RTX 5080 - premature review - it sucks

[Outback Bronze]

@JustBenching
Remember when UT2004, Doom 3 and FEAR completely shattered the kneecaps of that day fastest GPUs on, for the time, top resolutions? I remember. The whole “fastest GPU should get 60 at least at the highest mainstream available resolution” was never a thing that happened till the last few gens (and even then). This is just a return to normalcy.

I most certainly do! Oblivion was quite chewy too!!

Crysis was the worst though. My 8800GTS 640mb
was reduced to 20-25fps fully maxed 1280x1024 (on a 55 inch DLP) and i loved every minute of it. If only i could enjoy 20fps like i was able to do back then.

 

[Carillon]

wow, it gets beaten by the 5070...

 

[Assimilator]

A new gen based on the same architecture and fab node is the definition of laziness, it's the same kind of stagnation Intel pulled for years.

Moving to TSMC 3N would have achieved very little, no good reason for NVIDIA to bother. Especially since Apple has bought up all 3N capacity.

 

[Dragam1337]

Well, I know this may sound harsh & hard to swallow, but...

HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA....

ps..... this is aimed at all those that fell for the same ole same same lame-assed marketing BS that nGreediya puts out, once AGAIN !

I'm mostly amused by that guy who kept claiming that the 5080 will be faster than 4090... lol

 

[Frizz]

I most certainly do! Oblivion was quite chewy too!!

Crysis was the worst though. My 8800GTS 640mb
was reduced to 20-25fps fully maxed 1280x1024 (on a 55 inch DLP) and i loved every minute of it. If only i could enjoy 20fps like i was able to do back then.

I miss the fact we could choose to dual/triple wield mid to low end graphics cards to get an extra 10-15 fps more lol, it looked sick as hell too. Imagine if AMD reintroduced Crossfire 2x 9070s > 5080/5090 for me for sure lmao.

 

[Fizban]

AMD could put something together from what they have now in one year.

Is this 8% improvement the lowest gen to gen improvement in history?

That's 8% over the 4080 super, that's not gen to gen. That's half a gen It's like 3090 Ti vs 4090, when 3090 vs 4090 would be gen to gen.

I miss the fact we could choose to dual/triple wield mid to low end graphics cards to get an extra 10-15 fps more lol, it looked sick as hell too. Imagine if AMD reintroduced Crossfire 2x 9070s > 5080/5090 for me for sure lmao.

View attachment 382311

SLI and Crossfire never worked well. In many games it resulted in stuttering and worse performance than 1 GPU. I'd never pick 2 9070's over a 5080.

 

[Vya Domus]

Well, I think it is more logical to presume that the basic computing units of a new generation product after 2,5 years will be improved and redesigned

GPU IPC is difficult to improve and generally not feasible because of the constraints, there is a reason a regular CPU has maybe ~100 FPUs and a GPU 10000+ FPUs.

I though that the number of the units and frequency can be used to judge performance differences of GPUs of the same generation.

That's what GPU makers would want you to believe so the focus can switch to other software features but the reality is performance per shader does not differ very much between generations.

 

[Fizban]

wow, it gets beaten by the 5070...

Uhh...what?!?

 

[TheinsanegamerN]

I miss the fact we could choose to dual/triple wield mid to low end graphics cards to get an extra 10-15 fps more lol, it looked sick as hell too. Imagine if AMD reintroduced Crossfire 2x 9070s > 5080/5090 for me for sure lmao.

View attachment 382311

So do I. It's a damn shame DX12 multi GPU was left to rot.

SLI and Crossfire never worked well. In many games it resulted in stuttering and worse performance than 1 GPU. I'd never pick 2 9070's over a 5080.

I loved my SLI rigs back in the day, and rarely had problems that couldnt be fixed by 3 minutes of tinkering. In exchange I got near GTX 580 performance for less than half the price.

DX12 multiGPU works a LOT better than either of the previous options, but game devs are too lazy to implement it outside ashes of the singularity. We have the ability to fix this problem.....

 

[freeagent]

SLi worked well for me too, most of the time..

I keep buying boards that support it, because I support it

 

[DemonicRyzen666]

That's 8% over the 4080 super, that's not gen to gen. That's half a gen It's like 3090 Ti vs 4090, when 3090 vs 4090 would be gen to gen.


SLI and Crossfire never worked well. In many games it resulted in stuttering and worse performance than 1 GPU. I'd never pick 2 9070's over a 5080.

The second comment is completely false.
The fact that even today games stutter is proof enough that it's false.
Secondly I'm fairly sure 99% of you have never even experienced using or playing mGPU games vs crossfire or S.L.I. games to compare against each other.

 

[freeagent]

It might suck in the sense that traditional raster does not see much of an uplift, but once all that AI stuff kicks in and is fully supported, I am sure that is where these cards will shine.. either that or these are the new Turing cards in the sense of RT power back then.

 

[_roman_]

It s online

NVIDIA GeForce RTX 5080 Founders Edition im Test - GeForce RTX 4080 Ti Super mit Blackwell-Genen | Seite 6 | igor´sLAB

Der heutige Artikel bietet einen umfassenden Review zur GeForce RTX 5080 Founders Edition (FE) und baut auf den Launchartikel zur RTX 5090 FE auf, den ich natürlich weiter unten auch noch verlinke.

www.igorslab.de

 

[eidairaman1]

It might suck in the sense that traditional raster does not see much of an uplift, but once all that AI stuff kicks in and is fully supported, I am sure that is where these cards will shine.. either that or these are the new Turing cards in the sense of RT power back then.

Now unplug the ethernet and disable the wifi and see how well they perform. Then for MP, go to a sandbox lan party (no connectivity to the web at all, just you, friends and a ethernet switch)

 

[Denver]

AMD could put something together from what they have now in one year.

Is this 8% improvement the lowest gen to gen improvement in history?

No, That title goes to the 4060ti, which, in some games, actually performed worse than its predecessor.
But talking about high-end series, yes, the worst ever...

Is the 9070XT OC = 5080? How pathetic that would be for Jensen... and even more pathetic that people will buy this for U$1200-1500.

 

[Zazigalka]

GPU IPC is difficult to improve and generally not feasible because of the constraints, there is a reason a regular CPU has maybe ~100 FPUs and a GPU 10000+ FPUs.

That's what GPU makers would want you to believe so the focus can switch to other software features but the reality is performance per shader does not differ very much between generations.

it seems the real big jump was the introduction of dual issue shaders on ampere, what came after were merely efficiency/clock/cache size adjustments. only tensor cores got a big upgrade going from ampere to ada, judging how hard the new ray reconstrucion model hits ampere, but not blackwell/ada.

 

[LittleBro]

Really? Did you ever have high end gpus in your life before? Can you tell me what gpu was able to max out every possible game ever? I've gone through plenty of high end gpus (gtx 770, 1080ti, 3090, 5870hd etc.), I couldn't max out games at over 60 fps ever. The 1080ti was actually struggling at 1080p on 2 year old games (watchdogs 2 for example).

Especially if you are talking about RT games, expecting a gpu to max it out at native 4k is just delusional. We are at least 15 years away from that.

Sure, I did. I had HD6970, 1080Ti for a while and also RTX 2080 Ti for a while. HD5870 was not a high end card, nor was GTX 770.
There were few games that I played which indeed were unable to achieve all time min. framerate of 60 fps at ultra on 1080p.

There were games like Crysis, Metro that were really demanding in terms of hardware resources, but they fuckin' looked good. What I'm trying to say is that today we get shitty games that not only look like shit, but they even require shitload of compute power to look that shitty. And that is insane. This is something that was not standard before. Many games are made for consoles, then just ported to PCs. Consoles are basically much less powerful than PCs. Still, they do optimize those games for consoles to achieve 60 fps framerate. Sure, graphics on PC can be much better

 

[Zazigalka]

No, That title goes to the 4060ti, which, in some games, actually performed worse than its predecessor.
But talking about high-end series, yes, the worst ever...
View attachment 382313
Is the 9070XT OC = 5080? How pathetic that would be for Jensen... and even more pathetic that people will buy this for U$1200-1500.

looks good. now let's see how much this 9070xt costs. hoping for 599, that would be awesome. If it's 699, I'd probably go for 5070Ti to get DLAA and a better RT denoiser. Frankly, if it's 699, I'll probably just skip this whole generation altogether, until I get something interesting in my price range (~600) again.

 

[AnotherReader]

Well, I think it is more logical to presume that the basic computing units of a new generation product after 2,5 years will be improved and redesigned and you will not be able to judge the performance just from numbers of these units or clocks alone.

I though that the number of the units and frequency can be used to judge performance differences of GPUs of the same generation.

GPUs aren't CPUs; they rely on increasing the number of cores. There are rather few instances when the per clock performance of an AMD Compute Unit or a Nvidia SMX has increased.

1. AMD: VLIW (HD 2900 XT --> HD 6970) to GCN (HD 7970)
2. Nvidia: Pascal to Turing
3. AMD: GCN (Radeon VII) to RDNA 1/2 (5700 XT to 6950 XT)
4. Nvidia: Turing to Ampere
5. AMD: RDNA 2 (modest increase compared to others) to RDNA 3

Maxwell isn't there, because the per clock performance of the SMX didn't increase. However, since it clocked higher than Kepler and each SMX had 90% of a Kepler SMX's performance at the same clocks, the increased number of SMXs led to dramatically higher performance.

 

[Frizz]

It might suck in the sense that traditional raster does not see much of an uplift, but once all that AI stuff kicks in and is fully supported, I am sure that is where these cards will shine.. either that or these are the new Turing cards in the sense of RT power back then.

People are still in disbelief with framegen, there are some true gamers that review these cards like Optimum that have tested and prefer 300-400 extra frames vs rasterization on the 90 on single player games. I know I would prefer 120 FPS in a game that no card in existence could reach 60 fps rasterization on at 4k-8k resolution with max eye candy despite some artefacting.

 

[Carillon]

Uhh...what?!?

because according to nvidia 5070=4090 and according to these leaked results 5080<4090.

 

[ShrimpBrime]

It might suck in the sense that traditional raster does not see much of an uplift, but once all that AI stuff kicks in and is fully supported, I am sure that is where these cards will shine.. either that or these are the new Turing cards in the sense of RT power back then.

I thought it means NV wants to SUCK your wallet dry.

 

[freeagent]

I thought it means NV wants to SUCK your wallet dry.

What company doesn't in 2025

None of these guys are our friends lol

 

[AnotherReader]

Moving to TSMC 3N would have achieved very little, no good reason for NVIDIA to bother. Especially since Apple has bought up all 3N capacity.

I think there's some capacity available for N3, given the existence of Arrow Lake and Turin dense.

 

[N/A]

N3 arrow lake is 150mm2 and relatively small. Intel is implementing tiles instead of a monolithic die for a reason, and Nvidia needs 600mm2 for 5090 and 300mm2 for 5080 as density improves by only 30%. This also means 60 series is just another refresh shrink just like 9800 GT 65 to 55nm with 25% density but slightly better. Yeah it's this bad.