NVIDIA Teases GeForce RTX 40-series SUPER for CES Unveiling

[btarunr]

The Twitter handle of NVIDIA GeForce just updated its header picture, with that seems to us like the silhouette of its upcoming GeForce RTX 40-series SUPER graphics card. NVIDIA is expected to debut three new high-end graphics card SKUs at its CES event on January 8, with reviews and product availability of the three expected to go live in the following weeks. The lineup includes the RTX 4070 SUPER, the RTX 4070 Ti SUPER, and the RTX 4080 SUPER. Market availability of the three, according to older reports, pin them at January 17 for the RTX 4070 SUPER, January 24 for the RTX 4070 Ti SUPER, and January 31 for the RTX 4080 SUPER.



View at TechPowerUp Main Site | Source

 

[Solaris17]

h-hurray

I am personally not impressed, and I dont look forward too the super release at all. The price to performance ratio is already a joke because the prices are way too high, and thats saying something coming from someone with a 4090. Dont indulge this crap. AMD and Intel are much more interesting in this distopian GPU landscape.

 

[gmn17]

Any chance of a 4090 Super release?

 

[Crackong]

I am so excited (poker face)

 

[Chaitanya]

nVidia's marketting team to audience today:

 

[Hyderz]

yay new gpus! great for those who skipped the vanilla 40 series, 30series and even 20 series... , and hopefully these provide more power, efficiency
if its gonna be more powerful than amd cards, i doubt its gonna stay at the same price, the price will increase... thus making them unappealing

 

[HisDivineOrder]

Hooray

 

[Hxx]

You just know they won’t be worth the asking price.

 

[enb141]

Well, this time at least the "low end" will be released earlier, and scaling up it release date per one week.

Now the question is if Nvidia will release their Vanilla cards with this variants.

 

[atomek]

Did they changed power connector or there is still risk of failure of high-end RTX?

 

[wNotyarD]

Did they changed power connector or there is still risk of failure of high-end RTX?

They should use the revised 12V2X6 connector, but were there 4080's melting or was it exclusive to the 4090's?

 

[pavle]

New top shyte!

 

[Gameslove]

RTX 4080 Super with 16 GB VRAM 256 bit will big disappointment , it must have double - 32 GB VRAM 512 bit , you can imagine...if NVIDIA release this card with 32 GB....

 

[ratirt]

RTX 4080 Super with 16 GB VRAM 256 bit will big disappointment , it must have double - 32 GB VRAM 512 bit , you can imagine...if NVIDIA release this card with 32 GB....

that would have never happened. 32GB on a 4080 super with higher mem bandwidth vs 24Gb on a 4090 with only 384-bit bus? It would have cost more than a 4090 and the price for that one is ridiculous already.
You should probably check your expectations about the supers'.

 

[pavle]

RTX 4080 Super with 16 GB VRAM 256 bit will big disappointment , it must have double - 32 GB VRAM 512 bit , you can imagine...if NVIDIA release this card with 32 GB....

Indeed, but then their precious Quadro RTX 5000 Adding machine wouldn't look so lucrative anymore and that would hurt poor nvidiots' pockets.

 

[Dr. Dro]

Any chance of a 4090 Super release?

Extremely unlikely considered the use of AD102 in more profitable markets and the fact that for the first time ever a cutdown consumer-grade GPU has attracted the attention of government regulators... in fact this may even affect the next-gen Blackwell processor's market performance in the short-term future.

RTX 4080 Super with 16 GB VRAM 256 bit will big disappointment , it must have double - 32 GB VRAM 512 bit , you can imagine...if NVIDIA release this card with 32 GB....

512-bit GPUs are probably never happening again, especially not at the high-end segment. Ultra-high-end/halo segment will likely stick to 384-bit for the foreseeable future as VRAM IC technology advances in data rate and it has a high but relatively acceptable PCB development and manufacture cost. High-end will likely remain 256-bit while the performance segment will be between 128 and 192-bit, like the 4070 Ti nowadays.

4070 Ti SUPER will use AD103 with cutdown configuration (currently presumed to be 66 out of 80 SMs) because 4070 Ti already maxes the AD104 silicon, which means it should carry the 4080's memory bus but with far fewer cores enabled. On the other hand, 4080 SUPER will not have a lot of extra resources over 4080 because the 4080 is already almost maxed out, with 76 out of 80 SMs present. This will be a similar situation when comparing the original RTX 2080 to the 2080 SUPER - they'll be 5% apart.

 

[wNotyarD]

4070 Ti SUPER will use AD103 with cutdown configuration (currently presumed to be 66 out of 80 SMs) because 4070 Ti already maxes the AD104 silicon, which means it should carry the 4080's memory bus but with far fewer cores enabled. On the other hand, 4080 SUPER will not have a lot of extra resources over 4080 because the 4080 is already almost maxed out, with 76 out of 80 SMs present. This will be a similar situation when comparing the original RTX 2080 to the 2080 SUPER - they'll be 5% apart.

Only other option would be if the 4080S was in fact a 4080Ti, using a further cut AD102 but still with margins over a fully-enabled AD103.

 

[R0H1T]

512-bit GPUs are probably never happening again, especially not at the high-end segment.

Yes & that's because they're not needed anymore ~

AMD’s CDNA 3 Compute Architecture

AMD has a long history of vying for GPU compute market share.

chipsandcheese.com

Infinity caches FTW

 

[Dr. Dro]

Only other option would be if the 4080S was in fact a 4080Ti, using a further cut AD102 but still with margins over a fully-enabled AD103.

International release of 4090D perhaps, but at that point we're talking a die worse than a GTX 465. That'd be a disaster card as far as efficiency goes.

Yes & that's because they're not needed anymore ~

AMD’s CDNA 3 Compute Architecture

AMD has a long history of vying for GPU compute market share.

chipsandcheese.com

Infinity caches FTW

I mean I wouldn't use CDNA as the justification for that (as it's not a traditional GPU but a pure compute accelerator and since it uses large amounts of HBM instead), but you're right. Last-level caches and higher bandwidth VRAM ICs will make up for the bus width quite effectively.