NVIDIA GeForce RTX 5080 Founders Edition

[mb194dc]

The sad thing is, we won't even know how bad the situation is.

Even if the cards reach the stores at +30 to +100% pre-scalped levels and just sit on the shelves, and Mindfactory sales show they aren't moving even at the Lovelace levels (which were at times outsold by AMD cards), Nvidia can just show the quarterly revenue as record high!

DeepSeek fiasco won't put a dent in Nvidia's revenue, as they have allegedly sold all the AI acceleration hardware they can build this year. And as for it being illegal to doctor the revenue numbers? Either dismissed, or laughably cheap to breach - and since Nvidia changed the sector name to "Gaming and AI PC", they can legally include all kinds of non-Gaming related stuff.

That's what you get when you expect a nice toy from a company that doesn't want to be associated with toy making.

It highly probably will, corporations won't just donate money to Nvidia. Unless there's a profitable reason to buy LLM gpus. Within a few quarters their revenue likely drops 90% or even more.

We'll see I guess.

 

[umeng2002]

Come on guys. Stop being poor, If you don't like the 5080, just get a 5090. It's a better value anyways. Be thankful nVidia™ is even making gaming GPUs.

 

[Legacy-ZA]

@W1zzard

I know you are stilly busy with board-partner GPU reviews, but... when will we see them? Slightly before sales go live or as they go live?

Come on guys. Stop being poor, If you don't like the 5080, just get a 5090. It's a better value anyways. Be thankful nVidia™ is even making gaming GPUs.

 

[Bwaze]

It highly probably will, corporations won't just donate money to Nvidia. Unless there's a profitable reason to buy LLM gpus. Within a few quarters their revenue likely drops 90% or even more.

We'll see I guess.

DeepSeek also needs LLM acceleration, it's not running on potatoes. And due to sudden panic that they left people's data and inputs in plaintext, there is a big drive to limit it's use in the West, Italy has already issued a ban from app stores, other countries are following.

And DeepSeek isn't perfect - I bet the reasoning goes - if it had more hardware capability, it would work even better, don't cancel your expensive AI hardware orders! Less is more? No, more is more!

 

[Frizz]

These cards will be for the losers.... who didn't secure a 5090 at launch

 

[3valatzy]

Come on guys. Stop being poor, If you don't like the 5080, just get a 5090. It's a better value anyways. Be thankful nVidia™ is even making gaming GPUs.

The more you don't buy, the more you save - JHH.

 

[remekra]

I would love to see Zotac 5080 Solid review, seems to me that it's the only AIB card that should be MSRP but still looks good and should have good cooling perfomance. Others are marketed as SFF or look bad.

 

[Legacy-ZA]

I would love to see Zotac 5080 Solid review, seems to me that it's the only AIB card that should be MSRP but still looks good and should have good cooling perfomance. Others are marketed as SFF or look bad.

TTL did a review, but he didn't tear the ZOTAC card down so that we could have had a closer look at the components and how they are cooled.

GALAX & ZOTAC so far, have MSRP cards, but I want to see what they look like under the hood. ZOTAC being at the front of my mind, with the juicy 3 year warranty + 2 years if you register, well, assuming they honour their warranty claims if something goes bad.

 

[_roman_]

That 16GiB VRAM seems to be another planned obsolescence.

1200€ according to hardwareluxx.de for 16GiB. And hardly anyone ever talks about that so far.

This card looks like a 3070, regardless of it'S labeling for 5080. I'm sure it will repeat again. 3070 got fast 8GiB issues, this 5080 will get fast 16GiB issues. The faster PCIE bus will fix it a bit. In my point of view it's very bad design practise to hide low VRAM with the fast bus transfer to move needed data faster to the graphic card memory. That graphic card memory in the first place is there so the pcie bus has less traffic and is not so much abused for the important tasks in the system.

Just watching until that topic will pop up. Maybe January 2027? - I need a replacement for the 5080 ...

More Watts used before the power supply with a bad efficiency for the hole system equals more frames.
People should be aware of the efficiency factor in the output curve from their power supply unit. Hint. Multiply that graphic card "shown wattage" with a correction factor of 1.1 - i would be realistic and use a multiplication factor of 1.2. Someone in your households pays the watt at the wall socket - not those fake Watts shown in the gpu driver overlay.

These cards would look much better with less wattage for the same frames. There should be the wattage number in [Number Wattage graphic card / number wattage total system at the wall socket] next to the "relative performance" chart

 

[Legacy-ZA]

That 16GiB VRAM seems to be another planned obsolescence.

1200€ according to hardwareluxx.de for 16GiB. And hardly anyone ever talks about that so far.

This card looks like a 3070, regardless of it'S labeling for 5080. I'm sure it will repeat again. 3070 got fast 8GiB issues, this 5080 will get fast 16GiB issues. The faster PCIE bus will fix it a bit. In my point of view it's very bad design practise to hide low VRAM with the fast bus transfer to move needed data faster to the graphic card memory. That graphic card memory in the first place is there so the pcie bus has less traffic and is not so much abused for the important tasks in the system.

Just watching until that topic will pop up. Maybe January 2027? - I need a replacement for the 5080 ...

More Watts used before the power supply with a bad efficiency for the hole system equals more frames.
People should be aware of the efficiency factor in the output curve from their power supply unit. Hint. Multiply that graphic card "shown wattage" with a correction factor of 1.1 - i would be realistic and use a multiplication factor of 1.2. Someone in your households pays the watt at the wall socket - not those fake Watts shown in the gpu driver overlay.

It really should have had 24GB, but that is probably reserved for the RTX5080Ti/Super 1 year from now. If Indiana Jones is used as an indicator, yes, it runs out of VRAM with everything maxed.

 

[Broken Processor]

Don't start with "AMD did this" nonsense. AMD is not responsible for the actions of Nvidia and is not obligated to make high end cards so you can buy Nvidia cheaper.

I didn't start I literally said it and the proof is in the 5000 lineup and the result is an entire generation from Nvidia not worth buying for gaming. They have AI customers to scalp. We need competition or we end up with this and don't forget Intel's lost decade but at least Nvidia is innovating.

 

[Frizz]

the 4090 cost 4000 aud here in aussieland, prices are out for the 5080 and its between 2100-2400aud, its still not a bad deal at all.

 

[remekra]

TTL did a review, but he didn't tear the ZOTAC card down so that we could have had a closer look at the components and how they are cooled.

GALAX & ZOTAC so far, have MSRP cards, but I want to see what they look like under the hood. ZOTAC being at the front of my mind, with the juicy 3 year warranty + 2 years if you register, well, assuming they honour their warranty claims if something goes bad.

Yeah GALAX/KFA should also be MSRP but the looks of that card are just throwing me away and since I have my PC right next to my face it is a factor
Gigabyte and Asus also at least in theory should have MSRP cards with Windforce and Prime and MSI with Ventus but like I said they have pretty small coolers compared to Zotac and since I have plenty of room in my case I would rather have bigger card and lower noise and temps.
Or I will be in luck again and will be able to buy FE as with 3080.

 

[Bwaze]

I believe Nvidia's monopoly in gaming has nothing to do with the situation we're in - basically we got a re-release of Lovelace.

Nvidia has been in this situation before, and even if there is no competition to force them, they still have to somehow convince buyers. With such a low performance increase of course no 40x0 owner will upgrade. Owners of older cards? If the performance and cost of Lovelace cards, with lower priced mid generational refresh didn't convince them, what do 50x0 cards bring new? Especially if the stocks are really as low and we'll see scalping prices right away.

Multi frame generation? I've seen almost unanimous consent that it's not a magical solution "to make 5070 defeat 4090" - too much issues with image quality and latency for majority of uses. I think if Nvidia would bet on that, we'd see more YouTubers and other influencers sing paid accolades.

Nvidia just doesn't care this round about gamers. So people won't buy gaming cards? Big deal, it's a drop in their revenue now, and to rub it in our faces, we'll see overpriced cards lying on shelves (after the inevitable initial rush that will grab everything) - but the "Gaming and AI PC" revenue will still break all records...

 

[tfdsaf]

This was a pathetic release even by Nvidia's garbage standards. 8% faster performance at the same price 2 years later? Nvidia really must think all of its customers are complete and utter imbeciles with no brain, just empty skulls.

16GB of vram for a $1000 dollar GPU? Its 2025, its not 2017 anymore, when even low end cards we were getting 8GB of vram. How did we went from 8-12GB of vram 8+ years ago to barely having 16GB of vram on $1000+ GPU's?

 

[JustBenching]

This was a pathetic release even by Nvidia's garbage standards. 8% faster performance at the same price 2 years later?

It's not the same price, the 4080 was 1200$. The 5080 in fact offers over 40% better perf / $.

 

[Bwaze]

It's not the same price, the 4080 was 1200$. The 5080 in fact offers over 40% better perf / $.

RTX 4080 Super was basically the same card as RTX 4080, and it was released for $1000 because 4080 cards were lying on shelves for $1200. Don't act as if that didn't happen.

 

[csendesmark]

Paul did a nice summary on this GPU, definitely worth watching:

They should apply for a new name, my suggestion is:
nGreedia

 

[JustBenching]

RTX 4080 Super was basically the same card as RTX 4080, and it was released for $1000 because 4080 cards were lying on shelves for $1200. Don't act as if that didn't happen.

The post I quoted said 2 years later. The 4080S didn't launch 2 years ago but last year.

 

[Bwaze]

The post I quoted said 2 years later. The 4080S didn't launch 2 years ago but last year.

Average FPS of RTX 4080 at 4K: 84.1, at $1199 one frame costs $ 14,2568


Average FPS of RTX 5080 at 4K: 96.7, at $999 one frame costs $ 10,3309

The difference is 38.0016%, not 40.%.

So you're also not correct.

 

[mb194dc]

DeepSeek also needs LLM acceleration, it's not running on potatoes. And due to sudden panic that they left people's data and inputs in plaintext, there is a big drive to limit it's use in the West, Italy has already issued a ban from app stores, other countries are following.

And DeepSeek isn't perfect - I bet the reasoning goes - if it had more hardware capability, it would work even better, don't cancel your expensive AI hardware orders! Less is more? No, more is more!

Deepseek has shown that you can train models with a tiny portion of the hardware that already exists, even with older chips. Nvidia has likely already sold enough LLM chips to power and create all the models anyone will ever need for at least 5 years and probably even longer than that.

So whilst orders may not be cancelled (some may be), why would more orders be placed? The hardware is already in warehouses because the data centres don't even exist yet. Prices for GPU time at already existing data centres is going to collapse as well. If you look at it from the end users perspective, it won't be long before models are created that can be sold to the end user for $1 a month, compared to the $20 that the likes of ChatGPT charge, because they'll be created without using massively expensive Nvidia hardware...

There should be a massive shakeout and it'll likely be a bit of a bloodbath as more efficient models supplant what exists now. The 5080/90 generation could well end up like the 8800 Ultra, with prices for future generations being much lower than now as they'll be mountains of excess capacity.

 

[x4it3n]

Yes. For whatever is being used for the Blackwell die wafers. To be honest, I haven't read up on what they're actually using.

Blackwell is on an enhanced 4nm node, hence the poor efficiency of RTX 50s.

I'm pretty sure the original design was on a TSMC 3nm node but Nvidia either didn't want to pay $30-40K per wafer or just couldn't get enough chips due to Apple and Qualcomm getting most of the chips, even though TSMC 3nm yields are around 90% right now so yeah...

 

[remekra]

Deepseek has shown that you can train models with a tiny portion of the hardware that already exists, even with older chips. Nvidia has likely already sold enough LLM chips to power and create all the models anyone will ever need for at least 5 years and probably even longer than that.

So whilst orders may not be cancelled (some may be), why would more orders be placed? The hardware is already in warehouses because the data centres don't even exist yet. Prices for GPU time at already existing data centres is going to collapse as well.

There should be a massive shakeout and it'll likely be a bit of a bloodbath as more efficient models supplant what exists now. The 5080/90 generation could well end up like the 8800 Ultra, with prices for future generations being much lower than now as they'll be mountains of excess capacity.

Every bubble sooner or later pops. AI will still be here and will advance but market definitely needs correction. Deepseek it's only a beggining, it's still tied closely to CCP but I bet we will soon more models like it coming from the West.

 

[x4it3n]

Average FPS of RTX 4080 at 4K: 84.1, at $1199 one frame costs $ 14,2568


Average FPS of RTX 5080 at 4K: 96.7, at $999 one frame costs $ 10,3309

The difference is 38.0016%, not 40.%.

So you're also not correct.

Whoever bought a 5080 for $1200 is a total idiot lol

 

[Bwaze]

Deepseek has shown that you can train models with a tiny portion of the hardware that already exists, even with older chips. Nvidia has likely already sold enough LLM chips to power and create all the models anyone will ever need for at least 5 years and probably even longer than that.

Are you saying that just a couple of days after announcement of Stargate, a $500 billion AI initiative that it's not worth it?

I bet we'll see attacks on DeepSeek from all guns - first were about security of searches - which are even worse we expected from Cinese company, it's not only accessible to Chinese government, but to all.

Next we'll see debunking of every metric - from it's accuracy and useability to claims about hardware requirements.

It will be of national interest for DeepSeek to be put into it's place. And for businesses to run as planned