Monday, January 6th 2025
First NVIDIA GeForce RTX 5090 GPU with 32 GB GDDR7 Memory Leaks Ahead of CES Keynote
NVIDIA's unannounced GeForce RTX 5090 graphics card has leaked, confirming key specifications of the next-generation GPU. Thanks to exclusive information from VideoCardz, we can see the packaging of Inno3D's RTX 5090 iChill X3 model, which confirms that the graphics card will feature 32 GB of GDDR7 memory. The leaked materials show that Inno3D's variant will use a 3.5-slot cooling system, suggesting significant cooling requirements for the flagship card. According to earlier leaks, the RTX 5090 will be based on the GB202 GPU and include 21,760 CUDA cores. The card's memory system is a significant upgrade, with its 32 GB of GDDR7 memory running on a 512-bit memory bus at 28 Gbps, capable of delivering nearly 1.8 TB/s of bandwidth. This represents twice the memory capacity of the upcoming RTX 5080, which is expected to ship with 16 GB capacity but 30 Gbps GDDR7 modules.
Power consumption has increased significantly, with the RTX 5090's TDP rated at 575 W and TGP of 600 W, marking a 125-watt increase over the previous RTX 4090 in raw TDP. NVIDIA is scheduled to hold its CES keynote today at 06:30 pm PT time, where the company is expected to announce several new graphics cards officially. The lineup should include the RTX 5090, RTX 5080, RTX 5070 Ti, RTX 5070, and an RTX 5090D model specifically for the Chinese market. Early indications are that the RTX 5080 will be the first card to reach consumers, with a planned release date of January 21st. Release dates for other models, including the flagship RTX 5090, have not yet been confirmed. The RTX 5090 is currently the only card in the RTX 50 series planned to use the GB202 GPU. Pricing information and additional specifications are expected to be revealed during the upcoming announcement.
Source:
VideoCardz
Power consumption has increased significantly, with the RTX 5090's TDP rated at 575 W and TGP of 600 W, marking a 125-watt increase over the previous RTX 4090 in raw TDP. NVIDIA is scheduled to hold its CES keynote today at 06:30 pm PT time, where the company is expected to announce several new graphics cards officially. The lineup should include the RTX 5090, RTX 5080, RTX 5070 Ti, RTX 5070, and an RTX 5090D model specifically for the Chinese market. Early indications are that the RTX 5080 will be the first card to reach consumers, with a planned release date of January 21st. Release dates for other models, including the flagship RTX 5090, have not yet been confirmed. The RTX 5090 is currently the only card in the RTX 50 series planned to use the GB202 GPU. Pricing information and additional specifications are expected to be revealed during the upcoming announcement.
82 Comments on First NVIDIA GeForce RTX 5090 GPU with 32 GB GDDR7 Memory Leaks Ahead of CES Keynote
Actually properly optimizing lighting is hard, optimizing in general is hard, that's why we've seen devs just not doing it. the same devs that keep cutting features and delivering half baked ideas (unless its the cash shop).
EDIT: as a prime example. cities skylines 2. Every sim/zim/whatever they call them had fully modeled individual 3d teeth, with multiple surfaces. WHY? ITS A CITY BUILDER! Dumb stuff like that is absolutely everywhere. I'd be more curious at something like the new DOOM using RT on idtech to see what is actually possible, but with carmack gone, who knows if its still got the same level of optimization.
Even to myself.
Coals, oil and natural gas are the ones you should look at as bad energy sources.
that's why nuclear power is often in its own category.
France is the best case example for nuclear power and how we could have fixed our many issues decades ago, if people didnt panic over the word "radiation".
It can get lumped in the 0% CO2 bracket (in terms of operating emissions), but that's it... it hasn't really got any 'green' credentials in terms of environmental impact.
In seriousness, nuke plants dont kill birds like windmills do, nor do they require the huge clearing of large solar farms, nor do they bake the soil like solar can do. The production of solar still produces lots of toxic by products which get dumped into rivers and the fluid from turbines still has no recourse once used up and easily pollutes water tables.
Back to the topic, you will need your own mini nuke for your 5090. I'm all for it, where do I get my complementary Chirkov sticker?
Even at $850 you still only get a 4070Ti Super. No thanks. That still isn't a top end GPU. Sure, it gets you close, but isn't top end. 4080 would be low-high end where as the 4090 would be top high-end. 4070 would be low-mid to the Ti Super being high-mid.
So if you want to use inflation back when the 980Ti was $650. The 970 was a mid tier card back then and sold for $330. With inflation for today's price you would be spending $445. What does $445 buy you for a GPU in this current gen? A 4060 Ti.....so you now went from a mid range card in 2015 at $330 to an entry level card (let's be honest, anything under a 4060 isn't really a good buy for a GPU at these prices) to an entry level card for $445 because you can't get a mid-ranged RTX 4070 since it is $600.
You want the latest x nm process tech... you's got to pay...
It's quite amusing that Apple (who have a history of playing hardball with suppliers on costs) are always so willing to pay the premium charged by TSMC for it...
There's other things to look at as well. Maxwell was built on a cheaper node. The 28nm TSMC node cost about $5000 back in 2014. Ada's node was around $20,000. so that's 4x the silicon costs. GDDR6/X memory is also significantly more expensive than GDDR5 was, the coolers are bigger (and quieter), ece. So yeah, the 4070 is about 104mm2 smaller than the 970, but it costs likely twice to three times as much to actually make then maxwell did. Do you expect nvidia to eat a loss there, so you can buy one for $330? *sigh*
Nvidia's margins during the height of maxwell: 55-57%
Nvidia's margins during peak COVID boom (on the much cheaper samsung 8nm node): 65%
Nvidia's margins during the height of Ada: 56%
Can we let this misinformation die already? I know it's easy to just blame everything on NgReEdIa but the reality is there's more to the world then Nvidia. The onyl reason their margins are so high right now is the AI GPU boom. If they were still primarily Geforce they'd still be about the same margin level.
and i'm not sure you are making complete sense, the covid had higher margins, now the margin are the same as in the maxwell era but you claim margins are "so high right now because of AI" :wtf:
In the graph I linked below, you will find that Nvidia's margins in april of 2023 were hitting roughly the same level they were hitting in late 2014. Shortly after that, the AI market began its boom cycle, and their margins shot up to 75%, where they sit today. They are selling a LOT of $40,000 AI cellerators.
But on purely Geforce sales, no, Ada's price was not "nvidia inflation". It was regular old inflation, courtesy of our glorious leaders printing more money than they spent FIGHTING WWII to react to a nasty cold, driving up the costs of everything to insane levels. No, I meant margins. I've posted this source many times before
www.macrotrends.net/stocks/charts/NVDA/nvidia/gross-margin
They get their info from Nvidia's quarterly statements.
There's something else interesting in there: During the days of $500 fermi flagships AKA the "good old days", nvidia's margin was only 31% during the 400s series and 38% during the 500 series. Those are closer to what AMD was doing before recently, and we saw how well that turned out.
and revenue in the segment we care about barely moved since the covid/crypto for the gaming segment, so it's clear they are selling a lot less and for a lot more money (product margin) to almost make the same overall
stockanalysis.com/stocks/nvda/metrics/revenue-by-segment/
you're assumptions are incorrect
My assumptions are sound.
4060s are low tier.
4070s are mid tier (4070 low-mid, 4070 Super and 4070Ti aren't far apart for mid and 4070Ti Super for top-mid).
4080s are high tier.
4090 is top end.
so nvidia inflation…
Then you still have the problem of scaling your energy up or down. which is far more difficult with both options than with a nuclear plant. If you want the world to really go green then nuclear is an unavoidable power source at this time unless fusion power makes a breakthrough.
The problem with nuclear plants is that all people see is Chernobyl. A badly designed nuclear plant from the start and then they forget it happened because of bad management/operational error. At the same time it was somewhat a blessing in disguise because the world learned so enormously much from what not to do and it's implications of when it goes horribly wrong.
@TheinsanegamerN
Thanks for the link to that margins chart, haven’t seen it before. Mighty curious information. I think that your take that the current elevated margins are caused by the AI demand outstripping supply is more than fair then.
What this originally stems from was the whole nuclear being lumped into the same bracket as renewables, primarily because it's 'carbon neutral' - that's really a mistake and I'm unsure if it's deliberate by the energy industry to try and lift it's PR rep with the public.
Chernobyl is an outlier - the Fukushima problems are actually more worrying; a supposedly well designed facility in a well off first world country having multiple containment failures. Yeah for sure, a natural disaster started the sequence of events, but it's a reminder that even after a shutdown / scram there is still quite an immediate problem of maintaining control for cooling everything down.