NVIDIA GeForce RTX 5090 and RTX 5080 Reach Final Stages This Month

[erek]

“… these details should be taken with a grain of salt.

Regarding raw power, the RTX 5090 is speculated to feature 24,576 CUDA cores paired with 512-bit GDDR7 memory. The RTX 5080, while less mighty, is still expected to pack a punch with 10,752 CUDA cores and 256-bit GDDR7 memory. As NVIDIA prepares to launch these powerhouses, rumors suggest the RTX 4090D may be discontinued by December 2024, paving the way for its successor. We are curious to see how the power consumption is handled and if these cards are packed efficiently within the higher power envelope. Some rumors indicate that the RTX 5090 could reach 600 watts at its peak, while RTX 5080 reaches 400 watts. However, that is just a rumor for now. As always, until NVIDIA makes an official announcement,”

Source: https://www.techpowerup.com/326292/...month-chinese-d-variant-arrives-for-both-skus

 

[d3athf1sh]

dang, so 5080 is going to be like 1/2 the performance of 5090? that's a big gap. i guess eventually there will be a 5080ti but that memory bus makes a huge difference in memory bandwidth and performance so wonder which the Ti will get? even if somehow it gets a 384 bit bus or if it comes with the 512 one, it may be the one hold out for if you can manage to get by till then?

 

[Shoganai]

dang, so 5080 is going to be like 1/2 the performance of 5090? that's a big gap. i guess eventually there will be a 5080ti but that memory bus makes a huge difference in memory bandwidth and performance so wonder which the Ti will get? even if somehow it gets a 384 bit bus or if it comes with the 512 one, it may be the one hold out for if you can manage to get by till then?

Probably half the price too.

 

[Dudhunter]

5090 sounds a hell of a lot like the older titan models. $3299 starting price is my guess.

 

[Gideon]

5090 sounds a hell of a lot like the older titan models. $3299 starting price is my guess.

I expect it to be more like vaporware, technically it exists but as long as they are selling AI hardware like hotcakes I expect them to provide minimal quantity of the 5090.

 

[LukeTbk]

Go from 600->550 and 400->350 in a day in the interweb, without any surprise.... We will have read every possible scenario before launch and someone will be right (and wrong)

There was indeed a 400W NVIDIA cooling module case earlier, but it is currently in the cancellation or suspension stage; as for the 600W cooling module case, from the beginning to Now, it has not stopped, and the module manufacturer has also confirmed that NVIDIA currently has 5 GeForce RTX graphics card cases in progress.

That could be: NVIDIA appears to be developing "D" variants for both cards, which are likely tailored for the Chinese market to comply with export regulations.

A really great news in a way.

dang, so 5080 is going to be like 1/2 the performance of 5090?

AD102: 18432
AD103: 10240

4090 was about 25%-30% stronger with 80% more cores (that could be raw core count and not the enable one and information that maybe even Nvidia has yet to fully decide themselve), the 4090 was significantly more cut-down than the 4080 was, if the demand and price for the L40, RTX 6000 replacement stay really high that could still be the case, like the 600w-400w being the power delivery-cooling capacity and not actual use, core count info would probably be similar.

What would be an interesting progression.
3080-3090, very similar performance, bought for future vram and halo status if you buy it for game, giant MSRP gap, performance per dollar was often twice for the 3080, that MSRP was impossible to actually work in pratice.

They adjust too much
4080-4090, performance gap is created, but the 4080 fall in a limbo, price is too close despite a reasonable fps / $ versus the 4090, $1200 non halo card is limited interest.

Readjust, make a cheaper 5080 but an even larger performance gap to keep the much more expensive 5090 interesting.

Could be a $1000 vs $1800 msrp launch scenario.

 

[LukeTbk]

I expect it to be more like vaporware, technically it exists but as long as they are selling AI hardware like hotcakes I expect them to provide minimal quantity of the 5090.

If they follow the ad102 model with it (the core count gap with the xx80 if true seem like they are), the gaming version will be a significantly cut down version of those AI hardware.

For lovelace the L40-RTX6000 had 98.6% of the core enabled, for a 600mm+ chips that should create a lot of dies that missed that mark, thus the probably over 1 million of 4090 (88% of the die)-4090D(79%) they made and sold (1% of all steam users according to the hardware survey that would 1.1-1.4 millions cards, it must be overcounted, enthusiast proud to fill the survey more than the random users.... but a lot of those cards are not in gaming PC with a steam account)

This will not go in laptop, probably not go into 5080TI, with the price they charge for those L40-RTX 6000 card could still go almost full die, if so, if they make 1 million+ of xx90 again it would not be surprising. This a great business for passed down "bad" dies and not a bad way for the 2028 college kids to learn your cuda environment if they can buy 32 GB of vram 5090 at a good price, like the current one can do with 3090s)