bugThere are games where a 4060 card won't average 60fps at FHD. What would you do with a 4050?

NVIDIA should really take the opportunity with the 5000-series to rebalance their product stack naming, so that we don't have everything clustered above 50.

I'm not sure why so many are surprised - figured they'd pull a refresh once they said no next-gen until 2025.
The big question is how much of a leap 50xx series will be, and if it's a repeat of 30xx (instantly eclipsing the 20xx Supers).

good ol Nvidia, pretend to not care about competition by not dropping prices and instead just introducing endless SKU's

IsaakI don't really care to play the latest and greatest AAA games at max settings. I'd just like an entry-level modern Nvidia GPU for gaming and image/video editing. I still play everything I want on my 1650S, but it's gonna die someday and I'm weighing my options. Guess 300€ is the new entry-point.

A true x50 hasn't existed since the 16 series. It's killing me that the market for decent GPUs you can slap into a shitty prebuilt with an OEM power supply (i.e. 75W cards) is being neglected, because it's so important for the PC gaming segment's existence. Bringing the entry cost of PC gaming from whatever it is now (~$700? Give or take?) back down to ~$300 or less means that PC gaming is still cheaper than console, which is why this market flourished so much in the early 2010's in the first place. If we stop producing budget parts, the PC market will continue to shrink, and become more and more inaccessible to more and more people. The abominations of the RX6400 (with limited PCIe lanes gimping it on the older systems it would typically be installed on) and the downright insulting GTX 1630 (a card that clearly should have been 30 watts in line with the other x30 series, but instead was brought up to 75 Watts yet still slower and more expensive than the 1650 that came out 2 years earlier in the 75W segment) almost look like GPU manufacturers are deliberately trying to kill this market for whatever reason.

>There are games where a 4060 card won't average 60fps at FHD. What would you do with a 4050?

Regarding this mindset - I would argue that I have very nearly as much fun playing a game at 1080p/low as I do at 1440p/high or even better. Sure, there are some games that have a "wow" factor that you may miss some part of, but that doesn't affect my enjoyment all that much overall. My first GPU was a GTX 950 when money was really tight, and boy did I love that thing. Going from being unable to play what I wanted on a shitty laptop to being able to play whatever I wanted was (and probably always will be) the biggest upgrade I've ever gotten!

Double-ClickI'm not sure why so many are surprised - figured they'd pull a refresh once they said no next-gen until 2025.
The big question is how much of a leap 50xx series will be, and if it's a repeat of 30xx (instantly eclipsing the 20xx Supers).

I don't think it's much of a question; just add the bandwidth from 32Gbps and the extra cache/bus width of a 512-bit card and you have an approximation of your answer.

The "?" is GB205, which is less clear. Considering it is '05' and not '04', I'm under the current impression it will be 192-bit/GDDR7 or 256-bit GDDR6(x). If it were 256-bit GDDR7, it would likely be called GB204.

As I've said before, it's likely 4070Ti Super will be aimed at not giving people 50TF (ootb, *maybe* with OC), which *appears* an important metric for many games currently. 50TF+ *could* give an extra bump in DLSS/FSR scaling at certain resolutions or scaling from a PS5pro, which it appears they will keep as the reason to buy a 4080. It's also likely around ~1/2 of what the 2028 consoles will be. nVIDIA likely wants to keep such a card below that metric to encourage selling new cards at that time.

Enter GB205, which will likely be slightly better than 7900xt/Navi 4/BM. Maybe 20% (if you do the math), or one product/half a segment higher; perhaps similar perf (but *just-enough* better than a 4070 Ti Super) for a cut-down version. While it will likely have improved RT versus a AD104 and perhaps even AD103 due to clockspeed (~3300mhz?), you'll have to ask yourself if something like that is worth waiting for and paying the likely premium (~$800 again?). I think it will be something like 9216sp @ higher than proportional clock than the 4080 (10240sp) refresh, and I will be thoroughly amused if it is still 12GB. That's ofc why we must wait and see, but you can start to see where they would pigeon-hole it and whom would be the potential market. If they went 256-bit/24gbps GDDR6(x) the clock could be *slightly* higher and be more in-line with what to expect from n3e products (~3700-3800mhz?) or a diff config due to cache, but still in that approx realm of performance. >20%, but less than any meaningful amount stronger (wrt raster).

TLDR: A meaningful update in raster from 4070 Ti Super, but perhaps not 7900xt/Navi 4/BM/4080. Once again, this is how they 'getcha', and also why AMD/Intel are targeting that exact performance level. I don't know how Navi 4/BM will stack up *exactly* (should be pretty close), but if you are one those people that aren't sold on (what I consider compromises for) RT, I think 7900xt is going to have a VERY long life all things considered. I don't know if the same will be true of 4070 TI Super, which to me looks like a stepping stone to people buying GB205.

why?

4070 Ultra would be a better naming. Ti Super sounds awful.

I would say why a bit late, January , ,but clearly they're happy slapping Christmas shoppers with a one month old and slightly not as good feeling.

Odd.

If the 4090 is going to remain the top card, then the Supers should really just be cheaper. There is a 20% gap in between the 4090 and 4080 at 4k. Say the 4080 Super shortens that gap to 10%, in the real world no one will notice 10% more frames or 10% less frames.

Here we are, release the next gen card a year and a half after the 4080 release and we get a 10% bump in performance. All I say is it better come with a price drop too if they want my money.

TheoneandonlyMrKI would say why a bit late, January , ,but clearly they're happy slapping Christmas shoppers with a one month old and slightly not as good feeling.

Odd.

Standard practice, they want to clear non-SUPER stock for BF and Xmas to create demand for the SUPERs when those launch. NVIDIA isn't launching SUPER because they feel extreme competition, they're simply doing a middle-of-the-road refresh/rebrand for investor optics, and for investor profits it makes most sense to try to clear out most old stock first.

Яid!culousOwONeed a 4060 SUPER

Don't you mean 4050 Ti!

I want a 4070 Supper please. Together with a 4080 Super Duper Ti Ultra Extreme Plus. And some fries with it.

This is a good news for me, who is rocking an aged 1080ti that could kick the bucket any minute now.

A super card that can do 4k 60 raster comfortably with a 200w power envelope and costs around $250-300 ?

Nvidia must already be developing an additional layer of exclusive technologies for the 5xxx series in order to justify the 20-30% increase in price.

I don’t find the naming confusing.
Super to me means a little extra performance, using the full or close to that, die.
The 4070Ti is, regardless the name, a completely unique product in the series. So it’s normal to get a bump as long as the dies allow.

A 4080 super using the full ad103 would be a joke. The bump is not enough to justify a new sku. It’s about 500 more CUDA cores.

So a cut down ad102 for a 4080Ti would be the only logical way to go.

will we see a 4070 super mobile laptop gpu?

Cool, a 4080 Super sounds like it could be my next GPU.

mrdumbasspremiumRTX 4090 price breakdown (these prices are grabbed from the half point of the cards launch till new generation of cards launch or present/averaged in its manufacturing peak)
(AD102 die) costs about 200-220 USD per die (roughly 18K-20K USD per wafer including the GPU wafer tax I believe) (the die enabled is only 88.89% of the full available)
(24GB 21Gb/s (VRAM)) costs about 95-120 USD for all 24GB (roughly 4-5 USD per 1GB of GDDR6X VRAM, the modules are 2GB each, 12 total but that doesn't impact price by over 5% at best)
The RTX 4090 and it's PCB costs around 20-25 USD at best? (just so you guys know the geforce GTX 285 had a pcb cost of 7.37 USD and the GTX 580 had a pcb cost of 7.29 USD)
The cooler probably costs around 80-100 USD to make if not less (raw metal costs barely account for the total price, it's mainly the machining and painting (this goes for any air cooler btw))
395-465 USD for the entire thing excluding software R&D.
This accounts for FE coolers only (which NVIDIA on THEIR END has to make)
I'm pretty sure AIB coolers are as cheap as they can get when it comes to machining as they have to do this every generation of GPU's


Now we have the RTX 3090 price breakdown (these prices are grabbed from the half point of the cards launch till new generation of cards launch or present/averaged in its manufacturing peak)
(GA102 die) costed around 62.5-67.5 USD per die (roughly 5.5K-6K USD per wafer including the GPU wafer tax I believe) (the die enabled is 97.62% of the full available)
(24GB 19.5Gb/s VRAM) costed around 385-430 USD for all 24GB (roughly 16-18 USD per 1GB of GDDR6X VRAM, the modules are 1GB each, 24 total but this also doesn't impact price by over 5% at best)
The RTX 3090 and it's PCB costed around 25-30 USD at best (the reason why it's more was because it's design was far different from the previous generation of GPU's)
The cooler probably costed around 120-135 USD at best to make (once again it's more expensive over 40 series because it was far different over the previous generation of GPU's and a far different design overall)
592.5-662.5 USD for the entire thing excluding software R&D. (once again this is for the FE's only)
The RTX 3080 10GB was around 355-400 USD to make
Please don't ask for sources, if you want sources find them yourself as I don't get paid for this, nor should I.

The pricing for 40 series across the board sucks, it gets worse after looking at this

quote “4090…395-465 USD for the entire thing excluding software R&D.” wow, exactly the nvidia reported gross margin of 70%. (70% GM is 1120, 1120+470 the MSRPof the 4090 1600us.

(Apple has a GM of 45% and AMD a GM of 47%) which means a 7900xtx costs the same as a 4090 to make…
also Nvidia could have sold the 4080 for the same price, if they had the same GM as AMD… (BTW, i wonder if a would have bought a 4080, instead of a 7900xtx if it was the same price as a 7900xtx)

redeyequote “4090…395-465 USD for the entire thing excluding software R&D.” wow, exactly the nvidia reported gross margin of 70%. (70% GM is 1120, 1120+470 the MSRPof the 4090 1600us.

(Apple has a GM of 45% and AMD a GM of 47%) which means a 7900xtx costs the same as a 4090 to make…
also Nvidia could have sold the 4080 for the same price, if they had the same GM as AMD… (BTW, i wonder if a would have bought a 4080, instead of a 7900xtx if it was the same price as a 7900xtx)

Calculating actual RTX 4090 cost from gross margin is useless since their margins vary a lot from a product to another, relatively speaking they don't generate much profit from their GeForce cards vs their other products.

RTX 4080 Super 16GB (AD103 Unlocked) with over 10,000 Cuda Cores Super Binned Refresh Hardware.

Bring on DLSS 3.5 Gaming!
$999. MRSP... What's MSI best again?

Looking forward for this upgrade, RTX 2080 NVlink setup to RTX 4080 Super 16GB good 5 years upgrade.

Cheers

ToxicTaZRTX 4080 Super 16GB (AD103 Unlocked) with over 10,000 Cuda Cores Super Binned Refresh Hardware.

Bring on DLSS 3.5 Gaming!
$999. MRSP... What's MSI best again?

Looking forward for this upgrade, RTX 2080 NVlink setup to RTX 4080 Super 16GB good 5 years upgrade.

Cheers

what do you need dlss for if you have a 80 series with 16GB super ram? native res plus taa or dlaa both with sharpen tweaks. dlss fake resoultion, you are actually loosing visible information because the ai is guessing what the frame is. You know that card is really powerful right? You dont need the gimmick stuff on. hell i have 3060 12gb and i dont use dlss or upscaling at all. simply tweak the other graphics features of the game and you will find that the fidelity you can acheive with eliminatoing certian useless features LIKE CHROMATIC ABERAATION and other tweaks

1440 gsync lg 240hz gigabyte eagle oc 3060 5600x x570 32gb 3600 win 11 23h2

Poochwhat do you need dlss for if you have a 80 series with 16GB super ram? native res plus taa or dlaa both with sharpen tweaks. dlss fake resoultion, you are actually loosing visible information because the ai is guessing what the frame is. You know that card is really powerful right? You dont need the gimmick stuff on. hell i have 3060 12gb and i dont use dlss or upscaling at all. simply tweak the other graphics features of the game and you will find that the fidelity you can acheive with eliminatoing certian useless features LIKE CHROMATIC ABERAATION and other tweaks

1440 gsync lg 240hz gigabyte eagle oc 3060 5600x x570 32gb 3600 win 11 23h2

Do you turn down your graphics to run your games properly?.... to me if you have to turn down your graphics means you need a new video card. Just my opinion of performance.

You're using an lower resolution 1440p wear I'm using MSI MEG G-Sync Ultimate 1600p (3840x1600) 144Hz / 175Hz OC. Much harder to run anything above 1440p.

Try running any AAA Games 4K+ Ultra/Extreme mode levels with 144fps solid performance. DLSS 3.5 is the only way to get the job done if you didn't have an AD102 GPU.

Unfortunately AI is the future and what defines Nvidia DLSS 3.5 from AMD FSR3.

Cheers

ToxicTaZRTX 4080 Super 16GB (AD103 Unlocked) with over 10,000 Cuda Cores Super Binned Refresh Hardware.

Bring on DLSS 3.5 Gaming!
$999. MRSP... What's MSI best again?

Looking forward for this upgrade, RTX 2080 NVlink setup to RTX 4080 Super 16GB good 5 years upgrade.

Cheers

Hi man,

Can I ask for your feedback on the NVLink experience? Did you get anything worthy in video games in general? Almost did it with a dual 2070 SUPER setup and always wondered what it would achieve.

DahitaHi man,

Can I ask for your feedback on the NVLink experience? Did you get anything worthy in video games in general? Almost did it with a dual 2070 SUPER setup and always wondered what it would achieve.

RTX 2080 8GBx2 NVlink setup was good as it lasted! Was backwards compatibility with all SLI and no conflicts with Hardware G-Sync & Ultimate. The problems were No support, Nvidia did not push NVlink consumer levels NVlink. I had to pay $100. Just for the NVlink device, another thing is cooling, 2.5 Slots video cards are basically touching my opinion temps of card one around 90c and the second card around 70c.

Now moving into the AD103-400 Unlocked Refresh... RTX 4080 Super 16GB probably MSI. Going on my new MSI MEG GODLIKE MAX motherboard. Should be good for awhile...

Cheers

DahitaHi man,

Can I ask for your feedback on the NVLink experience? Did you get anything worthy in video games in general? Almost did it with a dual 2070 SUPER setup and always wondered what it would achieve.

RTX 2080 NVlink setup worked well around 50% fps more not doubled two cards level. NVlink was backwards compatibility with all SLI supported.

RTX 4080 Super 16GB (AD103-400 Unlocked) Refresh running DLSS 3.5 is the level I'm buying into.

Cheers