Overclock.net banner

So... the 5070 is DOA?

1 reading
5.7K views 32 replies 19 participants last post by  wericsson6  
#1 ·
12GB Vram, hyped as a 4090- tier GPU so people will buy it for 1440P & 4K, the new tech only shaves 400MB Vram usage...

Games at these resolutions also eat 16GB on UE5.

So 550 for a GPU that will have to make do with lowered textures to use RT at 1080P or lower.


 
#2 ·
I have a sneaky feeling the "4090-class" claim is with the 4090 using DLSS3 FG and 5070 using DLSS4 FG. Kinda like how NV claimed omgwtfbbq% gainz from 3090Ti -> 4070Ti- but only with FG enabled on the Ada card.
Regardless, I agree that 12GB VRAM is gonna be cramped. I guess we'll have to wait for 5000 Super for an improvement there with 3GB chips for 18GB on 192-bit & 24GB on 256-bit bus.
 
  • Rep+
Reactions: Sir Beregond
#3 ·
The marketing spin borders on the absurd and the card is a pretty modest upgrade from the 4070 Super, but I would not call it dead on arrival.

12GiB is sufficient, though barely, and the performance probably won't be that bad for $550.
 
#4 ·
There's even a small disclaimer on their presentation that 40 series uses Frame Gen and 50 series uses Multi Frame Gen meaning up to 3-4x generated frames. Don't get me wrong, I am not against this, especially since apparently it also looks better and doesn't impact input lag any more than FG so thats a win, but the performance claims are wildly misleading. Even the 5080 is not going to be faster than a 4090, not with 10k cores, 16GB VRAM and 980GB/s.

And this whole MFG could literally just be enabled on 40 series GPUs and get the same benefits and my guess is, it will be in about a year or so. Literally no reason it cannot run on 40 series.

I feel like this generation Nvidia just focused more on software development and less on actual GPU upgrades since they are unchecked by AMD. If you look at the specs, the jump from 4090 to 5090 is very modest in terms of SP cores, ROPs, TMUs, Tensor and RT cores. Yes, its an architectural uplift, but once the die shots will be available, I bet it won't be any revolutionary architectural design, just a scale up from the 4090, slapped faster VRAM on it and voila, 5090 at $2000.
 
  • Rep+
Reactions: StreaMRoLLeR
#5 ·
I feel like this generation Nvidia just focused more on software development and less on actual GPU upgrades since they are unchecked by AMD. If you look at the specs, the jump from 4090 to 5090 is very modest in terms of SP cores, ROPs, TMUs, Tensor and RT cores. Yes, its an architectural uplift, but once the die shots will be available, I bet it won't be any revolutionary architectural design, just a scale up from the 4090, slapped faster VRAM on it and voila, 5090 at $2000.
The 5090 has over 30% more SMs than the 4090, memory bandwidth is also increased by more than 80%
For all of that, Nvidia is claiming an incredible 30-40% performance improvement in Far Cry 6 with RT and A Plague Tale Requiem with RT and DLSS3.

The 5070 is supposed to have similar gains over the 4070, despite not getting nearly as much added hardware, so it's not a matter of GPU size. AD102 and GB202 just seems to scale extremely poorly.
 
#6 ·
The 5070 looks to be a card that performs roughly similar to the 4070 Ti Super, but also supporting DLSS 4x FG. The VRAM situation is kind of bad, but it's really not a terrible deal for $550.

What's the alternative? Buying a 7900 GRE, without FSR4 or DLSS support, and worse performance, for a similar price?
 
#7 ·
Braindead take, most new games especially the aaa one will have dlss4 support which makes it as fast as a $1600 RTX 4090, the rest is pure talk and to be fair quite delusional, i mean what did you expect RTX4090 performance in pure raster?
Just lol.
The RTX 5070 is aimed at RTX 4070 users, in raster it will be about 20 to 30% faster which isnt bad, in dlss probably 2x the performance, more features like reflex 2 etc, power consumption should be the same.
To me looks reasonable, not amazing but definitely not DOA as you say
 
#8 ·
Gen Z take.
If I take to the give and give to the take, then 12GB Vram is some how carrying a 4090.

The rest is pure zoom zoom.
 
#10 ·
No worries after a year they release RTX 5070 Super with 18Gb Vram by using 3gb modules ;)
 
  • Rep+
Reactions: 726650
#11 ·
all fake FG marketing bs - can we pay nvidia with fake money for fake frames?
 
#15 ·
What 5070 did is effectively kill any remaining resale value in there 3080/3080ti and maybe 3090. So those people upgrading to 5080/5090 effectively may have to give there old cards away.
 
#16 ·
AMD has a opportunity here to come with 9070 XT for $500.
9070 XT 16GB vs 5070 12GB is much better positioned than 7800 XT 16GB vs 4070 12GB simply due to 12GB becomes less and less 2 years after.
If FSR4 is decent and RDNA4 RT shows good improvement it might be the card to get.
 
  • Rep+
Reactions: BeeHiveClusterFck
#17 ·
Possibly, AMD did claim perf numbers are incorrect from leakers, so that could mean less or more from the main product, but the press kit did state 4070 Ti - 7900 XT level perf.
Which.... if as you say the software is decent it may catch on for some buyers.
 
#19 ·
5070 = 4090 is definitely going to be only when using DLSS4 multi-frame gen.

Only other improvements are going to come from architecture, move to G7 memory, and marginal improvement on process for the silicon (I believe move to N4P from N4?), but looks like power may be going up a lot too for some of these.

Of note, here's the CUDA core count difference between the two gens otherwise:

4090 (16,384) -> 5090 (21,760) = 32.81% increase
4080 Super (10,240) -> 5080 (10,752) = 5% increase
4070 Ti Super (8,448) -> 5070 Ti (8,960) = 6.06% increase
4070 Super (7,168) -> 5070 (6,144) = 14.29% decrease
4070 (5,888) -> 5070 (6,144) = 4.35% increase

Definitely going to be smaller improvement on traditional rendering performance unless the memory was the bottleneck I figure.
 
#23 ·
To be fair, that was in BlackOps 6 (if we're talking abt the sneaky unofficial CES journo benchmark run) which heavily favors Radeon- 7900XT also matches 4080S and 7900XTX handily beats it- but it does show promising performance. 9070 matching 7900XT in raster is a good result if the price is right.
 
#25 ·
The 9070XT needs to be 550 USD and come close to a 5080. AMD has a branding problem with gamers and they need to overcome that with a super aggressive price AND decent performance. the 9070 needs to 400 to 450 and trade blows with the 5070 OR beat it.
 
  • Rep+
Reactions: Ohio Buckeye 06
#27 ·
I really hope they do, I don't want to pay nvidia tax anymore and I want to move to SteamOS and I figured AMD GPU's are better due to drivers in linux
 
  • Rep+
Reactions: Darkpriest667
#29 · (Edited)
Nvidia is in full Apple mode by now, so they are only willing to sell something minimally decent that will last if you have $750 to spare for the 5070 Ti. And even then the segmentation is far from optimal, with obvious gaping holes for the Super series in a year or so.

As to the RTX 5070? It's a card designed FOR Nvidia, not for consumers. It's a way to upsell the 5070 Ti. Anyone who wanted a 12 GB card with around that level of performance should have bought an AIB RTX 4070 Super OC model in the past year. My advice if you really want something in this price segment: either wait for the competition to show what they have, get a good deal on an outgoing RTX 40 series GPU or wait for the Super series that will come with higher density VRAM chips and more VRAM.

Buying a 12GB card in 2025 for $550 with a 250w (!) TBP is all sorts of wrong. It's not a good forward looking purchase, simply put. I honestly don't know how they can reconcile the 12 GB frame buffer with the marketing BS that says "4090 performance". Yeah, 4090 performance with the tires of a bicycle. It doesn't add up and they know it. It's all DLSS 4 multiple fake frames anyway.

Let's face it, they have no competition and they have better uses for their allocated wafers at TSMC with AI cards for enterprise.

Nvidia doesn't care. This $50 discount on the 5070 and 5070 Ti is their absolutely minimum spec. redemption tour when they have to pretend they care after Ada Lovelace, but they really don't. They have no competition. Not to mention that they are passing the increased power consumption costs of these new cards to consumers.

-------------------

On another note, it's concerning how much people are ok with ever diminshing performance tier upgrades. Back in the Pascal series a GTX 1060 had the performance of the previous high-end GTX 980, so we got a two tier upgrade, and the 1060 has more VRAM (the initial 6 GB version, of course).

Then, in the Turing line-up, the RTX 2060 had the performance of a 1070 Ti (it has since improved a bit to around 1080 level because newer games use more DX12 features), but has less VRAM. We're now down to one and a half tier upgrade, with more performance coming after a few years, but with less VRAM.

Then Ampere - the RTX 3060 has the performance of an RTX 2070 and has more VRAM (but let's be real, they put 12 GB because 6 GB would be ridiculous after the 2060 Super and also because 8 GB wasn't practical at the time with the available technical solutions). Ok, down to a one tier upgrade.

Ada Lovelace - hahahah, well, the RTX 4060 has less VRAM than the 3060 and can't even match the 3060 Ti and sometimes even loses to the 3060 because of a lack of VRAM, heck, sometimes not even the 4060 Ti can match the 3060 Ti because of the narrower memory bus and half the PCIe lane count, while still only sporting 8 GB on the initial model.

And now Blackwell, where the RTX 5070 will at most match a 4070 Ti without DLSS 4 multiple fake frames trickery.

So, we are now down to an upgrade of half a tier.

Half a tier. The more you buy, the more you save, I guess, somehow.