Thursday, September 5th 2024

NVIDIA GeForce RTX 5090 and RTX 5080 Reach Final Stages This Month, Chinese "D" Variant Arrives for Both SKUs

by
AleksandarK
 Discuss (87 Comments)
NVIDIA is on the brink of finalizing its next-generation "Blackwell" graphics cards, the GeForce RTX 5090 and RTX 5080. Sources close to BenchLife indicate that NVIDIA is targeting September for the official design specification finalization of both models. This timeline hints at a possible unveiling at CES 2025, with a market release shortly after. The RTX 5090 is rumored to boast a staggering 550 W TGP, a significant 22% increase from its predecessor, the RTX 4090. Meanwhile, the RTX 5080 is expected to draw 350 W, a more modest 9.3% bump from the current RTX 4080. Interestingly, NVIDIA appears to be developing "D" variants for both cards, which are likely tailored for the Chinese market to comply with export regulations.

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, these details should be taken with a grain of salt.
Sources: BenchLife, via Wccftech

Related News

Add your own comment

87 Comments on NVIDIA GeForce RTX 5090 and RTX 5080 Reach Final Stages This Month, Chinese "D" Variant Arrives for Both SKUs

#26
Dr. Dro
Shou MikoI also seen on X that RTX4090 D og RTX 4080 variant with double the amount of memory.

I didn't think Nvidia allowed board-partners to do, but I do not remember if it was a "custom" chinese job.
4090 D is AD102 base, it's the same thing as the 4090 but with additional cores disabled. The 4080 and 4080 Super were not reduced in specification for the Chinese market as they do not achieve nor exceed the "performance score" required for export controls defined by the US Government to kick in
closeYou need to know both prices before determining which one is worth it.
Well, I'm assuming they are going to target the same price brackets as Ada, which are as high as the market will bear.
Posted on Reply
#27
AusWolf
Dr. Dro4090 D is AD102 base, it's the same thing as the 4090 but with additional cores disabled. The 4080 and 4080 Super were not reduced in specification for the Chinese market as they do not achieve nor exceed the "performance score" required for export controls defined by the US Government to kick in
Thinking about this RTX 5000 D variant is weird. Basically, the D version only exists to comply with the performance limit of US export controls. So if the 5080 is too fast, they need a watered-down D version for China. But then, what is the 5090 D all about? If the 5080 was already too fast for export, and you had to make a 5080 D, then how can you make an even faster 5090 D? :wtf: Unless both the 5080 D and 5090 D are below 5080 normal performance level, which I highly doubt based on the core count differences.
Posted on Reply
#28
Dr. Dro
AusWolfThinking about this RTX 5000 D variant is weird. Basically, the D version only exists to comply with the performance limit of US export controls. So if the 5080 is too fast, they need a watered-down D version for China. But then, what is the 5090 D all about? If the 5080 was already too fast for export, and you had to make a 5080 D, then how can you make an even faster 5090 D? :wtf: Unless the 5090 D is below 5080 normal performance level, which I highly doubt based on the core count differences.
Yeah, I thought the same. My best guess is that the reasoning behind this is roughly the same as AMD's when they released the 2048SP version of the RX 580, that being that the Chinese market is receptive to alternate configurations to ensure availability, and they might be able to increase production output by releasing an inferior spec variant at the roughly same price in that region, as they will require less high-grade chips that qualify as a full 5080/5090 to build the same amount of cards, maximizing profits.

Sure sucks to be in China if you want bleeding edge graphics hardware, that's for sure.
Posted on Reply
#29
AusWolf
Dr. DroYeah, I thought the same. My best guess is that the reasoning behind this is roughly the same as AMD's when they released the 2048SP version of the RX 580, that being that the Chinese market is receptive to alternate configurations to ensure availability, and they might be able to increase production output by releasing an inferior spec variant at the roughly same price in that region, as they will require less high-grade chips that qualify as a full 5080/5090 to build the same amount of cards, maximizing profits.

Sure sucks to be in China if you want bleeding edge graphics hardware, that's for sure.
True that, but I mean, if the 10k cores in the 5080 are too much for export, and you need to cut it down to a 5080 D with less, then how much will you have to cut the 5090 down with 24k cores? You're gonna be using less than 33% of the chip at this point (unless you slap a 5060 GPU on it instead, and call that a 5090 D).
Posted on Reply
#30
Dr. Dro
AusWolfTrue that, but I mean, if the 10k cores in the 5080 are too much for export, and you need to cut it down to a 5080 D with less, then how much will you have to cut the 5090 down with 24k cores? You're gonna be using less than 33% of the chip at this point (unless you slap a 5060 GPU on it instead, and call that a 5090 D).
The 5090 D might be for export controls, the 5080 D might be just for yield and marketability reasons. In this scenario a full 5080 wouldn't be banned from export, but also unlikely to be released in China, leaving just the pair of D cards in the market in the upper range, and resuming as usual with the 5070 Ti and below.
Posted on Reply
#31
AusWolf
Dr. DroThe 5090 D might be for export controls, the 5080 D might be just for yield and marketability reasons.
Possibly. We'll see when they come out, I guess.
Posted on Reply
#32
N/A
The metrics subject to export limitations are based on AI performance. roughly 456 tensor cores.
Posted on Reply
#33
Bwaze
Could it be that Nvidia will (try to) limit the AI performance of Chinese export models with drivers, firmware, and leave gaming, CUDA etc. performance largely intact - just like they tried with “LHR” editions of anti-cryptomining RTX 30x0 cards?

That would indeed enable Nvidia to sell RTX 5080 and 5090 "D"cards.

And what is more, it would also enable them to "accidentally leak" firmware, drivers that enable the full performance of gimped cards, just like they did with "LHR" ones, and noone could ever find the culprit!
Posted on Reply
#34
Dr. Dro
BwazeCould it be that Nvidia will (try to) limit the AI performance of Chinese export models with drivers, firmware, and leave gaming, CUDA etc. performance largely intact - just like they tried with “LHR” editions of anti-cryptomining RTX 30x0 cards?

That would indeed enable Nvidia to sell RTX 5080 and 5090 "D"cards.

And what is more, it would also enable them to "accidentally leak" firmware, drivers that enable the full performance of gimped cards, just like they did with "LHR" ones, and noone could ever find the culprit!
Probably won't be an acceptable solution to regulatory agencies especially considering how LHR ultimately failed to stop miners and increased speculation pressure on the GPU market.

People using these for AI will have the know how to bypass any such lock, defeating the purpose of export controls. Hardware must be physically unable to do it
Posted on Reply
#35
Gucky
FoulOnWhitePeople who buy these don't care about efficiency, they care about having the best and gaming at the highest settings they can. I don't get people buying a 4090 and undervolting it, you bought the best GPU nvidia made and strangle it so it uses less power.
Do you use air conditioning? Such a statement only comes from those who use AC.
I only have a gas powered hot water radiator as a heater, which is normal in germany. We usually don't have AC.
That heater in my room (12m²/30m³) has a ~550W output. Now I ask you, would you use a heater in the summer? A stock 4090 with CPU and monitors are actually outputting MORE heat then my heater.
Of course I want the maximum amount of FPS with highest graphic, but using a (PC) heater in the summer grants you >32°C in the room.
Every 100W from the PC is about +1°C in my room with an half opened window while it is colder outside.
Tl;dr Undervolting my PC reduces the roomtemp by 1.5-2.5°C.
Posted on Reply
#36
Bwaze
Dr. DroProbably won't be an acceptable solution to regulatory agencies especially considering how LHR ultimately failed to stop miners and increased speculation pressure on the GPU market.

People using these for AI will have the know how to bypass any such lock, defeating the purpose of export controls. Hardware must be physically unable to do it
But regulatory agencies won't be looking at the success of LHR editions - because they weren't released to satisfy any regulations, they were released to paint Nvidia as if it's doing something to satisfy gamers in a market where they couldn't compete with people who planned to profit from running GPUs. PR stunt from the beginning.

And we can all laugh and point to Nvidia for breaking the limitations themselves, but officially and legally I'm sure they weren't guilty. So if they deliver the cards that show on official tests performance below regulated, they have to allow the product to be sold?
Posted on Reply
#37
lexluthermiester
AleksandarKThe RTX 5090 is rumored to boast a staggering 550 W TGP, a significant 22% increase from its predecessor, the RTX 4090. Meanwhile, the RTX 5080 is expected to draw 350 W
Good fricken grief NVidia! Seriously with all this wattage?!? Hard core pass! If you can't make cards that perform better than the previous generation that also use less power, you're doing something wrong. VERY wrong.
Posted on Reply
#38
kondamin
Dr. DroSure sucks to be in China if you want bleeding edge graphics hardware, that's for sure.
there is probably a thriving grey market, governments aren't capable of stopping drugs/weapon/human trafficking and that's something they all agree on is a bad thing to do.
Posted on Reply
#39
Minus Infinity
5090 is using 448 bit bus, 512 would be full phat Titan card.,
closeYou need to know both prices before determining which one is worth it.
5080 25% the price of the 5090 ROFLMAO. Try 5080 $1200, 5090 $2000. 5070 won't even be 25% the price.
Posted on Reply
#40
Bwaze
Minus Infinity5090 is using 448 bit bus, 512 would be full phat Titan card.,


5080 25% the price of the 5090 ROFLMAO. Try 5080 $1200, 5090 $2000. 5070 won't even be 25% the price.
RTX 4080 is $1200. Even without AI craze Jensen told us there is no More's Law any more, any increase in performance will bring increase in price.

You can take this as a rough guide:

$1200 + 50% = $1800, and that doesn't even cover the inflation!
Posted on Reply
#41
Dr. Dro
kondaminthere is probably a thriving grey market, governments aren't capable of stopping drugs/weapon/human trafficking and that's something they all agree on is a bad thing to do.
Oh, no doubt there is a market. But they likely cater exclusively to enterprise gear and the AI market, not gamers.
Posted on Reply
#42
Pumper
AusWolfThinking about this RTX 5000 D variant is weird. Basically, the D version only exists to comply with the performance limit of US export controls. So if the 5080 is too fast, they need a watered-down D version for China. But then, what is the 5090 D all about? If the 5080 was already too fast for export, and you had to make a 5080 D, then how can you make an even faster 5090 D? :wtf: Unless both the 5080 D and 5090 D are below 5080 normal performance level, which I highly doubt based on the core count differences.
I'm not sure, but aren't the restriction mostly about AI performance? So the 5090D can still be faster than a 5080 in gaming, but slower than 5080 in AI computation.
Posted on Reply
#43
Dr. Dro
PumperI'm not sure, but aren't the restriction mostly about AI performance? So the 5090D can still be faster than a 5080 in gaming, but slower than 5080 in AI computation.
No, it cannot be done. Tensor cores (Matrix multiplication units) are inextricably linked to their associated portion of the processor (SM unit). Disabling them will inevitably reduce the amount of shaders, ray acceleration units and associated cache.
Posted on Reply
#44
phints
A lot of clueless comments here. My 1.5 year old 4070 is still top tier performance/watt and when it came out it was unparalleled (in fact 4070S didn't improve this their TDP went up with the +15% performance). Slightly UV'd it's a 160W GPU that performs awesome at 1440p. This is one area no one can fault Nvidia. When it comes to pricing and VRAM sure you can fault them, but their efficiency has been excellent.

The main reason why 3000 -> 4000 is so much more efficient is they went from Samsung 8nm to TSMC 5nm. The jump from layout/feature optimization will not be as big this time if they only go to TSMC 4nm and GDDR7. So yea efficiency will improve but to see the typical +30-40% gain TDP will have to go up too.

Mostly these rumors are BS just like they were last time and should be flagged as nonsense but sites gotta post their clickbate.
Posted on Reply
#45
GhostRyder
The gap clearly signifies a 5080 ti is being held back for later. Either that or the scaling up of more cores does not yield a huge performance uplift. The problem with TDP of these cards is going to be cooling requirements more than anything. I don't see the 5090 being fun to use without liquid cooling even with massive air coolers (At least from my perspective).

I feel like something is missing though, I find it hard to believe these are truly a year out.
Posted on Reply
#46
Onasi
GhostRyderI feel like something is missing though, I find it hard to believe these are truly a year out.
Half a year. Closer to a third, really. CES is early January. And this is in line with what we’ve heard about 5000 series being planned for early 2025. It makes some amount of sense - NV doesn’t have real competition to rush them along, the 4000 series still sells well and getting rid of existing inventory before launching new cards is prudent and for now NV is probably more keen on using fab allocation on enterprise Blackwell accelerators to fully capitalize on the demand.
Posted on Reply
#47
neatfeatguy
Only thing I hope doesn't happen is that people get screwed by Nvidia again for melting power connectors. We saw enough burning 4090s, let's just hope they don't screw things up again with the 5090.
Posted on Reply
#48
nguyen
Oh boy so 4090 will remain king of the jungle for more than 24 months, previous best record was 2080 Ti at 24 months.

Grabbing the 5090 will get me set for 2+ years :cool:
Posted on Reply
#49
Vayra86
dgianstefaniThis is categorically false.




As you can see, (much) slower GPUs often use (much) more power.
Reading comprehension skills need a serious update with you. Re read my posts pls. The point has eluded you clearly - despite me repeating it 3 times.
Posted on Reply
#50
the54thvoid
Super Intoxicated Moderator
nguyenOh boy so 4090 will remain king of the jungle for more than 24 months, previous best record was 2080 Ti at 24 months.

Grabbing the 5090 will get me set for 2+ years :cool:
My 2080ti had way more class than your 4090. :D

(I'm sure you also had a 2080ti).
Posted on Reply
Add your own comment
Sep 10th, 2024 17:04 EDT change timezone

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts