NVIDIA 5000 SERIES

[Sedgey123]

So from everything i have read and watched it just seems more and more like the price tag of all the new series cards is pretty much for a bit of software that is effectively over inflating there numbers due to Multi Frame generation as the rasterisation perfomance uplift on each card does look to be only about 25-30% ish from there predeccessor.

That is if i am understanding frame generation correctly of course.

 

[mrk]

The 30 series suffered from them, not the 40 series.

I was running my 13900K and 4090 on a 750w PSU for over 8 months with no issues at all.
Only swapped the 750 for an 850 because I was building my daughter a rig and thought it best I have the more powerful PSU of the two and the Asus Loki matches my white build

The 40 series reviews showed they brought transient spikes under control but it seems normal around here for people to recommend PSU's far exceeding what people actually need

Yes 40 series had no transient issues. My 3080 ti FE needed a new PSU as the transients knocked out my then SuperFlower but only in one single game, Cyberpunk. It was a hot topic in this forum at the time as I spent ages reseraching and diagnosing before discovering the issue.

I have no trouble seeing a decent 1000W PSU doing just fine with a 5090 really.

 

[AStaley]

Well I'm staying with the 4090 so there is that

Whaat you were getting the 5090 last night lol. @TNA

Don’t worry just yet, there’s a few weeks left for him to realise his mistake and buy the 5090.

 

[Kirby Wurm]

It's obviously too early to say yet before reviews, but the generational improvement for the 5080 appears to be smaller than the 3080 to 4080 based on A Plague Tale; that was about 50%, whereas this looks more like ~30% with DLSS4 to help justify the price (and the lack of competition).

I'm frustrated that itr doesn't have 24GB VRAM. The mobile 5090 uses 3GB modules to achieve it, so as has been speculated already it's surely because Nvidia are holding back for a 5080 TI/Super to plug the gap between the 80 and 90 (though if the 5080 is supposedly using the full GB203 chip then would that only leave cut down GB202 chips?). I'd like to get one but I worry whether it will be enough to last more than a couple of years at 3440x1440p/4K. On the other hand maybe it will run out of grunt first. It's not a 3070 situation by any means, but still.

 

[Sedgey123]

It's obviously too early to say yet before reviews, but the generational improvement for the 5080 appears to be smaller than the 3080 to 4080 based on A Plague Tale; that was about 50%, whereas this looks more like ~30% with DLSS4 to help justify the price (and the lack of competition).

I'm frustrated that itr doesn't have 24GB VRAM. The mobile 5090 uses 3GB modules to achieve it, so as has been speculated already it's surely because Nvidia are holding back for a 5080 TI/Super to plug the gap between the 80 and 90 (though if the 5080 is supposedly using the full GB203 chip then would that only leave cut down GB202 chips?). I'd like to get one but I worry whether it will be enough to last more than a couple of years at 3440x1440p/4K. On the other hand maybe it will run out of grunt first. It's not a 3070 situation by any means, but still.

In your situation if you are running a 6800XT i would be doing as you said and waiting for reviews/benchmarks.. if you are not fussed about 2nd hand a 4090 might be more suitable if you dont care about the multi frame generation as it sounds like it will be on par with the 5080 on rasterisation or there abouts so that would give you the extra VRAM you wouldn't have to worry about.. or again wait for the 5080 Ti/Super as yes there is a massive hole there for Nvidia to put something in, but that is more than likely a year or so away at a guess.

 

[Kirby Wurm]

In your situation if you are running a 6800XT i would be doing as you said and waiting for reviews/benchmarks.. if you are not fussed about 2nd hand a 4090 might be more suitable if you dont care about the multi frame generation as it sounds like it will be on par with the 5080 on rasterisation or there abouts so that would give you the extra VRAM you wouldn't have to worry about.. or again wait for the 5080 Ti/Super as yes there is a massive hole there for Nvidia to put something in, but that is more than likely a year or so away at a guess.

Yeah, that's sensible. I don't think the 4090 will devalue much though, the extra VRAM justifies a higher price to some degree. And waiting just leads to that cycle of when to upgrade, as by the time they arrive the next full generation will be a year away. Decisions! I look forward to the 30th.

 

[ICDP]

I mean I know not many bought that sku but hard to talk of the topic. @ICDP have you monitored that on yours at all?

No but I don’t play modded games. I do tend to use upscaling if possible but that’s more to keep things cool and quiet. 16GB VRAM has not been an issue at 4K for me and I suspect it will be fine for a few years yet.

 

[sx_turbo]

6800XT is beasting it - abit faster than 3090 in several titles these days - fine wine drivers, must admit I've never got so much value from a card than a 6800XT 16gb - it eventually beat out Nvidia's old flagship.

Are you sure about that?

I've a 6800xt, very happy with it but can't believe it's magically better than a 3090, unless Nvidia are doing an apple and hamstringing the drivers on older GPUs to force owners to upgrade

 

[Slam]

Will all 5080 cards use the Nvidia power connection or can partners use multiple traditional pci-e cables?

 

[Grim5]

It's obviously too early to say yet before reviews, but the generational improvement for the 5080 appears to be smaller than the 3080 to 4080 based on A Plague Tale; that was about 50%, whereas this looks more like ~30% with DLSS4 to help justify the price (and the lack of competition).

Rtx3080: 29 Teraflops

Rtx4080: 49 Teraflops (68% improvement)

Rtx5080: 56 Teraflops (15% improvement)


Don't need to be a rocket scientist to work out the performance improvement from 4080 to 5080 is much, much smaller than 3080 to 4080


Right now the waters are muddied due to Nvidias marketing BS, but come review day, I'm willing to bet there is gonna be a lot of sad face unhappy review videos

 

[muon]

It'll also be obvious from the transistor count.

GP102 was 12bn transistors when fully unlocked as a Titan Xp
GA102 was 28.3bn transistors when fully unlocked as a 3090 Ti
AD103 was 45.9bn transistors when full unlocked as a 4080 Super
GB203 is ???

Whilst we don't know what GB203 is, GB202 (5090) is 92bn transistors and 744mm^2. GB203 is 377mm^2 so should be around 46bn transistors.

This is because of the process used.

Pascal was TSMC 16nm
Turing was TSMC 12nm
Ampere was Samsung 8nm
Ada was TSMC 4nm
Blackwell is still TSMC 4nm

The 3nm process isn't quite ready.

3 nm process - Wikipedia

en.wikipedia.org

 

[mrk]

 

[Mike25]

Will all 5080 cards use the Nvidia power connection or can partners use multiple traditional pci-e cables?

Looking at the partner cards that have pictures on the OCUK store, they all use the 16 pin connector. At least some of them seem to come with an adapter, but might be best checking the manufacturers website.

 

[Grim5]

Will all 5080 cards use the Nvidia power connection or can partners use multiple traditional pci-e cables?

All the FE cards for all models and partner cards I've seen are using a single 16 pin PCIE12 connector, Nvidia is really pushing this connector hard. If you want to use PCIE 8 pin, it will be an AMD card.
As above double check, but I believe all cards will come with an adaptor because about 1% of PCs have a PSU with a PCIE12 power connector in them

 

[HEADRAT]

I’ll be interested to see proper benchmarks of AAA games using

1 Rasterisation
2 Ray Tracing
3 DLS4

I get a feeling NVIDIA is pushing 3 hard as performance uplift of 1 and 2 is poor compared to the hefty price tag of 5090, time will tell

 

[MCFC_ANDY]

I think everyone is looking forward to 3rd party benches. Plus dls 4 as looks too good to be true.
I only grabbed a quick look but did i see like just 27fps raster RT in cyberpunk, what card was that?

 

[brostradamus]

It'll also be obvious from the transistor count.

GP102 was 12bn transistors when fully unlocked as a Titan Xp
GA102 was 28.3bn transistors when fully unlocked as a 3090 Ti
AD103 was 45.9bn transistors when full unlocked as a 4080 Super
GB203 is ???

Whilst we don't know what GB203 is, GB202 (5090) is 92bn transistors and 744mm^2. GB203 is 377mm^2 so should be around 46bn transistors.

This is because of the process used.

Pascal was TSMC 16nm
Turing was TSMC 12nm
Ampere was Samsung 8nm
Ada was TSMC 4nm
Blackwell is still TSMC 4nm

The 3nm process isn't quite ready.

3 nm process - Wikipedia

en.wikipedia.org

i dont get why they regressed on transistor density, i was dreaming of 110-120 billion transistors on 5090 and was very confident about it crossing 100 billion as a very very very conservative estimate
..still a 30% perf improvement is not bad

 

[Varys]

Looks very good indeed

 

[ne0]

I still can't get over seeing the 5090 can only manage 28fps Wukong at native 4k. That's just insane.

The whole refresh seems entirely based on frame gen which just seems completely off. Frame gen should be a way to boost a game where you need more frames not the default way to play everything and the main selling point, surely?

I'm a 3090 user so still makes sense for me to upgrade this time but still it's disappointing and just not as exciting with so much emphasis on frame gen providing the perception of more performance instead of actual extra grunt.

 

[Raiden85]

Really is insane spending £2k+ only to reach 28fps natively . And the game to me doesn't even look that good.