to give you more real performance with AI). I agree frame gen using AI is not the way
Thanks for sharing your thoughts Bryan. I am in the market for a new card and like the proposition of the refresh 5080 (Super or Ti variant). Saying this because I believe it will have 24GB VRAM and not cost me 4K AUD. I just don't think 16GB is viable going forward. Thoughts?
I'm very interested in the mechanical design. I hope that it's something other manufacturers copy because it solves the final "dead zone" in the ATX spec where air recirculates.
By the way, only the 5090 and 5080 have the new cooler design, the 5070 doesn't get it.
I didn't know that about the 5070 cooler, 5080 founders might be the card to get for mini itx.
5070ti is only AIB…
5090, 5080, 5070 get Nvidia FE editions…
@@haukikannel The 5070 AFAIK isn't double flow-thru, just single like the 40-series.
@ is it? Well I am Little bit dissapointed. The biggest upgrade in 50 series is the cooler! That is actually cool new invention! Now 5070 is even more boring…
I'm probably sticking with my 4090 in my main PC, but will most likely upgrade my mini itx 4070 ti living room tv pc to a 5080.