New Leak Reveals NVIDIA RTX 5080 Is Slower Than RTX 4090

[Hxx]

Not even remotely the case. 5090 @ 170 SMs doesn't beat 4090 by 50% in a single game. 5080 @ 84 SMs would need astronomical core clock to even become close to 4090. It's going to be 20% faster than a plain 4080 in the most 5080-favouring scenarios with an average advantage hovering around 10%. Full stop.

lol your math aint mathin because a 4090 is 20% faster than a 4080 in overall relative performance as per TPU. If as you say a 5080 is 20% faster than a 4080, well im no math whiz but it would imply a 5080 is about on par with a 4090 in relative performance lol which is what i stated. Anyway I stand by what i said we shall see in 5 days

 

[mkppo]

What rubbish, the 5070 is going to match the 5090. Fake news

lol your math aint mathin because a 4090 is 20% faster than a 4080 in overall relative performance as per TPU. If as you say a 5080 is 20% faster than a 4080, well im no math whiz but it would imply a 5080 is about on par with a 4090 in relative performance lol which is what i stated. Anyway I stand by what i said we shall see in 5 days

That's a lot of complicated stuff right there but just thought i'd mention that at 4K the 4090 is 29.8% faster than the 4090 in the 5090 TPU review.

 

[Hxx]

What rubbish, the 5070 is going to match the 5090. Fake news



That's a lot of complicated stuff right there but just thought i'd mention that at 4K the 4090 is 29.8% faster than the 4090 in the 5090 TPU review.

agreed i think it depends on the test but i think TPU calls it Relative Performance I am assuming "overall"
NVIDIA GeForce RTX 4090 Specs | TechPowerUp GPU Database

 

[Dr. Dro]

Clearly this is not going to be the case. 50 series is DOA. But there is not going to be an even more cut down 5090. Nvidia wants to sell 5090 at astronomical prices. The rest is a second wave of super refreshes and keep selling that.

If the 50 series are DOA, the competition might as well hang up the ol' hat and go home, pack their bags and enjoy retirement in the Bahamas.

 

[N/A]

5070 5% more shaders than 4070, 5070 Ti to 4070 TiS 8960/8448 same, 5080 to 4080S 5%. 40 series aren't bandwidth starved it seems and whoever bough already has no incentive to upgrade. I'm not falling for this frame gen software shenanigans.

 

[x4it3n]

lol your math aint mathin because a 4090 is 20% faster than a 4080 in overall relative performance as per TPU. If as you say a 5080 is 20% faster than a 4080, well im no math whiz but it would imply a 5080 is about on par with a 4090 in relative performance lol which is what i stated. Anyway I stand by what i said we shall see in 5 days

At Native 4K the 4090 is about 30% faster than the 4080 SUPER in Raster and about 30-40% faster in RT/PT with Cyberpunk even reaching almost 50% at 4K with PT DLAA (due to the game relying more on RT Cores, but it's unplayable anyway, it runs around ~24fps lol).
The reason why the 4090 has such a poor scaling is due to Memory Bandwidth bottleneck. The 4090 has 60% more cores but only 37% more bandwidth therefore all of its core cannot be filled. Also Memory Bandwidth doesn't scale like CUDA Cores or Core clocks either so you would need more than 60% more Bandwidth (vs 4080 SUPER) to fill them! Hence the 5090 having a 512-bit bus and almost 1.8TB/s of bandwidth. But knowing that the 5090 has 33% more Cores than the 4090 it is probably Memory starved too lol

 

[HisDivineOrder]

Didn’t we already know this?



Have you watched or read Mark Cerney’s presentation about raster performance increases?

I’ll summarize for you: Sony and AMD are also saying rasterized graphics are tapped out.

Mark Cerny is saying what he has to say because AMD is also angling to become an AI company. Sony doesn't want to build anymore custom GPU's, so they have to rely on one of three companies and every single one is either an AI company or daydreaming they are one. AMD just hasn't gotten there yet and may never get there if Nvidia's intrinsic advantages remain overwhelming.

There are no graphics card companies anymore and probably never will be. So raster gets brushed aside and AI features get pushed to the forefront and Mark Cerny's just saying what's left to say to try and keep people from realizing what happened.

Or are you suggesting that had the AI boom not happened Nvidia would have given up on the graphics market? Or that the AI boom just so happened to perfectly coincide with GPU raster suddenly being incredibly difficult to meaningfully improve? What a weird coincidence, that.

 

[Metroid]

No news to me, I knew it would be slower. Only idiots believe what nvidia say and this is not new, nvidia has been doing it for a long time.

 

[FamousFighter]

I hope you're being sarcastic because otherwise the delusion is real lol

No I was being serious and that sucks the 5080 can't match 4090 performance.

 

[Icon Charlie]

No I was being serious and that sucks the 5080 can't match 4090 performance.

True but IMHO this is part of the part of Ngreedia's plan all along. Because...

1. They knew this series was crap which is why all of the BS market speak.
2. There is not a lot of them being made right now... LIMITED INITIAL RUN to keep prices up with what they have out.

The other reason is once again MARKET SEGMENTATION.

They'll crank out the SUPA and the TI later. Not a lot but juse enough to keep the market prices from falling down hard.

 

[Outback Bronze]

So the 5080 really is just a 4080 Super Duper with Updated Framegen with no price increase.

You know what, this is starting to remind me of the 2xxx series where the RTX 2080 was equal to the 1080 Ti and the 2080 Ti cost a bomb. But check out the prices back then vs AIB cards. The FE is more expensive. This seems to have been reversed by what cards (5090) W1z has just published.

RTX 2080 Ti	$999	4352	64	1350 MHz	1545 MHz	1750 MHz	TU102	18600M	11 GB, GDDR6, 352-bit
RTX 2080 Ti FE	$1199	4352	64	1350 MHz	1635 MHz	1750 MHz	TU102	18600M	11 GB, GDDR6, 352-bit

 

[lexluthermiester]

A set of newly leaked benchmarks has revealed the performance capabilities of NVIDIA's upcoming RTX 5080 GPU.

When official review benchmarks are shown, I'll believe it, not until then.

 

[AsRock]

They have no incentive to do so. They are turning into Intel when AMD forgot how to make good CPUs, too bad for now AMD forgot about GPUs this time. That and AI craze.

Let's hope that next gen we will see competition from top to bottom.

Yes they are just trying to buy time in hope that they will come up with some thing.

 

[starfals]

Great. Not that we all didn't know this was coming but still. I was kind of hoping it would just repeat what every other 80 card did in the past. Passing the last generations easily. This is the first time i see an 80 card that can't do it. Why is it so cut down? Why can't at least past with 1-2% or at least be even? 5080 will give us last gen performance. The only next gen FPS will be in 5090, which is stupidly priced. DOA generation. F off Nvidia. Same for AMD, what a fail of a company. Nobody will be getting my money i guess, and im here and waiting to give it to ya. Give me a real next gen card and you get 1000-1300 bucks. It's not exactly peanuts money either. They can't satisfy a power user like me. They only make happy the budget users under 1k, and the rich people above 1600. Give me 20GB, faster than 4090 and u got 1300 dollars today. Don't do that and i wont give you a dime.

Cya in 2027 i guess.

 

[alwayssts]

More like fifty 8%-y....amirite?

I know...'boo this man'. But fair is fair.

All it really has to do is beat the maximum potential of N48 when itself is running stock...which I'm sure nVIDIA will make sure it will by .00000001% (only minor exageration) on avg.

They know that, you know that. We all know that. That's how they do it. Also, gotta make it so 9216sp/18GB/192-bit is an upgrade on 3nm and piss all the then-5080 owners off.

It's the same old song and dance.

Buy nVIDIA card or buy cheaper AMD card and overclock to the speed of the nvidia card, more-or-less. AMD even knows this, and will probably adjust their capabilties as such if possible I would imagine.

I understand some people have different points of view, but to me the answer has always been obvious.

I'll take the $500 16GB 4060Ti, the $600-700 4080 20GB, and the >4080 but couple hundred dollar less 24GB.

Yeah, you call them overclocked 7800xt/7900xt/7900xtx...whatever.

Or you can buy a good-but-expensive nVIDIA 90 card for 2+ cycles/2x margin for 2x most-any would pay and/or overclock a similar-performing/more-spendy nvidia card for no tangible resolution/fps gain.
Props if you can justify the former.

Have no doubt this gen will be similar, and in short order we prolly won't even be having the DLSS/FSR argument (for the most part).
Still just about RT. Will still argue it doesn't matter until next-gen of cards (3nm) when it coalesces.
If you want in early to pay for mostly nVIDIA-paid showpieces most people don't actually play, more power to you. You keep running through that one 2077 spot, gamer. Enjoy your 'bar' bench(mark).
BTW, does nVIDIA demand that for review samples or do people subject themselves to that willingly? If they need help, blink twice next time you talk about it. Maybe we can send The Fixer.
Ok, or maybe just get demands amended.

9070 OC probably won't beat a stock 4070 Ti Super/7900xt but be damn close and a hell of a lot cheaper.
9070XT OC will probably not quite hit a stock 5070ti, but damn close and a hell of a lot cheaper.
If there is a 9070XT(X/I) it will probably not OC to beat the stock speed of a stock 5080 but damn close and a hell of a lot cheaper.

Give or take damn close with actual parity to insignificant amounts faster. At least, that's my theory of how nVIDIA is planning for it to go down. If AMD can do it IDK, but that's certainly the hope.

IOW, replacements for the 7800xt/7900GRE, 7900xt, and 7900xtx but with less ram in the later cases (and less absolute [OC] performance) to facilitate a cheaper price.

AMD got it right the first time, but cheaper is good. I still think the 7800/7900 series will be good value wherever they settle (given the extra ram, their own OC potential, and if given FSR4 support).

I don't think anything is going to keep me from holding out for the next gen and whatever is ~96-112 ROPs.
Pair something like 6*1920sp/11520sp and 256-bit/40gbps 24GB of ram, that'd be choice; a 4090 for conceivably 1/2 price.
Nvidia might make something slightly better, but I bet the cut-down model is similar to that. Plus, knowing them they'll cut it down to 224-bit and 22GB or something.
I truly do think >10752 on 3nm will be good-nuff long time.
Unless someone wants to sell me a 4090 for that price earlier, but I don't think that's going to happen.

I like the potential AMD options this go round (in theory), and wouldn't fault anyone for buying them (especially if you upgrade every gen or think 16GB is enough for your setup for awhile)...
...but these GTX50 kinda look a joke when you think about what's coming in a ~year or the comparative value wrt raster/vram.
I still think nVIDIA only makes sense if you have to have this-gen addative RT and are willing to pay for it.
In both price AND higher-end playable performance (compared to w/o). Some people might be; this guy ain't.

I get it, 'you can wait forever' (which is a straw-man argument bc it's really about when to jump on to get the best value/longevity)...but next-gen is we'll probably see sane-priced 4090 performance and...
...>16GB vram. Even >16GB on 192-bit cards (if they exist) and 5080 perf. The fact 5080 (even if next-gen same-bracket will have slightly-higher raster performance) doesn't have 24GB is borderline criminal.

Naw. I'll wait. For high-end performance for what high-end cards USED to cost...or i'm *personally* willing to spend for a decent 4+ year card.
If I'm going to spend that much it better be at least parity PS6 in every possible way. While that system might have ~32GB of combined memory, I doubt it will need use more than 24GB for the GPU.
16? Prolly. 18? Idk.

To me this all adds up to "wait"; for 18GB 192-bit 5080-performance at the least.
For price/VRAM alone, if not stabilizing RT performance (which even if 9216sp could be higher than 5080 potential given potential clock differences/capabilities). I don't know if many people realize that.

edit: damn formatting...I thought I had it perfect. Pardon the awful line breaks as I don't feel like fixing it. I know it looks awful and is probably a pain to read as-is. Points still stand.
2nd edit: Cleaned it up a little anyway because I'm anal like that.

 

[Zazigalka]

not surprised at all. 4080 + 20% is not 4090.

I am so regretful that I sold my 4090 for $1400 two months ago to buy the 5000 series, I don't know what to buy now. I may go with AMD this time to protest Nvidia.

yeah.
protest ? what exactly ? the fact that you have ants in your pants ?

 

[Broken Processor]

But no no no it's faster than a RTX 7090 with DLSS 4 enabled! /S

 

[ypsylon]

Well IMHO anyone on higher end 4000 series shouldn't upgrade at all. It's not any shocker that 5080 is slower than 4090. This release is more for folks still on 2000/3000 series. And if you need VRAM then 5090 is: YES.

Me personally maximum CUDA and VRAM which makes 5090 obvious jumping from 3090 for 2.5x rendering uplift/card. But we'll see still plenty of optimizations when it comes to compute in next 4-6 months so it may be even nicer. For gaming surely nobody needs 5090.

 

[JustBenching]

This is the first time I can remember an 80 class Nvidia card not matching or beating the previous flagship even the weak RTX 2080 at least came out slightly ahead of the 1080ti.

#progress



It's sad even though they could only do so much with the 5090 without making it 800mm2 they could have done more with every card below it a 1200 usd 4090 like card would have been fine maybe they don't want it all that close to their precious 90 class anymore.

Hopefully reviews are better than expected but I'm not holding my breath.

Yea, don't know, im one of those people that don't really care about my card losing value, so i'd love for the 5080 to slaughter the 4090 ---> better / heavier / nicer RT in games. But this gen was a disappointment. I like the performance of the 5090 but man, the power draw is insane, and it doesn't look like you can really scale it down. It loses up to 26% (!!!) by being locked to 400 watts. Imagine what happens at the 320w i wanna run mine at. I think the new and more vram might have something to do with that.

The biggest thing this gen was the new DLSS, the performance mode old vs new is huge. I legitimately can actually game on performance mode, it doesn't look totally crap like before. Not saying that it looks great, but i think it's better than playing at 1440p native on a 1440p monitor.

If the 50 series are DOA, the competition might as well hang up the ol' hat and go home, pack their bags and enjoy retirement in the Bahamas.

Well 50 series is a disappointment, amd isn't.

Spoiler

In order to get disappointed you have to have some expectations. Who has expectations from amd other than "nvidia in raster for 9,99$ less money and less features"?

 

[oxrufiioxo]

Yea, don't know, im one of those people that don't really care about my card losing value, so i'd love for the 5080 to slaughter the 4090 ---> better / heavier / nicer RT in games. But this gen was a disappointment. I like the performance of the 5090 but man, the power draw is insane, and it doesn't look like you can really scale it down. It loses up to 26% (!!!) by being locked to 400 watts. Imagine what happens at the 320w i wanna run mine at. I think the new and more vram might have something to do with that.

The biggest thing this gen was the new DLSS, the performance mode old vs new is huge. I legitimately can actually game on performance mode, it doesn't look totally crap like before. Not saying that it looks great, but i think it's better than playing at 1440p native on a 1440p monitor.

I'd honestly love if the 550 usd 5070 was as fast as my 4090 dgaf about resale value either I care more about progress.

I haven't tried the new Tramformer model but personally only like quality DLSS even at 4k or 1440p UW DLAA. Hoping the new version makes balanced 4k usable for me.

 

[alwayssts]

You know what, this is starting to remind me of the 2xxx series where the RTX 2080 was equal to the 1080 Ti and the 2080 Ti cost a bomb. But check out the prices back then vs AIB cards. The FE is more expensive. This seems to have been reversed by what cards (5090) W1z has just published.

RTX 2080 Ti	$999	4352	64	1350 MHz	1545 MHz	1750 MHz	TU102	18600M	11 GB, GDDR6, 352-bit
RTX 2080 Ti FE	$1199	4352	64	1350 MHz	1635 MHz	1750 MHz	TU102	18600M	11 GB, GDDR6, 352-bit

I think FE were originally binned better. IMHO, and I don't know this for a fact, I imagine AIB's threw a stink. Now they get the better bins but nVIDIA undercuts them by keeping same margins per chip.
See, a key factor in business is make the other side think they've won...and you can imagine how AIB's may have thought that by getting better bins/allocation after initial FE releases.
Can't you see Huang telling AIB's in the beginning "This is a novelty, we're not going to compete against you..."..... .... *under breath*...."right now." Fast forward to today.
They were originally sold as just that; a novelty. The more you spin, the more you margin. Novelty wears off; undercut partners competition bc you can. Shrewd, but makes sense. Literal cents.
Isn't it also fascinating it's not only the cheapest, but also clearly a metric shitload of design effort was put in to make sure it's a two-slot card their 'partners' couldn't compete with in size (right now)?
I laugh when people still haven't figured out how Huang does business. Not talking about you, I just mean in general. This is the guy that cleverly used his CES speech to get cheaper memory pricing.
AMD likely would've said "We don't use Micron ram because it's literal comparative shit. They can't even make 20gbps GDDR6 and needed pam4 to make it which is fucking bananas."
nVIDIA be like "Samsung/Hynix be cheaper, don't care if not shit...(actually we do, but we'll literally use/promote Micron if you don't)." BOLO 36gbps ram on next-gen nVIDIA cards when AMD 40gbps...
You think that larger L2 is a design reason? Yeah, it is. The design reason is because they can buy cheaper ram from Micron that doesn't perform as well...by the difference of the cache between AMD.
(Sorry that Hynix sale sheet loads so slow, they usually keep that shit pretty well-hidden now-a-days, so that's the best I can do.)
The man is a literal genius, he truly is. There's so many things that go on that people don't notice wrt their choices people think are coincidence, happen-stance, or 'slips of the tounge'.
EVERYTHING he/they does/do is planned wayyy in advance. Sometimes years, with plans to use as leverage or acclimate the market to their what's financially positive for them. Just like RT, etc.
I could go on for literal pages about how many things he does which almost nobody appears to notice. Sometimes people complain about them, but he's taught people to think it's a 'joke'. Genius.


FWIW, The value in 1080Ti/2080Ti was on the used market because of the cost principle you see. The price dropped SUBSTANTIALLY pretty quick and were the best deal for a good, long while.
For instance 2080Ti looked bad vs 3080 bc that's how nvidia do but given 8nm was shit and 12(16)nm was not, and the overclocking difference was massive...the performance difference was neglibable.
Pretty well-kept secret, I think, but dare to compare (even look at Wizard OC scores for both cards). IMHO it did lead to the idea of the only really good value for nVIDIA cards:
1. Buy 90-class new (or late-releasing Ti back in the day) on new node...which are absurdly overpriced but specced over the rez/fps cutoffs at maxed settings in most games vs their counterparts.
2. Buy 90/Ti used when next 80 comes out for cheap bc it looks bad in reviews. That 90 will still perform often at the same perf tier as that 80 and perform better than the new AMD competition for less.
3. This excludes 3000 series bc it's kinda shit. Not trying to be mean to those owners...It's just that Samsung's node sucked and the scaling doesn't line up with everything else before or since.

Won't hold true this gen, but will for 3nm vs 4090. That said, 4090 will have lasted the original owners two generations instead of one. In total they could last a very long time simply bc node/ram.
Huang is right in the regard of "the more you buy the more save" in the notion that upper Ti/90 cards truly are set up to last an extra gen, hence the upcharge.
When you own a market, this happens. They can expect and/or seed and/or pay devs to target whatever they want to seperate the performance tiers of their cards, and they do. Sometimes by 1fps.
There's a reason why the 2080ti is still on the review graph and still averages 1440p60 according to W1z's suite. They want to hammer that point home about the 'highest-end longevity' to excuse price.
5090 is wonky simply bc it's simply limited by power; looks very back-ported from 3nm. Weird card. Cool to see the best performance possible on 4nm though (factoring in higher clock for yields).
Don't get me wrong, I'd never buy one new...but I see the appeal for those that want the best or see it as a true investment on a new node. 1080Ti took forever to die, same for 2080Ti.
4090 probably the same. It's not by accident. Neither is the fact they know you won't upgrade for another generation, hence the added margin of that lost sale factored into price.

It's stuff like this that makes me recommend AMD.
What nVIDIA does wrt marketing/planning, it truly does work. But it's dirty bc so much of it is limited/false innovation to spur the largest margins...It's unnatural how products seperated.
(and obsoleted to the connoisseur...outside a '90'.)

I loved reccomending the 2080ti to people when it was a steal, but I don't know if that'll happen with 4090.

6800xt are cheap and damn good for what they are (if you overclock them they're VERY similar to 7800xt OC; good-nuff for many; essentially a stock 4070Ti with 16GB in raster).
Also, 9070 vanilla will probably vicariously cause those same cards to be even more cheap before-long, so that's probably the new value basement, without a doubt. Not bad for ~/<$400!

 

[Vayra86]

Gently caressing my crystal ball here.

I still think the 5080 would match the 4090 performance . In apps/games where the 4090s wider memory bus and added memory play a greater role those I would expect a 4090 to win but otherwise I expect a 5080 to match or surpass it . We shall see how this plays out next week

You can stop thinking. Just apply logic and count the shaders. Its not rocket science.

This is the first time I can remember an 80 class Nvidia card not matching or beating the previous flagship even the weak RTX 2080 at least came out slightly ahead of the 1080ti.

#progress



It's sad even though they could only do so much with the 5090 without making it 800mm2 they could have done more with every card below it a 1200 usd 4090 like card would have been fine maybe they don't want it all that close to their precious 90 class anymore.

Hopefully reviews are better than expected but I'm not holding my breath.

They simply keep the 4090 relevant and by doing so, sell the illusion the card holds value and imply the x90 are very cost effective. Remember... the more you buy the more you save. This has rang true for the high end which should be read as 'more'. The midrange is a huge clusterfuck now, complete stagnation, its just a replacement for anything that broke.

Considering there is no real difference between Ada and Blackwell besides cool powerpoint slides selling half truths, the gpu stack is populated adequately with a 1k, 1,5k and 2k $ card.

 

[dyonoctis]

Mark Cerny is saying what he has to say because AMD is also angling to become an AI company. Sony doesn't want to build anymore custom GPU's, so they have to rely on one of three companies and every single one is either an AI company or daydreaming they are one. AMD just hasn't gotten there yet and may never get there if Nvidia's intrinsic advantages remain overwhelming.

There are no graphics card companies anymore and probably never will be. So raster gets brushed aside and AI features get pushed to the forefront and Mark Cerny's just saying what's left to say to try and keep people from realizing what happened.

Or are you suggesting that had the AI boom not happened Nvidia would have given up on the graphics market? Or that the AI boom just so happened to perfectly coincide with GPU raster suddenly being incredibly difficult to meaningfully improve? What a weird coincidence, that.

Don't forget how coincidental it is that TSMC decided to jack up the price of their wafer, and ASML also jacked up the price of their machines, which caused TSMC to start to complain.
ASML’s new machines are too expensive: TSMC - Taipei Times

 

[AusWolf]

The 4090 has 16k cores. The 5080 has 10k. They're made on the same node, so clock speeds should be similar. So in what strange parallel universe is this a surprise?

 

[Krit]

When official review benchmarks are shown, I'll believe it, not until then.

Just look at raw specs it's not a rocket science to say that there will be a small jump besides that nvidia already published ~15% jump. RTX 5070 Ti will be slightly better in performance uplift in general it should be better than RTX 5080/5070