• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

GeForce RTX 5090 is $1,999 5080 $999 5070 Ti $749 5070 $549 (Availability Starting Jan 30 for RTX 5090 and 5080)

Gaiff

SBI’s Resident Gaslighter
4080 Super:
CUDA Cores: 10,240
RAM: 16GB GDDR6X
Memory Width: 256-bit
Clock: 2.3 GHZ
TFLOP: 52.2

5080:
CUDA Cores: 10,740
RAM: 16GB GDDR7
Memory Width: 256-bit
Clock: 2.3ghz
TFLOP: 56.3

Where is this 40% coming from, exactly?
From the new architecture, improved ray tracing, and massive bandwidth increase thanks to GDDR7.

I also love how you ignored the lower end of the range.
 
Last edited:

Bojji

Member
4080 Super:
CUDA Cores: 10,240
RAM: 16GB GDDR6X
Memory Width: 256-bit
Clock: 2.3 GHZ
TFLOP: 52.2

5080:
CUDA Cores: 10,740
RAM: 16GB GDDR7
Memory Width: 256-bit
Clock: 2.3ghz
TFLOP: 56.3

Where is this 40% coming from, exactly?

Yeah I don't see it too.

5080 has few more TF than 4080S
5070ti has THE SAME amount of TF as 4070tiS
5070 has LESS TF than 4070S.

I don't see big performance uplifts in raster games based on this.

They even lowered clocks in some gpus vs 4xxx (to reduce power draw?). Only big IPC uplift would make those GPUs perform much better (maybe it's there...).

From the new architecture, improved ray tracing, and massive bandwidth increase thanks to GDDR7.

RT improvement won't help raster/UE5 games. Memory BW is needed in some games/settings but not much in others.

IPC could be there but without reviews we don't know that.
 
Last edited:

scydrex

Member
Yeah I don't see it too.

5080 has few more TF than 4080S
5070ti has THE SAME amount of TF as 4070tiS
5070 has LESS TF than 4070S.

I don't see big performance uplifts in raster games based on this.

They even lowered clocks in some gpus vs 4xxx (to reduce power draw?). Only big IPC uplift would make those GPUs perform much better (maybe it's there...).



RT improvement won't help raster/UE5 games. Memory BW is needed in some games/settings but not much in others.

IPC could be there but without reviews we don't know that.
Good points. We have to wait for reviews.
 

Gaiff

SBI’s Resident Gaslighter
RT improvement won't help raster/UE5 games. Memory BW is needed in some games/settings but not much in others.

IPC could be there but without reviews we don't know that.
The Far Cry 6 benchmark has it outperforming the regular 4080 by over 30%. This is without DLSS. Those charts if put in their proper context have been generally reliable.

The 4080S is 5% faster than the 4080. Clearly, you can’t just look at TFLOPs and derive your numbers from there.
 

Bojji

Member
The Far Cry 6 benchmark has it outperforming the regular 4080 by over 30%. This is without DLSS. Those charts if put in their proper context have been generally reliable.

The 4080S is 5% faster than the 4080. Clearly, you can’t just look at TFLOPs and derive your numbers from there.

Teraflops work great to some performance point - even between gens: 4070 and 3080 have very close number of TF and perform like that on settings that don't hammer 4070 memory BW. For bigger ampere cards vs. Ada they started to fuck up.

You see the same problem with 4090 and 5090 - it has almost the same performance uplifts in FC6 and PT as other cards but unlike them 5090 is better in all aspects and has 33% more compute units (and 2x memory BW). While others have weak upgrades and show the same performance gains?
 

Gaiff

SBI’s Resident Gaslighter
Teraflops work great to some performance point - even between gens: 4070 and 3080 have very close number of TF and perform like that on settings that don't hammer 4070 memory BW. For bigger ampere cards vs. Ada they started to fuck up.

You see the same problem with 4090 and 5090 - it has almost the same performance uplifts in FC6 and PT as other cards but unlike them 5090 is better in all aspects and has 33% more compute units (and 2x memory BW). While others have weak upgrades and show the same performance gains?
Now compare the 4080 with the 4090 which has 60% more compute performance but is only 30% faster.

Again, those charts are generally reliable. Far Cry 6 on the 4080 and 4080S performs the exact same or within 3%. Either that benchmark is a lie, or there are improvements that we aren’t of that allows Blackwell to increase performance without a massive increase in compute.

A: Far Cry 6 benchmark is fake

B: Blackwell performs inexplicably well in that game
 
Last edited:

KyoZz

Tag, you're it.
EDIT: This is not a bait or attack post. Please do not school me on this. I am not baiting or attacking. Take it easy guys and have fun with new graphics cards. I am considering 5080 myself since I love 30fps
Wink Whatever GIF by Crossroads Church


Even with FGx4 we are still getting lower ms than on consoles so WTF are you talking about? In fact it's even lower than native but as always you are here to gaslight pc players.
 

Bojji

Member
Now compare the 4080 with the 4090 which has 60% more compute performance but is only 30% faster.

Again, those charts are generally reliable. Far Cry 6 on the 4080 and 4080S performs the exact same or within 3%. Either that benchmark is a lie, or there are improvements that we aren’t of that allows Blackwell to increase performance without a massive increase in compute.

A: Far Cry 6 benchmark is fake

B: Blackwell performs inexplicably well in that game

They used those games for some reasons only known to them.

They also omitted super cards to make jumps look better. I don't know, something doesn't look right to me... I expected clocks to gave cards big uplifts but clocks are lower in some cases.

I guess we have to wait for reviews, without them we are on Nvidia PR mercy.
 

Gaiff

SBI’s Resident Gaslighter
They used those games for some reasons only known to them.
They couldn’t have used 10 million games because it’d have effectively been a review. If they had used other games, you would have said the same thing.
They also omitted super cards to make jumps look better. I don't know, something doesn't look right to me... I expected clocks to gave cards big uplifts but clocks are lower in some cases.
They didn’t omit them. This is a 5070, 5070 Ti, 5080, and 5090. They replace the 4070, 4070 Ti, 4080, and 4090. They will almost inevitably have a 5070S and especially a 5080 Ti or Super since there is $1000 separating the 80 from the 90.
I guess we have to wait for reviews, without them we are on Nvidia PR mercy.
That’s why I only used FC6 and Plague Tale and ignored that MFG wank. Go look back at their launches, the straight comparisons have always materialized. NVIDIA won’t lie, but they will mislead. It’s up to us to parse through the bullshit and uncover the truth.
 
Last edited:

Bojji

Member
They couldn’t have used 10 million games because it’d have effectively been a review. If they had used other games, you would have said the same thing.

They didn’t omit them. This is a 5070, 5070 Ti, 5080, and 5090. They replace the 4070, 4070 Ti, 4080, and 4090. They will almost inevitably have a 5070S and especially a 5080 Ti or Super since there is $1000 separating the 80 from the 90.

That’s why I only used FC6 and Plague Tale and ignored that MFG wank. Go look back at their launches, the straight comparisons have always materialized. NVIDIA won’t lie, but they will mislead. It’s up to us to parse through the bullshit and uncover the truth.

No pure raster game on this chart is disappointing. RT in FC6 is shit but it's still RT and for sure 5xxx will have better RT performance. Some people suggested CPU limit in FC6 but with that 4090 and 5090 would perform the same...

Super cards replaced regular cards, so they should compare against them because outside of 4070 they discontinued all other cards in last XX months.

Yeah, Nvidia usually don't lie in charts (at least in recent years) but they cherry pick for sure.
 
Last edited:

Xyphie

Member
Unless there's a radical improvement to maximum boost frequencies to like 3.5GHz I just don't see large per SM improvements happening.

Just looking at the die sizes kinda tells the story IMO:

4080 Super - AD103 - 379 mm² - 5nm
5080 - GB203 - 377 mm² - 4nm

TSMC density improvement is 4%, so negligible. Can validate those numbers by doing 5090's 92 billion transistors / 744mm² = 123.65M / mm², which is about the same as the Ada dies. Both are the full bin, so no difference there.

Would nVidia be able to find radical perf improvements in the same transistor budget, without non-traditional improvements like better ML/RT? I think you could see something like that from Intel with their very PPA-poor B580 and worse software.

I think we'll see like like +10% from the 80 -> 84SM move, like 100-200 MHz higher GPU clock from the higher TBP and node improvements and some outliers in which the extra bandwidth from GDDR7 plays a role.
 
Last edited:
4080 Super:
CUDA Cores: 10,240
RAM: 16GB GDDR6X
Memory Width: 256-bit
Clock: 2.3 GHZ
TFLOP: 52.2

5080:
CUDA Cores: 10,740
RAM: 16GB GDDR7
Memory Width: 256-bit
Clock: 2.3ghz
TFLOP: 56.3

Where is this 40% coming from, exactly?
I wonder if my OC'ed 4080S (59.99TF, 820GB/s memory bandwidth) will be able to somehow close the gap.
 

Rea

Member
5090 price is damn ridiculous for merely a few more frames without MFG. Thanks Jensen. Seems like my 4090 can last until 6090. 🙋😇
 

The Cockatrice

I'm retarded?
RT improvement won't help raster/UE5 games

I think this needs to be said more often. The majority of games are and will be using UE5 which uses Lumen so all those RT improvements to the gpus equals to nothing outside of 1 or 2 big games coming once a year.

I dont see how MFG from a technical pov is exclusive to only 5x cards and can't be used in 4x ones. This new lineup feels like a fucking scam.
 

Mithos

Member
4090 vs 5070, king dethroned! Also confirms wukong is 29fps maxed out using 5090 🤡


I keep seeing videos with Nvidia Frameview having a grey/black background on the statistics, but I see no such options in my Frameview App, anyone know how to turn that on?
 
Last edited:

HeisenbergFX4

Gold Member
I don’t get this . 4090 is 25 26 frames for wukong. Does that mean there is barely no resta performance gain for the 5090 over 4090 ?
Its how it looks that Wukong picked up single digit frames over the 4090 and DLSS 4 is coming to all RTX cards

Obviously we have to wait for real benchmarks from trusted sites but if my 4090 had not died I would be waiting for much more details before jumping on the 5090, maybe :)
 
I don’t get this . 4090 is 25 26 frames for wukong. Does that mean there is barely no resta performance gain for the 5090 over 4090 ?

Yeah there is something seriously wrong with Blackwell. Like given the extra cores + memory bandwidth + power draw, you would expect 33% faster at a bare minimum. Probally won't get an answer as to where the regression is until reviews tho.
 

//DEVIL//

Member
Its how it looks that Wukong picked up single digit frames over the 4090 and DLSS 4 is coming to all RTX cards

Obviously we have to wait for real benchmarks from trusted sites but if my 4090 had not died I would be waiting for much more details before jumping on the 5090, maybe :)
I don’t care about multi frame gen . This means nothing to me 3x or 4x. Loseless scaling does that and u don’t need an rtx even. Might be not as good but still. I am only interested in actual performance . If this is not going to have a resta power I’ll just buy a used 4090.
 

SmokedMeat

Gamer™
I’m going to wait it out and see some benchmarks from AMD as well. I’m not expecting them to match Nvidia on fancy new tech, but if they’re making great strides in FSR4 (looks to be the case), and RT is improved, I’ll absolutely give them a look.
Sounds like they’re going to be aggressive on pricing this time, but who knows.

The bottom line is everyone in the mid range should be rooting for them, because we need an alternative to Jensen.
 

dcx4610

Member
How are we feeling about the 5070 Ti compared to the 5070? Is 4GB extra RAM really worth $200? I'm just wondering if it will even be noticeable. I'm playing 1440p...
 

Celcius

°Temp. member
How are we feeling about the 5070 Ti compared to the 5070? Is 4GB extra RAM really worth $200? I'm just wondering if it will even be noticeable. I'm playing 1440p...
How many years do you plan to keep this card?
I wouldn't buy a card with less than 16gb in 2025 personally
 
Last edited:

dcx4610

Member
How many years do you plan to keep this card?
I wouldn't buy a card with less than 16gb in 2025 personally
Yeah... That's what I'm thinking too. I currently have a 2070 and don't buy GPU upgrades much. I might as well future proof just a little bit.

I guess the question now though is 5070 Ti or 5080? I play on 1440p so I'm debating if 5080 is overkill or if it's just future proofing.
 
Last edited:
4080 Super:
CUDA Cores: 10,240
RAM: 16GB GDDR6X
Memory Width: 256-bit
Clock: 2.3 GHZ
TFLOP: 52.2

5080:
CUDA Cores: 10,740
RAM: 16GB GDDR7
Memory Width: 256-bit
Clock: 2.3ghz
TFLOP: 56.3

Where is this 40% coming from, exactly?
The more you look at the 5000 series line up, the more suspect it looks. The more you look at the games chosen for benchmarks along with the settings, the more suspect it looks.... This isn't an intel where they show like the performance of 20 games or something. The math ain't adding up. Personally, I don't see how 40% on average is going to manifest. It doesn't seem plausible from the specs.
 
Last edited:
From the new architecture, improved ray tracing, and massive bandwidth increase thanks to GDDR7.

I also love how you ignored the lower end of the range.
No new architecture in recent memory has delivered IPC gains in the 40% region while staying on a similar node. Not all games are bandwidth sensitive either. I don't think the average RT and Raster increase will be 40% or anything close to that. There may be outliers but yea. Maybe Nvidia are just hiding the boost clocks and the cards actually boost to like 3ghz or something.
 
The more you look at the 5000 series line up, the more suspect it looks. The more you look at the games chosen for benchmarks along with the settings, the more suspect it looks.... This isn't an intel where they show like the performance of 20 games or something. The math ain't adding up. Personally, I don't see how 40% on average is going to manifest. It doesn't seem plausible from the specs.

Not to mention that you should expect the gap between the 4090 and 5090 to be way more than the other models.

I'd say the odds the 5070 is slower than the 4070 Non Super in raster games is pretty high.
 

Gaiff

SBI’s Resident Gaslighter
No new architecture in recent memory has delivered IPC gains in the 40% region while staying on a similar node. Not all games are bandwidth sensitive either. I don't think the average RT and Raster increase will be 40% or anything close to that. There may be outliers but yea. Maybe Nvidia are just hiding the boost clocks and the cards actually boost to like 3ghz or something.

I said 20-40%. The FC6 benchmark is 31%. I suspect it will be much closer to 20% most of the time than 40%. 40% is for extreme outliers where there are enormous performance differences depending on the architecture. Think of Alan Wake 2 in which the 3070 is around 15% faster than the 2080 Ti when they are usually equal.

You guys are using 40% as if I claimed this would be a regular occurence. It won’t be more than 40%. That’s the absolute max. Generally will be around 20%.
 
Last edited:
I said 20-40%. The FC6 benchmark is 31%. I suspect it will be much closer to 20% most of the time than 40%. 40% is for extreme outliers where there are enormous performance differences depending on the architecture. Think of Alan Wake 2 in which the 3070 is around 15% faster than the 2080 Ti when they are usually equal.

You guys are using 40% as if I claimed this would be a regular occurence. It won’t be more than 40%. That’s the absolute max. Generally will be around 20%.
If the average is around 20%(I suspect it'll be less when compared to the super), the 5000 generation is completely skippable.
 

Gaiff

SBI’s Resident Gaslighter
If the average is around 20%(I suspect it'll be less when compared to the super), the 5000 generation is completely skippable.
Judging by the specs sheets alone, it will be forgettable. Those results for FC6 are insanely impressive given that the hardware of the 5080 isn’t a lot more impressive on paper than its predecessor. Bandwidth aside, it’s paltry.
 
Last edited:

tylrdiablos

Member
Should be after 2 years going by usual Nvidia cadence

I expect first there will be a 5080 Super next year, then a 5080 Ti in 2027, then a 5090 Ti late in the generation not long before the 60 series debuts
Really? Seems like a long wait between models.
The 3080 was released in September 2020 and the 3080 Ti released in June 2021. So 9 months.
The 4080 was release in November 2022 and the 4080 Super (no Ti version) released in January 2024. So 14 months.

Maybe you're right. Ahh well. Back to waiting and seeing what AMD and Intel do I suppose.
 

V1LÆM

Gold Member
I'd wait for a 5080 Ti with 20-24GB RAM but I thought we'd get a 4080 Ti and 4090 still remains the best GPU available to this day and for the rest of this month. If we do get a 5080 Ti it's gonna be 2026, more expensive at ~£1,200-1,300 and at most it'll have 24GB RAM. It's not gonna beat the 5090.

£1,940 is a lot but at least I won't have to wait around to see what happens. I can buy a 5090 knowing it'll be the best GPU for the next 2 years (That is unless Nvidia put out a 5090 Ti) and will obviously last much longer than that. I won't need to upgrade again until at least the 8090.
 
Last edited:

Gaiff

SBI’s Resident Gaslighter
I'd wait for a 5080 Ti with 20-24GB RAM but I thought we'd get a 4080 Ti and 4090 still remains the best GPU available to this day and for the rest of this month. If we do get a 5080 Ti it's gonna be 2026, more expensive at ~£1,200-1,300 and at most it'll have 24GB RAM. It's not gonna beat the 5090.

£1,940 is a lot but at least I won't have to wait around to see what happens. I can buy a 5090 knowing it'll be the best GPU for the next 2 years (That is unless Nvidia put out a 5090 Ti) and will obviously last much longer than that. I won't need to upgrade again until at least the 8090.
There’s $1000 separating the 5080 from the 5090. NVIDIA could easily slot 3 or even 4 cards in there.

5080S: $1300

5080 Ti: $1600
 

FireFly

Member
I think this needs to be said more often. The majority of games are and will be using UE5 which uses Lumen so all those RT improvements to the gpus equals to nothing outside of 1 or 2 big games coming once a year.

I dont see how MFG from a technical pov is exclusive to only 5x cards and can't be used in 4x ones. This new lineup feels like a fucking scam.
RT improvements will help Hardware Lumen which is being used as the base for Epic's MegaLights technology.

Edit: Ah, I see it was already mentioned above. However the key point is that Hardware Lumen was previously too expensive to use on consoles, but with UE 5.5 Epic are aiming to make 60 FPS with Hardware RT achievable. So if it becomes the norm on consoles, that will filter down to PC.
 
Last edited:

Kenpachii

Member
I'd wait for a 5080 Ti with 20-24GB RAM but I thought we'd get a 4080 Ti and 4090 still remains the best GPU available to this day and for the rest of this month. If we do get a 5080 Ti it's gonna be 2026, more expensive at ~£1,200-1,300 and at most it'll have 24GB RAM. It's not gonna beat the 5090.

£1,940 is a lot but at least I won't have to wait around to see what happens. I can buy a 5090 knowing it'll be the best GPU for the next 2 years (That is unless Nvidia put out a 5090 Ti) and will obviously last much longer than that. I won't need to upgrade again until at least the 8090.

Waiting game is kinda iffy in my vision with all the tech innovations going at lightning speed these days, u don't want to buy a GPU to close to a new generation anymore because u fall behind faster. And that's what happens when u wait for the TI versions which makes them less value in my view.

Sitting on a GPU for multiple generations is also what is not a good idea anymore these days. Imagine buying a 3080 ti, a 4080 absolute levels it with framegen a year later. With path tracing pushed forwards in that gen, u need framegen to get stable framerates. a 4080 vs a 3080 u see a massive performance difference because of it and sadly path tracing is off the menu. While if u bought a 3080 at launch, u got a full 2 years with games launching around that tech before a change happens towards newer tech. Imagine having a 3080 and path tracing releases and u can't because u lack a feature, imagine when 5080 launches and sit on a 3080 when the next title makes full use of 4x framegen and pushes visuals forwards, that 3080 will sit at single digits trying to run it.

This is also why i am convinced that sitting for multiple gens with higher end GPU's is a dumb thing to do. A 5080, with 4x framegen absolute levels a 4080 in newer titles, games that are going to take advantage of that with higher settings will come out in the next 2 years as devs are going to use it. so buying a 5080 at launch u get 2 full years of having a gpu that games are made for is a good idea, waiting for a 5080 ti, that's 1 year u lose out of it.

I think the best thing to do is, before nvidia announces there next gpu's, is to sell your current high end gpu. sell that 4080 for 900 bucks ( they go for 900-950 second handed in my country ), add ~300 bucks and have a 5080 and u are ready to go for this gen. Do the same with the next generation when it arrives. 4090's lose there value harder here it seems like u can pick them up now second handed so u return is far worse, but even there u can do it with, just sell it earlier then the announcement and u should be good.
 
Last edited:
Top Bottom