Rumored Specs for NVIDIA GeForce RTX 5070 and RTX 5070 Ti Surface Online Showing Up to 300W TDP

Peter_Brosdahl

Moderator
Staff member
Joined
May 28, 2019
Messages
9,064
Points
113
With less than two weeks until the expected official RTX 50 Series reveal more rumored specs for NVIDIA's RTX 5070/5070 Ti have been leaked.

See full article...
 
Ok, so it's still rumor

But still.

300W for a 70 series?
 
I read rumors that nvidia having some new fancy tech that cuts down memory usage, but then again I read the same thing with Lovelace and that never happened.

12GB just doesn't cut it anymore for 2k gaming much less 4k.
 
12GB on the rumored 5070 is an absolute joke in 2025.
Yeah, but Nvidia would have to sacrifice a tenth of a percent of margin to run the extra memory channels for the wider bus!

Cause it's not like they're going to make a 24GB variant, lol.
 
I read rumors that nvidia having some new fancy tech that cuts down memory usage, but then again I read the same thing with Lovelace and that never happened.

12GB just doesn't cut it anymore for 2k gaming much less 4k.

Neural rendered compression of material textures - but IIRC it still requires dev support.
 
Yeah, but Nvidia would have to sacrifice a tenth of a percent of margin to run the extra memory channels for the wider bus!

Cause it's not like they're going to make a 24GB variant, lol.
I've seen rumors on a 24GB RTX 5080, but its been a while.
 
I'd be interested to see memory usage testing at 1440P, which is what this card is aimed at. Will the 5070 even be able to push 1440P at max settings to saturate 12 GB of VRAM? I'd think the 5070 Ti would be the card to push limits, and that comes with 16 GB.
 
Yeah, but Nvidia would have to sacrifice a tenth of a percent of margin to run the extra memory channels for the wider bus!
Cause it's not like they're going to make a 24GB variant, lol.
They tried "creative" approaches for VRAM capacity before - with the segmented 128 + 64 bit bus on the Geforce GTX 660 Ti and later the infamous 224 + 32 bit bus on the GTX 970 and they had to pay dearly for that in the class action lawsuit (yes, I know, mostly went to the lawyers). I don't think they would want a repeat of such a scenario and I'm not entirely sure how to effectively inform an already rabid, bloodthirsty customer base without it becoming at least a major PR disaster. It's a shame because while I don't design GPU's, I believe with todays technology that config could be made to work pretty effectively, heck Nvidia even has a pretty decent main RAM over PCI-E swap algorithm these days, some people don't even notice it.

That said, as far as raw capacity goes I'm cautiously hopeful that Intel may actually be able to light a fire under their *** this time, given how popular the B580 has been so far so the next generation after Blackwell may actually have an adequate amount of VRAM, though that won't help Nvidia buyers today.
 
Become a Patron!
Back
Top