Saturday, January 2nd 2021
NVIDIA Could Give a SUPER Overhaul to its GeForce RTX 3070 and RTX 3080 Graphics Cards
According to kopite7kimi, a famous leaker of information about NVIDIA graphics cards, we have some pieces of data about NVIDIA's plans to bring back its SUPER series of graphics cards. The SUPER graphics cards have first appeared in the GeForce RTX 2000 series "Turing" GPUs with GeForce RTX 2080 SUPER and RTX 2070 SUPER designs, after which RTX 2060 followed. Thanks to the source, we have information that NVIDIA plans to give its newest "Ampere" 3000 series of GeForce RTX GPUs a SUPER overhaul. Specifically, the company allegedly plans to introduce GeForce RTX 3070 SUPER and RTX 3080 SUPER SKUs to its offerings.
While there is no concrete information about the possible specifications of these cards, we can speculate that just like the previous SUPER upgrade, new cards would receive an upgrade in CUDA core count, and possibly a memory improvement. The last time a SUPER upgrade happened, NVIDIA just added more cores to the GPU and overclocked the GDDR6 memory and thus increased the memory bandwidth. We have to wait and see how the company plans to position these alleged cards and if we get them at all, so take this information with a grain of salt.
This is only a mock-up image and is not representing a real product.
Sources:
@kopite7kimi (Twitter), via VideoCardz
While there is no concrete information about the possible specifications of these cards, we can speculate that just like the previous SUPER upgrade, new cards would receive an upgrade in CUDA core count, and possibly a memory improvement. The last time a SUPER upgrade happened, NVIDIA just added more cores to the GPU and overclocked the GDDR6 memory and thus increased the memory bandwidth. We have to wait and see how the company plans to position these alleged cards and if we get them at all, so take this information with a grain of salt.
105 Comments on NVIDIA Could Give a SUPER Overhaul to its GeForce RTX 3070 and RTX 3080 Graphics Cards
I'd be happy with a 3080ti 12gb frankly.
Still using my old RX480 and I will continue to use it until normal prices and availability return. If it takes about 4 years so be it.
I paid $300 cnd for this 8GB 256-bit card, with all these generation improvements I would expect no less than double the performance of the RX 480 for the same price $300 cnd.
First generational release stack = preliminary fire, early adopter territory, availability hit/miss. We saw this to some degree with Turing as well, but then Turing wasn't even nearly as much of a leap as Ampere so obviously only specific and especially top end SKUs were hard to get. 2080ti really - they didn't make many, but they did sell them ('all' of them so to speak).
The lower SKUs had only a few, maybe even just one strong contender: the 2060 which dropped the 1080 performance to a lower tier price point. The rest of the non-SUPER stack (and RTX, not GTX, I consider that a one-off for Turing alone to bridge the gap between old and new) was totally uninteresting: the performance was already available for years with the Pascal line up, and at equal or even lower price to boot.
Nvidia gauged the market and response, and then launched SUPER. The only reason to do this is because they think they'll make more money with a better offer within the same gen. After all, its not truly a refresh is it... rather a rapid replacement of old stuff with new. When the volume and stack (perf/dollar/feature) is conservative, its easy to move to a SUPER refresh especially as the node gets better over time.
So maybe SUPER is the new staggered launch Nvidia has always had, but rolled out differently: its no longer feasible to save the top end SKUs for last or even first, they want to hit the entire mid-to-top of the stack virtually at the same time instead of spacing stuff months apart for x104 > x102 (>106) to appear. Advantage: they can recalibrate everything instead of just doing final tweaks on the stuff they have yet to release. In a way, their generational launches are 'more agile'... and more easily changed if the market responds in a strange way. They can sample the market twice per gen, too, helping them for future releases and with low volume, there aren't many cannibalized sales.
This would answer MANY questionable moves of late but more importantly, this approach allows Nvidia maximum risk mitigation, which isn't a bad idea given that they are still pioneering with RT.
And, if their initial bet was OK, they can just postpone SUPER until competition arrives -another thing you don't easily do with an entire SKU like say the x102 die products. Maybe they also took a lesson from Pascal which, for two FULL generations never even needed the 1080ti to begin with. Let's face it... it was lonely at the top for many years and I think that one killed a LOT of Turing sales.Amen to that, baby steps are for toddlers.
BTW, 58 posts per day, bud? Thought I was hardcore :rockout: Welcome :D
Then again... beefier "Super" cards means they'll be even more attractive to mining. So... y... unless cryptocurrency crashes hard, I don't see much chances of people getting RTX cards during this year,... much less getting them at decent / fair prices. Samsung would have to create a production miracle and nVidia would need to be in the mood of risking flooding the market with cards in order for having enough for miners and enthusiasts... which won't happen. Scarcity and high profits are a much safer bet.
And even if it was, why would a "SUPER" card be more attractive? Energy efficiency is king there, not raw power, and that should be the same.
So, super will be... GA102 based?
Anyhow, current RAM configs are nonsenseical, 3070 has less VRAM than current gen consoles, NV does indeed need to "do something".Tables turn, if you only consider newest games, it's 5/2 in 6900XT favor.
AMD seems to have grabbed GPU market share:
realAMD/comments/kp1rob
On the other hand, NV has all the Samsung's fab space to it nearly exclusively, but 3080s are nowhere to be seen. (3070 for 730 Euro is yours, so is 3090 for 1700+). As if that GPU at that price point was an outright lie. (On AMD front it ain't much better pricing wise, 6800 for 850, 6800XT for 950)
And "NV didn't expect 2080Ti perf for $500 would be in high demand" doesn't quite fly.
What they obviously didn't expect, was RDNA2 kicking Ampere's lower bottom.It's been 2 years, chuckle, and we went from Unreal Engine 4 almost ignoring AMD chips, to UE5 ignoring hardware RT capabilities in it's new engines and being demoed on a console with AMD APU, and newest and greatest AAA game often arguably looking better with RT disabled (let alone, faster).
What needs to happen for us to admit that this RT approach will get us nowhere?
Adoption of a new tech is tricky, but this time, it fails on main promise: ease of development. (that "effects unachievable without hardware RT" was a lie, was obvious upfront)
In particular, given how amazing (also light/shadow/reflection effects wise) God of War 2018 is, running on 7870-ish GPU.
This running without any hardware RTing is simply embarrassing:
In Ampere world, there is barely any room for that.
The 3070 costs 25% more than the 3060Ti, despite delivering only 10% more performance.
The 3080 costs 75% more than the 3060Ti with a 44% performance increase, and is 40% more expensive as compared to the 3070 with a 31% performance increase(that's kinda okay).
The 3090 is also a strange one, with 10% performance(only) over the 3080 with a 114% price increase. Seriously, if the 3080Ti is going to be faster than the 3080 and slower than the 3090, where would it sit? At any price it'll make either itself or the 3090 look bad.
Thinking about what happened with the Titan Black and the 780Ti, I'm inclined to think the 3080Ti will be cheaper yet at par with the 3090 or faster, with 12GB of GDDR6X VRAM.
Because the SUPERs would have more memory and more bandwidth? (no expert in mining, though)