Discussion Nvidia Blackwell in Q1-2025

[Heartbreaker]

Rubin got pulled into later this year because pretty clearly Blackwell in general is having issues, N4 is just not enough for it.

So now it's coming in 2025. Sure...

I'd like to see your evidence of that.

Because you can't just "pull things in" a year ahead of schedule, in chip design.

 

[CakeMonster]

I've seen so much wishful thinking about the next gen on various forums the last couple of days. If anyone has any minimally credible sources that it will arrive faster than usual, please provide them. If not I suspect its mostly cope and wishful thinking. We've been on a pretty steady 24-28m cycle for a long time, at best ~22 months. And that has been even in situation with more competition and with better nodes available, still they keep to the cycle.

 

[Win2012R2]

So now it's coming in 2025. Sure...

Rubin for "AI", appears to be - "Rubin is a microarchitecture for GPUs by Nvidia announced at Computex in Taipei in 2024 by CEO Jensen Huang... It is scheduled for mass production in late 2025 and will be available for purchase in early 2026"

Rubin (microarchitecture) - Wikipedia

en.wikipedia.org

 

[CakeMonster]

What I'd love to see is a CPU scaling review, where they take the fastest CPU out there (probably 9800X3D) and run it a different speeds, including OC, to try and identify where the CPU limitations are. We can make good guesses with different CPU's already, but I think a lot of people need to see it for themselves. Just to be clear, I'm throughly unimpressed by how NVidia chose to spec this generation, but I think a lot of hardware enthusiasts still don't realize that we're increasingly seeing CPU's struggle with these really beefy GPU's.

 

[Grooveriding]

While they haven't stopped working hard at GPU R&D, for production, they seem be allocating the wafers overwhelmingly to their insanely profitable AI chips.

This wouldn't be a total surprise.

I think it's worse than paper really. I'd say it's almost a vapour-launch like he infers in this video. 4090 I had to try to get one, but not too hard, just had to be ready when some came available on Newegg. I don't think Newegg had any stock of 5090s at all so far. If so, just a very, very few. As I've never seen them come in stock on the site. There is just nothing out there, and as he says on the 5080, the only reason to buy one is if you don't have a 4070ti Super and up, and don't have a card. I reckon if don't start seeing 5080s available all the time very soon, there is a major production issue going on. 5090 may continue to be a unicorn like 3080s were during Covid.

 

[SolidQ]

What I'd love to see is a CPU scaling review, where they take the fastest CPU out there (probably 9800X3D) and run it a different speeds, including OC, to try and identify where the CPU limitations are

Here AMD CPU test from 1600 to 9800

 

[Kepler_L2]

Rubin got pulled into later this year

It hasn't, if anything it actually got delayed.

 

[Win2012R2]

It hasn't, if anything it actually got delayed.

"Nvidia has asked SK hynix to move up its delivery timeline for next-generation HBM4 memory chips by six months, according to SK Group Chairman Chey Tae-won, reports Reuters.
Initially, SK hynix planned to ship its HBM4 chips to customers in the latter half of 2025. Following Nvidia CEO Jensen Huang's request, the timeline was shortened, though the exact new schedule was not specified. "

5 nov 2024 - https://www.tomshardware.com/pc-com...-hbm4-chip-delivery-by-six-months-says-report

At least HBM won't be holding it back, maybe Nvidia will delay it because Blackwell is clearly delayed and it must have been pre-sold

 

[Ranulf]

Owen on the 5080 vs 4090:

 

[Heartbreaker]

Rubin for "AI", appears to be - "Rubin is a microarchitecture for GPUs by Nvidia announced at Computex in Taipei in 2024 by CEO Jensen Huang... It is scheduled for mass production in late 2025 and will be available for purchase in early 2026"

Rubin (microarchitecture) - Wikipedia

en.wikipedia.org

You said: "pretty clearly Blackwell in general is having issues", which makes it about the gaming GPUs we are discussing.

There is no sign gaming GPUs are going to break form the typical 18-24+ month schedule, which is what most of care about.

So again, I'd expect ~2 year life for gaming GPUs, with a mid cycle Super Refresh, with some 3GB VRAM chips for some models.

 

[GodisanAtheist]

Hell maybe NV is like "You people don't like our cards?! FINE! Then you don't get any and we'll see how you like THAT!"

After a year of no NV GPUs on store shelves people will greedily lap up whatever slop NV shovels in front of them.

Seems perfectly rational, it's what I'd do if I was a $3T dollar company...

 

[Heartbreaker]

Hell maybe NV is like "You people don't like our cards?! FINE! Then you don't get any and we'll see how you like THAT!"

After a year of no NV GPUs on store shelves people will greedily lap up whatever slop NV shovels in front of them.

Seems perfectly rational, it's what I'd do if I was a $3T dollar company...

Doesn't quite hold up. It likely cost several hundred million up front for the new chips.

If you don't care and are just going to short stock, it makes it that much harder to recoup those upfront costs.

Why not simply save all the up front costs and rebrand Ada as 50 series. Rebrands have happened before. That way you save all those up front costs...

If I was running a $3T company and was going to backburner gaming GPUs, that is what I would have done.

 

[Mopetar]

the 980 was not GM200, it was GM204 - only the 980 Ti and Titan were GM200

You are correct. Perhaps I should have compiled all of my information into a some kind of table in a program like Microsoft Excel in order to avoid making this kind of mistake.

Of course then someone on the Internet might call me a nerd.

 

[ToTTenTranz]

Rubin for "AI", appears to be - "Rubin is a microarchitecture for GPUs by Nvidia announced at Computex in Taipei in 2024 by CEO Jensen Huang... It is scheduled for mass production in late 2025 and will be available for purchase in early 2026"

Rubin (microarchitecture) - Wikipedia

en.wikipedia.org

Sounds like an AI-only arch, like Volta and Hopper.


When was the last time Nvidia took less than 24 months between architectures? I think it was 20 years ago, between Geforce FX and Geforce 6.
Things are different now.

 

[Grooveriding]

Just got lucky and got an order in for an AIB on Best Buy here. These cards are robbery. The markup over MSRP is ridiculous. The card I got was $600 more than MSRP.

 

[maddie]

Just got lucky and got an order in for an AIB on Best Buy here. These cards are robbery. The markup over MSRP is ridiculous. The card I got was $600 more than MSRP.

But you just checked the "I consent to be robbed" checkbox.

 

[Grooveriding]

The AIBs are the scalpers now. I didn’t want an FE, I don’t trust the high memory temps in tandem with a new type of GDDR on the cards. Just had to bite the bullet as I had already sold my 4090 and needed a card.

I feel a sucker though. Nvidia keeps raising prices and I keep buying. Where is the cutoff lol.

 

[CakeMonster]

Yeah I prefer AIB as well, for the same reasons, but not the very largest cards. One day I might want to drop a monster like this into a secondary PC or give it away to a friend, and the largest ones simply don't fit most cases. So I'm in a bit of a bind, just considering the 'smaller' ones. Also, Asus has 2 HDMI ports, which is nice, although obviously not worth a huge premium either.

 

[Grooveriding]

The Asus Astral is $4100 Canada bucks. Just insane.

 

[gdansk]

Next time they should auction off the initial stock instead.
People seem willing to pay it. @moonbogg was right again!

 

[Win2012R2]

You said: "pretty clearly Blackwell in general is having issues", which makes it about the gaming GPUs we are discussing.

Yes, Blackwell in general having issues - AI parts which are the most valueable been delayed a lot due to heat issues, so Rubin is based on public info is shifted to be out earlier than originally planned, would it also have a gaming card or be like Volta gen? I don't know but given how poor Blackwell in gaming too it makes me think that they would bring 60 gen earlier than 2 years.

Frankly even with limited N3 capacity it 2026 they will be able to make more gaming cards than got "released" yesterday.

Obviously a lot depends on AMD's plans for UDNA/RDNA5, if they get release next year then I can't see Nvidia just ignoring it - Blackwell gaming is the worst release in a very long time, Jensen knows it and he is not the guy who likes to lose.

 

[adroc_thurston]

so Rubin is based on public info is shifted to be out earlier than originally planned

It's not.

 

[Win2012R2]

It's not.

And Jensen asked Hynix to speed up supply of HBM4 by 6 months just for kicks?

https://www.reuters.com/technology/nvidias-huang-asked-sk-hynix-bring-forward-supply-hbm4-chips-by-6-months-sks-2024-11-04/

Blackwell is clearly a hot steaming t**d.

 

[adroc_thurston]

And Jensen asked Hynix to speed up supply of HBM4 by 6 months just for kicks?

Yea.

 

[Win2012R2]

Yea.

Yeah, and he must have paid big bucks for it too, because that's what he likes doing for kicks.

What's your rational why Rubin isn't pulled in quicker than originally planned? They clearly need new process