Is there any hope for an 18GB 5070 refresh? Not sure where we stand on the odds of that happening.
Nvidia GeForce RTX 50-series product value
- Thread starter Boss
- Start date
My guess is there will be a 16GB 5070 super with ~64 SMs a year from now.
So no 3GB modules? Not sure when they are supposed to be available.
DegustatoR
Legend
They are, Nvidia is using them for laptop parts right from the start.
Yeah but I'm guessing they will use 3GB modules on the 128-bit cards first.
digitalwanderer
Legend
So it appears the MSRP of $750us for the 5070 is total BS:
Was this a glitch? There are 2 MSRP models on the MSI store (linked from videocardz article).
All out of stock obviously. So at least you can not buy a 5070Ti for $750 instead of having to not buy one for a 20% markup over MSRP.
Subtlesnake
Regular
So if Nvidia had simply "refreshed" the 4090, like with the 5080/4080, the 5070 would suddenly be better value?
The idea is if prior x50 cards had 24% of the execution units of the corresponding x90 then that should hold true for the next x50 vs x90. When that doesn't happen it's "shrinkflation". There's no doubt the 50 series is a sideways generation and generally underwhelming but there's no rule that says all SKUs must grow in size equally.
Yep, the thinking is it would actually be better if the 5090 were smaller and slower.
DegustatoR
Legend
x70 you mean? But that makes about as much sense as suggesting that every lineup of all goods must always fit into the same price range and never ever change its positioning inside the lineup in any way. I.e. it never actually happened previously and never will happen. And that's even prior to us discussing the pointlessness of any "name vs name" comparisons.
So I don't know what that comparison should be "laying out" to anyone.
Subtlesnake
Regular
You're getting effectively the same architecture on the same process, with some additional features. Where does the "next-gen" part come in, apart from on the feature side?
It's assigning meaning to a marketing name that was never advertised.
The GTX 275 had 100% of the shading units of GT200B. Every 70 card after that was highway robbery. The GTX 570 was 94% of GF110. Kepler is where it really fell apart. The GTX 770 was only 53% of GK110B. And now the 5070 is a stingy 25% of the 5090. So the 70 class card can be anywhere between 25% and 100% of the big honcho.
Of course you need to ignore that entirely new tiers of performance, die size and cost were introduced over time but that's inconvenient.
I see the "encircled" areas got a $150 price jump between each.
I see the 30- and 40-series "encircled" SKUs got basically zero memory bandwidth bump, but the 50-series got more than a 2x increase over the "encircled" predecessor.
The actual performance bump between the 3050 and 4060Ti is around 47% (Source) and the 4060Ti to the 5070 is about a 53% bump (Source)
So, the 50-series has the same price bump as 30-series to 40-series as encircled in your chart, and the 50-series enjoys the same performance bump as the 30-series to 40-series as encircled in your chart.
Looks like this chart isn't supporting the argument you appear to be making. Cost, performance, and value remained functionally equal between generations in your selected data.
Generally speaking, there are rules about posting a picture or a graph without any of your own context. In your next response, please lay out the argument you're making with the content you posted.
I've had this discussion with a few people in real life in the office at work, and I like to use this thought experiment:
Let's rewind back to the 40 series, and let's assume that NVidia found a near-perfect solution for bonding multiple compute dies into one logical GPU, sort of like how Apple's doing it.
For the sake of argument, let's say that they that they had the tech to bond 2x, 4x, 8x, and 16x AD102 dies together on an interposer with >90% performance scaling, and doing so allowed them to create 4 new SKUs at those performance levels, with price increases roughly aligning with their performance increases.
With the above assumed to be true, some questions:
1) Does the existence of a hypothetical 4090x16 at ~$25,000 with performance to match materially change anything at all about the value of the lower part of the product stack? Does a 4070 now have 16x lower perceived 'value' than before because it has only ~1.5% of the performance of the new top card in the stack?
2) If the answer to the above is 'no', would it change anything if some breakthrough (but expensive) process tech suddenly appeared out of nowhere, allowing them to scale the top end product(s) in the stack into the stratosphere? Like imagining a hypothetical world where TSMC suddenly figured out a 5 angstrom process, but with wafers costing $300,000 each. Or alternatively, a way to break the current reticle limit, and essentially letting Nvidia create wafer-scale chips of arbitrary size, Cerebras-style, so they could make a monolithic ~10,000mm2 monster with ~16x the performance of AD102.
3) If the above answer is also no, then why do people assume the arbitrary spot where Nvidia (or AMD) in any generation chooses to place the top end SKU in the stack matters?
Or alternatively, looking at it another way, imagine a hypothetical world where Nvidia just straight up didn't make an AD102 for the 40-series generation, and the 4080 Super was the top card in the stack, perhaps even given the name 4090. In this hypothetical world, does the 4070 having 58% of the functional units of the top card in the stack somehow increase its value proposition, making it now 'one of the best 70 series cards in history?'
What argument am I making? I haven't written anything. Albquerque Mod mode: This is called bait, and it's a potentially bannable offense. For others reading, I'm leaving this much here as a warning -- do not emulate this behavior.
Last edited by a moderator:
Then your posts are in violation of the rules, thanks for pointing it out. I'll solve the problem..
MOD MODE:
Anyone is welcome to post a chart or a graph or a link to a story, however they also must provide context for their posting of said data / graph / linkage. If you (the royal you, not just Boss) have a position, then you need to state it. If you don't have a position and just want to make things argumentative, don't click submit on your post. This is not a troll forum, nor a troll thread, and I will absolutely get rid of a troll without any remorse.
I've removed a few responses which might make the thread a little hard to follow for a bit. The article itself can be reposted if and when context and poster's position can be stated in the same reply and not used as "gotcha!" baiting to other respondents.
Last edited:
Similar threads
