NVIDIA GeForce RTX 5070 Founders Edition

[RedelZaVedno]

So 5070 = 4070 SUPER? Same price, same performance... Why wait a year to get this GPU?

 

[Vayra86]

Right. Next

 

[alwayssts]

Right. Next

Next. Right.

 

[EatingDirt]

Meh, you don't have to run everything at Ultra. I use a 4K monitor with a 6700 XT and while I'd like a faster card, it does fine with some settings tweaks. In many older games, it can run 4K with most if not all details at max. I don't want to back off of 4K because I need it for work.

The 5070 is actually best in class at 4K (that class being ~$550). Probably a result of its use of GDDR 7.

So, for at least some users who want or need 4K but don't want to pay $1000+, this is not a horrible choice if it is actually $550. It won't be of course, probably next best choice is 9070 if it can be had for $550 (dubious), and then a 7900 GRE or 7800 XT if the prices come down when these newer cards hit the market.

"Best in class" is meaningless when we're going to be getting the 9070 & 9070XT in, what, 2 days?

Still wouldn't consider this a 4k card. As someone that plays on 1440p Super Ultrawide with a 6800XT I don't consider it a 4k card either, even with 16GB or RAM. Lots of games that barely run at 60fps at medium settings, but at least I know that the VRAM isn't the issue.

12GB is 1080p & 1440p territory. You'll probably never have to worry about VRAM limits at those resolutions before a GPU upgrade, but at 4k, VRAM & textures will always be something you would need to worry about with a 5070 at 4k for some current games, and definitely some future games.

 

[calhau]

So no improvement over the 4070 super as expected, the only thing I can say is:

B580
Ram:12gb gddr6; tdp :190w; die:272mm²
→250$

5070
Ram:12gb gddr7; tdp :250w; die:263mm²
→550$

I'll let you think how much over BoM this garbage sells.

 

[Icon Charlie]

Magic 8 ball... "How's the Nvidia 5000 series video cards so far"?

Magic 8 ball: "NOT so good".

 

[RedelZaVedno]

Steve's Gamers Nexus priceless review summary of 5070: "Nvidia said f@ck you to gamers".

 

[RandallFlagg]

"Best in class" is meaningless when we're going to be getting the 9070 & 9070XT in, what, 2 days?

Still wouldn't consider this a 4k card. As someone that plays on 1440p Super Ultrawide with a 6800XT I don't consider it a 4k card either, even with 16GB or RAM. Lots of games that barely run at 60fps at medium settings, but at least I know that the VRAM isn't the issue.

12GB is 1080p & 1440p territory. You'll probably never have to worry about VRAM limits at those resolutions before a GPU upgrade, but at 4k, VRAM & textures will always be something you would need to worry about with a 5070 at 4k for some current games, and definitely some future games.

Yeah it really doesn't work like that. Most games let you select HD, 2K, or 4K textures for example. At least to my eyes, there are a lot of settings that are either completely undetectable or only marginally detectible.

And most of the time, meaning like 90%, the games I play are not brand new AAA titles.

That means they are designed to run well on 4+ year old flagship cards at 4K. At 4 years ago, a 2080 Ti was top of the line, and the 6700 XT is only 11% slower than 2080 Ti at 4K. So basically anything made in 2020 or earlier is going to run well on a 6700 XT, even at 4K.

 

[bubbleawsome]

This is the first time I'm seeing the noise-normalized cooler performance chart and I like it a lot, awesome addition.

 

[alwayssts]

So no improvement over the 4070 super as expected, the only thing I can say is:

B580
Ram:12gb gddr6; tdp :190w; die:272mm²
→250$

5070
Ram:12gb gddr7; tdp :250w; die:263mm²
→550$

I'll let you think how much over BoM this garbage sells.

Yeah, well, I do. When I explain to people 5080 is literally somewhere around 250% over BoM they don't believe me...but it's still true.

 

[vacsati]

This is not that bad for 550, pretty okay speedbump over the 4070.
Perfect 1440p card and can do many less demanding title in 4k medium.
Could have been 16gb, but doesnt really matter, since the gpu chip cannot really handle 4k high detail levels.

I guess this will be my next card, lets hope the reds dont fck up like they usually do, and the price of the 5070 will go down after a few wek.

 

[Dawora]

If FE does +13% performance from oc, then a Vanguard/Aorus can do 15-16%. This is above 4070ti. But at what cost?

Is the connector enough for +15% speedup from oc?

in games FPS boost is sometimes even bigger i was testing 5080 and i got almost +20% more FPS in some games. but even +15% is great OC

Realy looking to test couple of 5070 best models to see real OC potential.

 

[Toss]

as fast as RTX 4090 LUL

 

[10tothemin9volts]

Considering the -50 bucks MSRP over the 4070 Super, still 2 biggest no-gos for me:

• 12 GB VRAM DOA Their [HWU] into is funny, the GPU is so bad, Steve was not only standing, not standing on chair, but standing on the roof (RTX 5060 8GB VRAM super DOA: Steve is standing on the chair on the roof?)
• Only an almost insignificant 7.5% (100%/93%) power efficiency improvement over my 4070 (it's within ± 10% when using the same node, so one could say no improvement) (RTX 50 series GPUs use the same process node like its RTX 40 predecessor, that's why no improvement)

The only chance I'd still consider upgrading from my 4070 to a no-power-efficiency-"successor" GPU is if there is either:

• A 18 GB VRAM (3GB per GDDR7 chip, instead of the current 2GB), or preferably
• A 24 GB VRAM (clamshell VRAM design: using 2GB GDDR7 chips on both sides of the PCB), or even more preferably
• A 36 GB VRAM (clamshell VRAM design + using 3GB GDDR7 chips)

versions for AI LLM self-hosting or, generally, when 12GB VRAM are not enough. Hopefully in next year's Refresh. 3GB GDDR7 chips will be used in the 5090 Laptop SKU (see table: both, 5080 Laptop and 5090 Laptop, SKUs use the same 256-bit GB203-400 chip and 24GB VRAM is only possible if 3GB per GDDR7 chip are used).

Regarding the Specs table's "AI TOPS" on NV's website: It looks like NV is comparing FP4 vs FP8, but doesn't disclaim it, arguing in bad faith.

---

Interestingly, the 5070' chip size is 263 mm² and it doesn't look like NV is reusing any RTX 40 chip design (nowhere near neither AD104' 294.5mm², nor AD103' 378.6mm²) (since it's the same, more ripe, node with maybe even higher yield, they probably just removed some buffer silicon (not sure how much removing of features like in the recent news would contribute to the smaller mm²), and called it a day).

 

[BMfan80]

Ehhh? I don't think it'll draw anything close to that. If they let it use 1.2v, which I doubt, I still don't think it would draw much over 400w or so. My theory craft would be it around ~425w max.
My gut says it'll be limited either by 1.1v or essentially the same thing, but through power-limitng/2x8-pin (~375w).

Where did you get that 375w figure?
My 4090 can pull 600w, I've seen it when I tried to max the power usage.
I have a SuperFlower PSU that has a dual 8 pin to the 12x6 cable.
The Corsair adapter is also a 2x8pin cable, which mean an 8 pin plug can provide atleast 300w on it's own.

 

[TheDeeGee]

 

[dir_d]

Steve at Gamers nexus was dropping hints that the RX 7090 was going to be around the RX 7900XT performance, If so this card is DOA.

 

[alwayssts]

Where did you get that 375w figure?
My 4090 can pull 600w, I've seen it when I tried to max the power usage.
I have a SuperFlower PSU that has a dual 8 pin to the 12x6 cable.
The Corsair adapter is also a 2x8pin cable, which mean an 8 pin plug can provide atleast 300w on it's own.

9070 xt is (typically) 2x8-pin, which is rated for 375w. Well, technically something like 365 or something. That's probably the PL if not limited by voltage. I don't know which way they went.

As you've probably seen they're rated at 304(329) and 340 (?), where the question mark is probably ~365-375w.

Your 4090 has a 600w power connector.

FWIW: Nice review round-up right here. One of them is not like the others. If you guessed which one, you are correct:

NVIDIA's new RTX 5070 is getting destroyed by reviewers

NVIDIA's RTX 5070 review embargo has passed, and it seems that reviewers are finally fed up with NVIDIA's marketing. Here's what you need to know before buying.

www.windowscentral.com

 

[EatingDirt]

Yeah it really doesn't work like that. Most games let you select HD, 2K, or 4K textures for example. At least to my eyes, there are a lot of settings that are either completely undetectable or only marginally detectible.

And most of the time, meaning like 90%, the games I play are not brand new AAA titles.

That means they are designed to run well on 4+ year old flagship cards at 4K. At 4 years ago, a 2080 Ti was top of the line, and the 6700 XT is only 11% slower than 2080 Ti at 4K. So basically anything made in 2020 or earlier is going to run well on a 6700 XT, even at 4K.

Most games don't list textures as HD, 2k or 4k. They are low-medium-high-ultra, etc. I can't list a games on one hand that use resolution settings for their textures. I prefer my textures cranked because textures, if you have enough VRAM, have practically no impact to FPS.

And sure, if you play games that are 5 years old right now, you won't have any issues. If you get a 5070 right now, and have that for another 5 years, and continue to play games that are 5 years old, and only games that are 5 years old, you will maybe only have issues with some of the games I listed.

I guess we'll see tomorrow if the 5070 continues to be 'best in class' at 4k.

 

[BMfan80]

9070 xt is (typically) 2x8-pin, which is rated for 375w. Well, technically something like 365 or something. That's probably the PL if not limited by voltage. I don't know which way they went.

As you've probably seen they're rated at 304(329) and 340 (?), where the question mark is probably ~365-375w.

Your 4090 has a 600w power connector.

It's a "600w" connector on the GPU, but the PSU uses TWO 8 pin connectors to send the 600w to the card.
Again, same as the Corsair adapter cable.

"While the included adapters that come with the GPU are fine, what you plug them into is of concern. Technically, 8-pin GPU connectors support a maximum of 150 watts, for a maximum of 450 watts on paper when you use three of them. In reality, most power supplies and most 8-pin plugs can support 324 watts (27 amps at 12 volt) or even 468 watts (39 amps at 12 volts) according to power supply maker Corsair. That, of course, depends on the power supply design. There’s no guarantee of what each PSU can support through those three 8-pin GPU plugs."
This comes from a PCWorld article

 

[regs]

So no improvement over the 4070 super as expected, the only thing I can say is:

B580
Ram:12gb gddr6; tdp :190w; die:272mm²
→250$

5070
Ram:12gb gddr7; tdp :250w; die:263mm²
→550$

I'll let you think how much over BoM this garbage sells.

But it's over 2.5 times as fast as B580

 

[GodisanAtheist]

As soon as Nvidia is able to actually supply the channel this is going to sell like gangbusters.

Virtually anyone not in the know associates Nvidia with GPUs and 5 is bigger then 4 therefor 5xxx series is going to be better than their 4xxx series.

Hell, I'd bet dollars to donuts we get posts IN THIS VERY FORUM of lost and confused normies wondering why the 5070 isn't performing any faster than their 4070 S/Ti after a few months.

 

[efikkan]

Most games don't list textures as HD, 2k or 4k. They are low-medium-high-ultra, etc. I can't list a games on one hand that use resolution settings for their textures. I prefer my textures cranked because textures, if you have enough VRAM, have practically no impact to FPS.

And it wouldn't make much sense to list it that way either, as texture quality settings refer to texel density, not the total size of the texture. E.g. an object that will be tiny on screen may have a texture of only 256x256 even on a 4K screen. Not to mention "important" objects may have much higher texture detail than terrain and landscape, something we still see in brand new games. So regardless of settings, many different texture resolutions are used at once.

Generally speaking, having a texture density which is greater than the rendered pixel density on-screen is preferable (so texels > pixels), especially for polygons which are at an angle.

Texture settings will have an impact at performance for either memory bandwidth or TMU-throughput. If you don't see a negative impact, then you're not bottlenecked by these for the game you're running.

 

[lilhasselhoffer]

Appropriate choice.

Note also that by the time of the later movies they were fighting literal ass blasters. No, that is not a joke. The evolution of the evolution of a giant sandworm was a small biped with leathery wings, that flew by literally farting and igniting said flatulence to fly.


Seems oddly appropriate that Nvidia would be the ones to get high off of the smell of their own farts, and extend their dominance of the market by proposing that they use said farts to provide unlimited and impossible levels of energy.

 

[alwayssts]

It's a "600w" connector on the GPU, but the PSU uses TWO 8 pin connectors to send the 600w to the card.
Again, same as the Corsair adapter cable.

"While the included adapters that come with the GPU are fine, what you plug them into is of concern. Technically, 8-pin GPU connectors support a maximum of 150 watts, for a maximum of 450 watts on paper when you use three of them. In reality, most power supplies and most 8-pin plugs can support 324 watts (27 amps at 12 volt) or even 468 watts (39 amps at 12 volts) according to power supply maker Corsair. That, of course, depends on the power supply design. There’s no guarantee of what each PSU can support through those three 8-pin GPU plugs."
This comes from a PCWorld article

Right, it *can* do that. I know that, guy. I very-much know how over-specced they are.
I think AMD/ATi has allowed cards from the factory (not counting unsanctioned v-mods or software before it was regulated) to go over the connector spec approx one time in all my years I can remember.
But they are not going to rate a card over the connector power spec, or they would just give them all 3x8-pin connectors.
That is like saying your 600w connector may be allowed to pull 1000w with the factory bios. It might go over a little, especially with transience, but not a ton.
Are you being factitious or do you really not understand?
I understand *some* might pull over 375w, again I don't know the bios/PL sitch, but I imagine it won't be extremely high bc memory bw is a concern and there's likely another sku they have to sell.
If they allow much over 375w, there's not going to be a lot of juice left in the tank.
Considering I think the high-end sku will be ~3.5ghz stock (and use 3x8-pin), and 1.2v/up to ~3700 might take ~425w (or ~475w if 32GB), I doubt this will encroach on it's territory. It might, but then just bw limited.

We'll know when W1z unhides the bioses tomorrow. The answer, as always, will be to scout out the Nitro+.