Wednesday, December 25th 2024

NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

Yesterday, 03:15 Discuss (132 Comments)

Thanks to kopite7kimi, we are able to finalize the leaked specifications of NVIDIA's upcoming GeForce RTX 5070 and RTX 5070 Ti graphics cards.
Starting off with RTX 5070 Ti, it will feature 8,960 CUDA cores and come equipped with 16 GB GDDR7 memory on a 256-bit memory bus, offering 896 GB/s bandwidth. The card is reportedly designed with a total board power (TBP) of 300 W. The Ti variant appears to use the PG147-SKU60 board design with a GB203-300-A1 GPU. The standard RTX 5070 is positioned as a more power-efficient option, with specifications pointing to 6,144 CUDA cores and 12 GB of GDDR7 memory on a 192-bit bus, with 627 GB/s memory bandwidth. This model is expected to operate at a slightly lower 250 W TBP.

Interestingly, the non-Ti RTX 5070 card will be available in two board variants, PG146 and PG147, both utilizing the GB205-300-A1 GPU. While we don't know what the pricing structure looks like, we see that NVIDIA has chosen to make more considerable differentiating factors between its SKUs. The Ti variant not only gets an extra four GB of GDDR7 memory, but it also gets a whopping 45% increase in CUDA core count, going from 6,144 to 8,960 cores. While we wait for the CES to see the initial wave of GeForce RTX 50 series cards, the GeForce RTX 5070 and RTX 5070 Ti are expected to arrive later, possibly after RTX 5080 and RTX 5090 GPUs.

Sources: @kopite7kimi #1, @kopite7kimi #2

Add your own comment

132 Comments on NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

#101

Hecate91

Vayra86This is the thing really. Its never a criticism at anyone in particular its just a shame we get so little hardware for that money, and that I think is something we should signify and stress every time. Because underneath that we really need to realize this is not the market functioning as it should. The game is rigged now, and the margins made are absolutely silly. - Margins made over OUR backs with OUR money.

Agreed, the amount of performance uplift for the cost compared to how it was with the GTX 10 series for example sadly just isn't something we get anymore, I always bought Nvidia cards but since the RTX 30 series the game is rigged towards pushing consumers towards spending more on the xx80 cards, the xx70 has been too much of a compromise for the money, IMO.

nguyenYou couldn't be more wrong, 1080Ti was even further removed from the rest of the stack than 4090 in term of performance

1080 Ti was 35% faster than 1080, meanwhile 4090 beats 4080 by 31%.

2070 Super has 30% less CUDA Core and 10% less Bandwidth than 1080Ti, yet still match 1080 Ti in older DX11 games

TSMC 12nm (2070S) was only a tweaked node vs TSMC 16nm (1080 Ti), TSMC introduce them as same node just like their N5/N4nm nodes.

Nvidia has been very consistent in the past decade in making 104 die perform just as well as previous gen 102 die, but now we will see 103 die (GB103) carry that torch.

The 1080Ti was only priced 16.67% more than the 1080, ever since the RTX 30 series the naming stack completely changed and the performance gap between tiers is much larger.

#102

Visible Noise

Sound_CardYou have a 4090. You spent somewhere between $1200 (doubt) to $2000 dollars for a graphics card, then you say complaints are just nitpicks. This is the most hysterical post I have read here in a while.

I have a 4090. Why is the non-stop, unoriginal complaining about its price being annoying so hilarious to you? In what way do you think it contributes to the conversation? Does posting the exact same thing 2,000 times instead of 1,000 make it better commentary?

hearing the same sour grapes day after day is boorishnes.

sepheronxFunny how some game developers have put limitations for using PT when it can be hacked and enabled and working on cards with 10GB of VRAM

nvidia/comments/1hbak60
So I also think game developers are working with nvidia to create artificial limitations.

So 12GB should be actually enough for a while. I just think that game developers are working in tandem with other companies to also create limitations that aren't necessary.

What are you talking about. They patched the game to run on 6GB cards. At least get your facts in order before going with the tin foil hat conspiracy theories.

#103

Onasi

nguyenYou couldn't be more wrong, 1080Ti was even further removed from the rest of the stack than 4090 in term of performance

1080 Ti was 35% faster than 1080, meanwhile 4090 beats 4080 by 31%.

Orly?

The 4090 was still and potentially IS still CPU bottlenecked. We are seeing it in real time now with the 9800X3D. In CPU heavy titles like BG3 anyway. Besides, I was comparing with the 4070Ti, since that was the measuring stick the previous poster used.

nguyenNvidia has been very consistent in the past decade in making 104 die perform just as well as previous gen 102 die, but now we will see 103 die (GB103) carry that torch.

Yes. And the full AD104 (4070Ti) is around 10% slower than the full GA102 (3090Ti). And I said that I can see the 5070Ti getting in that ballpark. Not sure what you are disagreeing with here.

#104

nguyen

OnasiOrly?

The 4090 was still and potentially IS still CPU bottlenecked. We are seeing it in real time now with the 9800X3D. Besides, I was comparing with the 4070Ti, since that was the measuring stick the previous poster used.

Yes. And the full AD104 (4070Ti) is around 10% slower than the full GA102 (3090Ti). And I said that I can see the 5070Ti getting in that ballpark. Not sure what you are disagreeing with here.

166% / 127% = 30.7% faster, not sure what you are getting at.

4090 is not the full die at all, it has 30% of total L2 cache disabled, 4090 is even more cut down than the 3090 that got beaten by 4070Ti. If the 5070 Ti get into the ball park of full AD102 die of course it will be faster than 4090.

#105

sepheronx

Visible NoiseWhat are you talking about. They patched the game to run on 6GB cards. At least get your facts in order before going with the tin foil hat conspiracy theories.

To run the game at 6GB cards or to run Path Tracing on below 12gb cards?

#106

Visible Noise

sepheronxTo run the game at 6GB cards or to run Path Tracing on below 12gb cards?

Both. You can path trace on your 3080 now.

Edit: Holy shit man, that Reddit post is over two weeks old. Get with the times.

#107

sepheronx

Visible NoiseBoth. You can path trace on your 3080 now.

Edit: Holy shit man, that Reddit post is over two weeks old. Get with the times.

good to know they finally removed the limit without mods.

#108

Onasi

nguyen4090 is not the full die at all, it has 30% of total L2 cache disabled, 4090 is even more cut down than the 3090 that got beaten by 4070Ti.

I… never said it is? There IS no full AD102 GPU to make a full comparison with. But it’s hard to argue that the amount of total reaources availiable between the 4070Ti and 4090 is not ridiculously skewed. I fully believe that we aren’t seeing its full potential yet. In fact, all depends on the games. If you take the averages from the older test bench the difference between the 4080 and the 4090 is around 36%. So let’s say a 30-35 performance delta. That’s between 102 and 103 chips, by the by, not 104. The 5070Ti might get close to the 4090, sure, but I don’t think it will beat it. Or, hell, if it will I firmly believe it will then cost even more than the already overpriced 4070TiS.
Frankly, judging by these leaks and the insane delta between the 5070 and the Ti model the 5070Ti is a xx70 class GPU in name only - it’s essentially a 5080. Which in turn is a 5080Ti if we go by generational nomenclature of the past and so on. Which, I guess, supports your point, huh. We’ll see.
Good shout earlier about the 12nm process, by the way, I forgot it was just a denser optimization and not a full new node. Not aure how it compares to the 5N to 4N at this stage. About the same, I reckon?

#109

nguyen

OnasiI… never said it is? There IS no full AD102 GPU to make a full comparison with. But it’s hard to argue that the amount of total reaources availiable between the 4070Ti and 4090 is not ridiculously skewed. I fully believe that we aren’t seeing its full potential yet. In fact, all depends on the games. If you take the averages from the older test bench the difference between the 4080 and the 4090 is around 36%. So let’s say a 30-35 performance delta. That’s between 102 and 103 chips, by the by, not 104. The 5070Ti might get close to the 4090, sure, but I don’t think it will beat it. Or, hell, if it will I firmly believe it will then cost even more than the already overpriced 4070TiS.
Frankly, judging by these leaks and the insane delta between the 5070 and the Ti model the 5070Ti is a xx70 class GPU in name only - it’s essentially a 5080. Which in turn is a 5080Ti if we go by generational nomenclature of the past and so on. Which, I guess, supports your point, huh. We’ll see.
Good shout earlier about the 12nm process, by the way, I forgot it was just a denser optimization and not a full new node. Not aure how it compares to the 5N to 4N at this stage. About the same, I reckon?

Well if Nvidia could make the 4090 2 years ago, they definitely could make an "cost optimized 4090" now using slightly smaller die and smaller bus, Nvidia has been very consistent doing so for a long time. The only question is what price will they charge for the "cost optimized 4090".

This is my guess:
5070Ti = 4090 = 1000usd
5080 = 4090 +10% = 1200usd
5090 = 4090 +50% = 2000usd

#110

Onasi

@nguyen
Yeah, possibly. I agree that the closest to “reasonable spec” enthusiast card will be at least 1000. It’s just how it is nowadays. Not sure if the 5070Ti and 5080 will only be 10% apart. Base 4070Ti and base 4080 were what, 25-30? I think NV will want to keep the same (roughly) delta so as not to compete with itself and find ways to choke/restrict 5070Ti performance.
If I had to take a stab at the full stack it would be thus:

5060 - roughly 4060Ti performance
5060Ti - base 4070, maybe a tad more. 4070S level if we are lucky
5070 - base 4070Ti to 4070TiS level
5070Ti - 4090 -10%ish
5080 - 4090 +10%ish
5090 - 4090 +40%ish at a ridiculous price

#111

Ibotibo01

Onasi@nguyen
Yeah, possibly. I agree that the closest to “reasonable spec” enthusiast card will be at least 1000. It’s just how it is nowadays. Not sure if the 5070Ti and 5080 will only be 10% apart. Base 4070Ti and base 4080 were what, 25-30? I think NV will want to keep the same (roughly) delta so as not to compete with itself and find ways to choke/restrict 5070Ti performance.
If I had to take a stab at the full stack it would be thus:

5060 - roughly 4060Ti performance
5060Ti - base 4070, maybe a tad more. 4070S level if we are lucky
5070 - base 4070Ti to 4070TiS level
5070Ti - 4090 -10%ish
5080 - 4090 +10%ish
5090 - 4090 +40%ish at a ridiculous price

So, the best GPU among them would be 5060 Ti because it has 16GB of VRAM and is priced at $399. Also, 5070 Ti would be the high-end option.

#112

Onasi

@Ibotibo01
It will be a decent “value” (insert me rolling my eyes here) option IF it’s actually 400 bucks. We don’t know that yet. For all we know, NV might decide to send the entire stack into stratosphere and price the 5060Ti at, like, 550 with a 350-400 base 5060. This is the same company who unironically asked for a 100 bucks more on the same 4060Ti with just double the VRAM.

#113

TumbleGeorge

Onasi5070Ti - 4090 -10%ish

-30% too small number of cores to have better ratio.

#114

Onasi

@TumbleGeorge
I am being very conservative/cautious after my talk with @nguyen, so I chose to err on the optimistic side here.

#115

forman313

gffermariNo, I mean if they bring neural texture compression in this gen, 12GB would be enough for a x70 class gpu.

That actually looks promising, thank you for the links.

I haven´t had my morning coffee yet and this is not my area of expertise, but to me this looks like it could work. Lets hope it´s not pushed too far. I.E. closing the 8-12 and 12-16GB gap, not 8-16 or something like 6-12GB.

#116

TumbleGeorge

The more data is compressed, the more performance will be required for compression/decompression. I wonder if there will be any resources left for actual work with the models while the logic is loaded with compression/decompression operations?

#117

Outback Bronze

oxrufiioxoPricing definitely sucks for everything but they really are raw dogging us no lube....

Yeah, you'd think Nvidia should be able to afford a pot of lube packaged with a 5090 :)

#118

Vayra86

OnasiI… never said it is? There IS no full AD102 GPU to make a full comparison with. But it’s hard to argue that the amount of total reaources availiable between the 4070Ti and 4090 is not ridiculously skewed. I fully believe that we aren’t seeing its full potential yet. In fact, all depends on the games. If you take the averages from the older test bench the difference between the 4080 and the 4090 is around 36%. So let’s say a 30-35 performance delta. That’s between 102 and 103 chips, by the by, not 104. The 5070Ti might get close to the 4090, sure, but I don’t think it will beat it. Or, hell, if it will I firmly believe it will then cost even more than the already overpriced 4070TiS.
Frankly, judging by these leaks and the insane delta between the 5070 and the Ti model the 5070Ti is a xx70 class GPU in name only - it’s essentially a 5080. Which in turn is a 5080Ti if we go by generational nomenclature of the past and so on. Which, I guess, supports your point, huh. We’ll see.
Good shout earlier about the 12nm process, by the way, I forgot it was just a denser optimization and not a full new node. Not aure how it compares to the 5N to 4N at this stage. About the same, I reckon?

You keep going like that and I'll start believing Blackwell is the new Pascal and I need to sell off my 7900XT

Honestly, if the 5080~5070ti gets a half decent price I might, just for shits and giggles. Given the large gap to x90 and the stagnant behaviour of the previous gen Blackwell might actually look quite good, because otherwise, what's Nvidia going to sell us anyway?

nguyenWell if Nvidia could make the 4090 2 years ago, they definitely could make an "cost optimized 4090" now using slightly smaller die and smaller bus, Nvidia has been very consistent doing so for a long time. The only question is what price will they charge for the "cost optimized 4090".

This is my guess:
5070Ti = 4090 = 1000usd
5080 = 4090 +10% = 1200usd
5090 = 4090 +50% = 2000usd

I'm kinda thinking the 5070ti might end up around 800,- as someone else predicted earlier somewhere. I also don't think it'll be a 4090 level perf card, but under it, and the 5080 might nip at its heels instead.

#119

Onasi

Vayra86You keep going like that and I'll start believing Blackwell is the new Pascal and I need to sell off my 7900XT

It might be… at three times the price, but who’s counting?

Vayra86because otherwise, what's Nvidia going to sell us anyway?

Well:
Option A - DLSS 4, now with quadruple the detail on generated frames!
Option B - who the fuck cares, we hold 90% of the market, we can release whatever with mild performance increases and it will sell.

#120

RedelZaVedno

I'm still sticking to my 20yo buying formula, so AMD and NGeedia can do whatever they please.
I buy new GPU when it's 50% faster in rasterization than what I have and costs the same adjusted for inflation.
I got 4070TIS for 700 bucks so the only upgrade path is something in a range of today's 4090 for 720 bucks max.
I don't see anything on horizon that would offer it this gen. So I'll just wait another 2 to 3 years, maybe even 5. No biggie.
If these companies don't want our money anymore and are all in on AI so be it. More than enough good old games I haven't played yet. I can wait.

#121

The Shield

gffermari5070 and 12GB ...is not that bad, if nVidia introduce something to compensate.

With the obscene price they are going to charge the 5070, the only thing they need to introduce are further 8 Gb to transform that 12 in a 20.

#122

Dawora

THANATOSOnce more only 12GB Vram? Not like I didn't expect It, but still.

Last time it was 8GB what ppls was expecting...

AtiAmdI hope AMD will release rx 9060 with 4070 ish perfomance at €400 :)

No need to hope, there will be 5060Ti 16GB

RedelZaVedno5070Ti looks like 4070TI Super +10% via faster memory.... and even 5080 specs don't look all that impressive. Ngeedia gives us no incentive to upgrade this gen.
It looks like all large dies are eaten by

U can OC 4070Ti +10%

So its kinda stupid to think 5070Ti is only 10% faster vs 4070Ti

5070Ti is close to 4090

Prima.VeraThat 5070 card should be a 5060, while the Ti version, should be a normal 5070....
And people are asking why we call nVidia, nGreedia.
Proof. ;)

Dont buy 5070
Buy 5060Ti or 5070Ti
Or buy AMD again

freeagentLol. People complaining about prices in every single thread need to find a new hobby :laugh:

500$ from nvidia GPUs is expensive and greedy
500$ from AMD GPUs is just fine.

Forums are full of fans who trash talk Nvidia atm i can feel the pain of Amd fans

#123

freeagent

DaworaForums are full of fans who trash talk Nvidia atm i can feel the pain of Amd fans

I think that people forget inflation is for real, and things are a lot more expensive then they used to be. Its ok to like one or the other, or even both. They make computer hardware.. devoting your life to spreading the word about hardware is a waste. People who have been bit before are highly unlikely to spend a lot of money on a piece of hardware from brand X just as an experiment. They could use that money and buy what they know works for them. This is why when you work on something, it is important to do the best you can, because people are watching.

#124

Dawora

DahitaI just want a $500 16GB decent card for 1440p, or a $1000 24GB . Was that really too much to ask for 2025?

Ppls only see one GPU, its like there is no other otions..

How about 5060Ti 16GB or 5070Ti 16GB?

#125

Krit

OnasiOrly?

The 4090 was still and potentially IS still CPU bottlenecked. We are seeing it in real time now with the 9800X3D. In CPU heavy titles like BG3 anyway. Besides, I was comparing with the 4070Ti, since that was the measuring stick the previous poster used.

Yes. And the full AD104 (4070Ti) is around 10% slower than the full GA102 (3090Ti). And I said that I can see the 5070Ti getting in that ballpark. Not sure what you are disagreeing with here.

Comparing GTX 1080Ti to RTX 4090 is complete mismatch and dishonor. GTX 1080 Ti has way, way better price/performance ratio than RTX 4090 in straight comparision. And 8800 GTX / 8800 GT was, way better than even GTX 1080Ti / GTX 1070...... Performance gains were tremendous for almost the same price! Where's the RTX 4090 is complete piece of shit in this region.

Add your own comment

NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

132 Comments on NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts

NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

Related News

132 Comments on NVIDIA GeForce RTX 5070 and RTX 5070 Ti Final Specifications Seemingly Confirmed

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts