Noctua says Nvidia doesn't have enough dies to make big, brown, RTX 5090 — RTX 5090 Noctua Edition may never see the light of day

(Image credit: Noctua)

The RTX 5080 Noctua Edition is one of the most overbuilt RTX 50 series graphics cards on the market, featuring three NF-A12x25 G2 fans installed above a massive heatsink boasting 11 heatpipes. Such a design would logically have been perfect for an RTX 5090 model, and according to Kitguru, Noctua thinks so as well. However, it turns out Nvidia is the bottleneck preventing an RTX 5090 Noctua Edition from existing.

KitGuru talked with Noctua's Jakob Dellinger, who divulged that Nvidia does not have enough GB202 chip production "at the moment" to build an RTX 5090 Noctua Edition graphics card. This problem is likely why Noctua has never built an xx90-series RTX graphics card so far. The only graphics cards it has built with Asus are cards based on the RTX 5080, RTX 4080 Super, RTX 4080, RTX 3080, and RTX 3070.

However, Dellinger did discuss what an RTX 5090 Noctua Edition could look like if events ever change. Noctua would allegedly focus on changing the heatsink design to be more optimized for the RTX 5090, potentially using a vapor chamber design similar to the Asus ROG Astral. The RTX 5080 Noctua Edition model's 11 heatpipes could also potentially be switched to 8mm heatpipes entirely for this theoretical RTX 5090 version. By comparison, the RTX 5080 model comes with four 6mm heatpipes and seven 8mm heatpipes.

Nvidia's difficulty producing GB202 dies is something it has been struggling with since Blackwell debuted, thanks to the RTX 5090's insane market demand since its release earlier this year. Even to this day, RTX 5090 AIB partner cards are regularly priced hundreds of dollars above MSRP. This problem has been compounded by Nvidia releasing more GB202-powered GPUs recently, including the Nvidia RTX Pro 6000 Blackwell workstation and server edition graphics cards.

It's a real shame we probably won't see an RTX 5090 Noctua Edition graphics card for the foreseeable future. Noctua and its partnership with Asus have produced some of the most overbuilt air-cooled Nvidia graphics cards ever created, with cooling performance that can compete with AIO and water-block cooled GPUs. The RTX 5090's sky-high 575W TDP makes it the perfect candidate for Noctua to put its cooling experience to good use.

The RTX 5080 Noctua Edition is the latest GPU to come out of Noctua's partnership with Asus, featuring a first-ever triple-fan cooler design for a Noctua GPU cooling solution. As a result, combined with the aforementioned NF-A12x25 G2 fans, Noctua's RTX 5080 variant allegedly produces some of the most competitive cooling and acoustical performance behavior out of any RTX 5080 to date. Noctua's own testing claims the RTX 5080 Noctua Edition produces 1.7C lower VRAM, 6.2C lower GPU, and 14.5C lower dB(A) temperatures and noise compared to the Asus RTX 5080 ROG Astral.

Cooling performance is so good on the RTX 5080 Noctua Edition, that Noctua benchmarked its card at just 500RPM, achieving 74.6C GPU temperatures, which is more than acceptable temperatures for modern Nvidia GPUs. By contrast, the Asus ROG Astral model is barely able to keep GPU temps under 90C at the same RPM.

If Noctua gets around to making an RTX 5090 version with the same cooler design, it could very well be the best air-cooled RTX 5090 on the market.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

See more GPUs News

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.

12 Comments Comment from the forums

jlake3

Is it because it would use a unique shroud that would require new tooling? Because I'd seen speculation from someone with industry contacts that the RTX 5090 Dhahab had a run the double digit range, and Asus did a Doom themed RTX 5080 that was a limited edition of 666. Not a recent example, but the Mars II was supposedly a run of only 1500, and that definitely was not sharing a board or cooler with any other design.

I imagine that Noctua would have a hard time pitching a 5090 collab to Asus as a business venture, but if Asus wanted to do it for marketing buzz and bragging rights, they definitely could find some dies and a production slot.
Reply
jonaswox

Isn't modern semantics great. What he meant to say is that Nvidia did not prioritize this. Which is fine. But acting like it's some sort of force majeure is ridiculous.
Reply
court655

Oh no! Another overpriced noctua version won't be made? How will we live?! Stop supporting the greedy money making efforts of these corporations by not buying these and maybe we can get decent price/performance like we used to. 🙄
Reply
emike09

I've instead replaced all my intake and exhaust AIO fans with industrial noctua fans. I dont mind the noise. OC temps are cold at all times. And its usually quite quiet unless the system is under massive loads, in which case - Let that system roar!
Reply
bill001g

jonaswox said:
Isn't modern semantics great. What he meant to say is that Nvidia did not prioritize this. Which is fine. But acting like it's some sort of force majeure is ridiculous.
This is because a huge number of people reading these articles don't seem to understand that NVIDIA even as big as they are does not actually manufacture chips. They have contract with TSMC. It is also highly likely that TSMC can not just increase production without a long lead time for them to build the equipment used to manufacture chips. What NVIVIA can likely do is change the mix of chips they ask TSMC to make for them. To get more 5090 GPU chips they must reduce some other kind of chip. Right now they still make far more money making chips for AI purposes so it seem they use the capacity they have agreements for to make those chips.

Some day maybe soon the majority of AI companies will fail which will reduce the demand for these chips....especially if the AI that survives does not use NVIDIA based solutions.
Reply
blueshark3d

May be Nvidia will prioritize this in future ? ? ? ?
Reply
passivecool

If you do not get this, you should look up the concept of economy of scale.

emike09 Have you replaced the screws in the case with V2a M6? Hard to imagine how the thing stays together...
Reply
passivecool

bill001g said:
This is because a huge number of people reading these articles don't seem to understand that NVIDIA even as big as they are does not actually manufacture chips. They have contract with TSMC. It is also highly likely that TSMC can not just increase production without a long lead time for them to build the equipment used to manufacture chips. What NVIVIA can likely do is change the mix of chips they ask TSMC to make for them. To get more 5090 GPU chips they must reduce some other kind of chip. Right now they still make far more money making chips for AI purposes so it seem they use the capacity they have agreements for to make those chips.

Some day maybe soon the majority of AI companies will fail which will reduce the demand for these chips....especially if the AI that survives does not use NVIDIA based solutions.
you kind of have to wonder what part is just binning and salvaging all the chips with suboptimal functionality.
NV does not use the most modern nodes, but.... This is generally for sure the most difficult stuff that humanity has ever produced. It has taken dedicated researchers decades to create the tools necessary to create the tools to make the tools to make the chips. I feel this is frequently undervalued.

Threat to NV is apparently now that Gogl, FB, AZ are all making their own, specialized chips for AI. And AMD also wants a piece of the pie, and has the muscle to do it. Intel grew glacial for a decade but now, volcanos, everywhere. Very dynamic and exciting times.
Reply
greenreaper

blueshark3d said:
May be Nvidia will prioritize this in future ? ? ? ?
Their priority is money. If they can sell the same chip for twice as much on a pro card, they will. That's their job; to make as much as possible for their shareholders.
Reply
Eliad Buchnik

passivecool said:
you kind of have to wonder what part is just binning and salvaging all the chips with suboptimal functionality.
NV does not use the most modern nodes, but.... This is generally for sure the most difficult stuff that humanity has ever produced. It has taken dedicated researchers decades to create the tools necessary to create the tools to make the tools to make the chips. I feel this is frequently undervalued.

Threat to NV is apparently now that Gogl, FB, AZ are all making their own, specialized chips for AI. And AMD also wants a piece of the pie, and has the muscle to do it. Intel grew glacial for a decade but now, volcanos, everywhere. Very dynamic and exciting times.
Most large corporations can design an accelerator. But creating a system wise solution of large scale array of interconnected accelerators is more difficult, creating software suit with all the libraries that are available for CUDA is the hardest part and the reason why solutions from Amazon, Google, AMD and other haven't got any serious traction - developing your software solution and migrating the existing libraries to new hardware solution takes years - something companies can't afford in the AI race, and for the foreseeable future why almost everyone will keep using CUDA despite if they have their own hardware solutions. While it may seem to be exiting, unless someone gets full software solution that rivals CUDA, it won't be very existing.
Reply

Show more comments