NVIDIA "Blackwell" Successor Codenamed "Rubin," Coming in Late-2025

[btarunr]

NVIDIA barely started shipping its "Blackwell" line of AI GPUs, and its next-generation architecture is already on the horizon. Codenamed "Rubin," after Vera Rubin, the new architecture will power NVIDIA's future AI GPUs with generational jumps in performance, but more importantly, a design focus on lowering the power draw. This will become especially important as NVIDIA's current architectures already approach the kilowatt range, and cannot scale boundlessly. TF International Securities analyst, Mich-Chi Kuo says that NVIDIA's first AI GPU based on "Rubin," the R100 (not to be confused with an ATI GPU from many moons ago); is expected to enter mass-production in Q4-2025, which means it could be unveiled and demonstrated sooner than that; and select customers could have access to the silicon sooner, for evaluations.

The R100, according to Mich-Chi Kuo, is expected to leverage TSMC's 3 nm EUV FinFET process, specifically the TSMC-N3 node. In comparison, the new "Blackwell" B100 uses the TSMC-N4P. This will be a chiplet GPU, and use a 4x reticle design compared to Blackwell's 3.3x reticle design, and use TSMC's CoWoS-L packaging, just like the B100. The silicon is expected to be among the first users of HBM4 stacked memory, and feature 8 stacks of a yet unknown stack height. The Grace Ruben GR200 CPU+GPU combo could feature a refreshed "Grace" CPU built on the 3 nm node, likely an optical shrink meant to reduce power. A Q4-2025 mass-production roadmap target would mean that customers will start receiving the chips by early 2026.



View at TechPowerUp Main Site | Source

 

[Space Lynx]

1 year cycle? I am surprised... I thought we were doing 2-3 year cycles for gpu's these days. wild

so rtx 5090 will be outdated within 12 months of launch, or is there something I am not understanding?

 

[G777]

1 year cycle? I am surprised... I thought we were doing 2-3 year cycles for gpu's these days. wild

so rtx 5090 will be outdated within 12 months of launch, or is there something I am not understanding?

It’s replacing the B100. We don’t know when Nvidia will be releasing the 60-series, but given that the R100 should be available early 2026, then the 60-series will probably be mid/late 2026 and early 2027.

 

[Space Lynx]

It’s replacing the B100. We don’t know when Nvidia will be releasing the 60-series, but given that the R100 should be available early 2026, then the 60-series will probably be mid/late 2026 and early 2027.

This makes more sense, ok so this is just for like openAI and big time buyers. consumer gamers are still on a 2-3 year cycle. noted.

 

[GodisanAtheist]

My bet is a major DLSS on steroids style initiative to lean very heavily on AI reconstruction of game scenes and AI generated frames to keep power draw and die sizes small.

All arrows point to AI Reconstructed games based on NV's trajectory, they might completely avoid chiplets etc and keep their die sizes and silicon costs under control through their software hegemony.

 

[cvaldes]

1 year cycle? I am surprised... I thought we were doing 2-3 year cycles for gpu's these days. wild

so rtx 5090 will be outdated within 12 months of launch, or is there something I am not understanding?

They are alternating enterprise and consumer product releases. So enterprise Blackwell is the focus one year followed by RTX 50 series the following year (consumer products based on the same architecture). Then enterprise Rubin the next year followed by a consumer RTX 60 series based on Rubin.

Clearly Nvidia knows where the money is. Datacenter now dwarfs their Gaming business by 6x. Their priority is getting next gen GPU architecture into their enterprise customers' hands first, not gamers.

So it's still a 2-year cycle if you look at one business or the other comparing apples to apples, oranges to oranges.

 

[Space Lynx]

It looks like they are alternating enterprise and consumer product releases. So Blackwell is the focus one year followed by RTX 50 series the following year (consumer products based on the same architecture). Then Rubin the next year followed by an RTX 60 series based on Rubin.

Clearly Nvidia knows where the money is. Datacenter now dwarfs their Gaming business by 6x. Their priority is getting next gen GPU architecture into their enterprise customers' hands.

So it's still a 2-year cycle if you look at one business or the other comparing apples to apples, oranges to oranges.

I'm genuinely surprised Jensen just didn't tell Nvidia gamers they would have to chill out for a few years, and go fully production AI to and become the worlds first 3-4 trillion valued company, they probably could have done that.

 

[cvaldes]

The Gaming business unit will still make decent gross margins because they can boost the retail price. And Nvidia doesn't have any real competition at the high end of gaming cards, they can charge whatever they want because a small number of people have the resources and desire to own the RTX 5090.

From a big picture perspective, supporting gaming products is still worthwhile in creating a larger software ecosystem. They release new Game Ready GeForce drivers with every new major game title release just like today. And they continue to add new titles to GeForce NOW and improve that offering.

It's better to think Nvidia more like Apple. While they book most of their revenue selling hardware, they think of themselves primarily as a software company whose software and services run best on their proprietary design hardware. For Nvidia, that's not just device drivers for GeForce gaming cards, it's also things like Nvidia Omniverse or the aforementioned GeForce NOW. Omniverse is a far more comprehensive and deep ecosystem than anything AMD has.

It's Nvidia's software that is the primary differentiator. A lot of these kind of companies think that their systems are about 70% software and 30% hardware.

And if rumors are to be believed, the next generation Nintendo gaming device (a.k.a. "Switch 2" for lack of an official name) will be powered by an Nvidia SoC. That firmly cements Nvidia's importance in a larger gaming universe beyond PCs.

And even if you hate cloud gaming and refuse to use GeForce NOW, much of what Nvidia learns from it can be used when they work with enterprise customers on their own cloud systems. The Intel nor AMD have cloud gaming services so that's one type of experience that Nvidia has and those two competitors do not.

 

[konga]

They are alternating enterprise and consumer product releases. So enterprise Blackwell is the focus one year followed by RTX 50 series the following year (consumer products based on the same architecture). Then enterprise Rubin the next year followed by a consumer RTX 60 series based on Rubin.

Clearly Nvidia knows where the money is. Datacenter now dwarfs their Gaming business by 6x. Their priority is getting next gen GPU architecture into their enterprise customers' hands first, not gamers.

So it's still a 2-year cycle if you look at one business or the other comparing apples to apples, oranges to oranges.

This isn't true. Datacenter GPUs are now on a yearly cycle. Blackwell became available for datacenters this year, while Rubin will be available for datacenters next year.

NVIDIA 2024-2025 data-center roadmap lists GB200 and GX200 next-gen GPUs - VideoCardz.com

 

[cvaldes]

This isn't true. Datacenter GPUs are now on a yearly cycle. Blackwell became available for datacenters this year, while Rubin will be available for datacenters next year.

NVIDIA 2024-2025 data-center roadmap lists GB200 and GX200 next-gen GPUs - VideoCardz.com

Interesting, so Nvidia has chosen to speed up the cadence, at least on the enterprise side of things.

That should make a positive impact on their financials and stock price. And yes, as an American with a retirement account, I have an indirect position in Nvidia (and AAPL, MSFT, GOOG, INTC, AMD, AVGO, QCOM, AMZN, FB, and many others).

Not sure it that will translate into a faster cadence of GeForce releases though. With current rumors pegging a late 2024 unveil for the first RTX 50 series card, it doesn't appear to be the case.

Anyhow Nvidia's Datacenter business has 6x revenue compared to Gaming. I understand why their priorities are on moving quickly here.

 

[Redwoodz]

The Gaming business unit will still make decent gross margins because they can boost the retail price. And Nvidia doesn't have any real competition at the high end of gaming cards, they can charge whatever they want because a small number of people have the resources and desire to own the RTX 5090.

From a big picture perspective, supporting gaming products is still worthwhile in creating a larger software ecosystem. They release new Game Ready GeForce drivers with every new major game title release just like today. And they continue to add new titles to GeForce NOW and improve that offering.

It's better to think Nvidia more like Apple. While they book most of their revenue selling hardware, they think of themselves primarily as a software company whose software and services run best on their proprietary design hardware. For Nvidia, that's not just device drivers for GeForce gaming cards, it's also things like Nvidia Omniverse or the aforementioned GeForce NOW. Omniverse is a far more comprehensive and deep ecosystem than anything AMD has.

It's Nvidia's software that is the primary differentiator. A lot of these kind of companies think that their systems are about 70% software and 30% hardware.

And if rumors are to be believed, the next generation Nintendo gaming device (a.k.a. "Switch 2" for lack of an official name) will be powered by an Nvidia SoC. That firmly cements Nvidia's importance in a larger gaming universe beyond PCs.

And even if you hate cloud gaming and refuse to use GeForce NOW, much of what Nvidia learns from it can be used when they work with enterprise customers on their own cloud systems. The Intel nor AMD have cloud gaming services so that's one type of experience that Nvidia has and those two competitors do not.

What makes you think locking gaming hardware into a monopoly run by Nvidia is a good idea?

 

[Minus Infinity]

AMD will be a spot of bother as they don't even have MI400 to compete against Blackwell yet and Rubin is coming next year and AMD originally was only having MI400 ready sometime mid-to-late next year, but it would be trashed by Rubin. AMD probably assume MI400 would be competing against Blackwell and would be surprised by Rubin's time frame for sure. MI350 needs to come out very soon (currently just slated for H2 2024) and MI400 bought forward 6 months at least.

 

[N/A]

only the A16 node would give them a relatively big 2.02x density increase. N3 vs. N5 only 30 - 60%

 

[Readlight]

Whatever, videocard prices are 500-1000€ not 100€

 

[Kn0xxPT]

Nvidia rushing next series early, must be to counter AMD APU's. the last thing Nvidia wants is companys buying APU's instead of dGPU's. Something like what happened with US Military buying PS3 machines because it was a better investment at price.

I can imagine, AMD APU with a perfomance of a 4070, in a single package. Nvidia Mid and Low end GPU's take a big hit, leaving them only with High-End.

Just a thought.

 

[stimpy88]

So Blackwell is a stop-gap product? Is this because RDNA 5 is going to kick it's arse?

 

[Tech Ninja]

The future is a small local SOC running GeForce Now on x080. This is how I play the newest games my 2080 can’t handle. I get a 4080 for a few $ a month!

 

[sephiroth117]

1 year cycle? I am surprised... I thought we were doing 2-3 year cycles for gpu's these days. wild

so rtx 5090 will be outdated within 12 months of launch, or is there something I am not understanding?

That was before their valuation reached 2 or 3 trillion ^^.

Of course those architectures are years in the making but they are in a merciless AI race, the consumer GPU market is just a small part now and we're getting the R&D leftovers.


Regarding the RTX 5090 or 5080 I'm interested in, I just have a simple rule: if it's still the same node 5nm as Ada lovelace, than it's really not worth picking this RTX 5000, it will, of course, have innovations and all but IMHO the 3nm is expensive and prioritized for AI & Datacenters and Rubin may indeed truly offer the efficiency breakthrough that comes with 3nm

I am not interested in a RTX 5090 that needs 600W to really differs from RTX 4090 because it's still is 5nm. I'm speaking way too soon as no one really revealed Blackwell consumer GPUs let alone Rubin but we'll see

 

[Space Lynx]

yeah waiting for new node shrink is probably smart. I am happy with my rig. I plan to upgrade the 7800x3d to 11800x3d whenever that launches in 2-3 years