Discussion Nvidia Blackwell in Q1-2025

[Saylick]

If the AI bubble pops, NV will pivot to robotics..

Nvidia already have announced that robotics is the “next big thing”. The pivot hasn’t happened yet but they are definitely trying to set themselves up for it if it happens.

 

[Ranulf]

5060ti rumor shows up with another round of 8GB and 16Gb variants. Maxsun brand supposedly prepping for it now.

https://videocardz.com/newz/maxsun-preparing-geforce-rtx-5060-ti-with-16gb-and-8gb-memory

 

[Mopetar]

Selling an 8GB xx60 Ti in this day and age ought to be criminal. I feel no remorse for anyone who buys one though. Some people need to touch the stove in order to learn.

 

[Heartbreaker]

Selling an 8GB xx60 Ti in this day and age ought to be criminal. I feel no remorse for anyone who buys one though. Some people need to touch the stove in order to learn.

4060 Ti showed my that most people pay so little attention to reviews. 4060 Ti was ripped to shreds in reviews, and then went on to sell like hot cakes.

 

[Thunder 57]

Selling an 8GB xx60 Ti in this day and age ought to be criminal. I feel no remorse for anyone who buys one though. Some people need to touch the stove in order to learn.

And they'd likely still outsell a 9060 12GB that performs the same or better at the same or lower price. Same should Intel come out with a more performant Battlemage at the same or lower price with more memory. Nvidia is the new Apple.

 

[poke01]

Nvidia is the new Apple.

I would even say worse than Apple now, Apple stopped selling 8GB Macs in 2024. NV knows that 8GB cards are not going to last very long and yet they sell them because they can. Apple caved due to market pressures ie Intel, AMD and QC starting their laptops at 16GB of RAM.

Who can pressure Nvidia?? NV marketing spits out DLSS, RT/FG and most think that 8GB VRAM is enough for those settings(its not). The only way to beat this is if AMD or Intel defeat the 5090 or any current gen xx90 class card because they brought the 4060 ti even after the bad reviews, this shows the halo effect is huge.

 

[Win2012R2]

Who can pressure Nvidia??

Benchmarks of popular games like Indiana clearly showing that 8 GB cards are dead, they will shift next gen to 50% more VRAM at least using 3 GB modules, maybe even refresh in 2026 because current 50 series are so bad

 

[Golgatha]

What a circus this is going to be.

Best Buy lists RTX 5080s from Gigabyte at up to 40% higher than MSRP — just one model at $999

That's a steep markup for custom models.

www.tomshardware.com

 

[SolidQ]

from reddit, true/not true, it's on your opinion

 

[MrTeal]

Looks like someone on reddit figured out how to divide 10752 by 9728.

 

[jpiniero]

https://videocardz.com/newz/geforce-rtx-5070-ti-launches-february-20-first-reviews-drop-february-19

Videocardz claims the 5070 Ti's release is Feb 20, with reviews a day before.

 

[Josh128]

from reddit, true/not true, it's on your opinion

100% true. Look at the shader increase.

 

[Josh128]

The end of tech as we know it? $6000 5090s? $2500 iphones?

Trump Reveals Intentions To Impose Up To 100% Taxes On TSMC's Chips Made In Taiwan; Prioritizing Domestic Production

The Trump administration has revealed plans to implement tariffs on chips made in Taiwan in an attempt to bolster domestic production.

wccftech.com

 

[Golgatha]

The end of tech as we know it? $6000 5090s? $2500 iphones?

Trump Reveals Intentions To Impose Up To 100% Taxes On TSMC's Chips Made In Taiwan; Prioritizing Domestic Production

The Trump administration has revealed plans to implement tariffs on chips made in Taiwan in an attempt to bolster domestic production.

wccftech.com

I guess no one informed Trump that TSMC has been there since 1987, that it would take a decade to build a similar operation and get up and running domestically, and you probably couldn't staff it appropriately using only US labor anyway?

 

[jpiniero]

and you probably couldn't staff it appropriately using only US labor anyway?

IIRC the TSMC Arizona fab is mostly (?) staffed by Tawianese citizens on worker visas.

 

[ToTTenTranz]

The end of tech as we know it? $6000 5090s? $2500 iphones?

Trump Reveals Intentions To Impose Up To 100% Taxes On TSMC's Chips Made In Taiwan; Prioritizing Domestic Production

The Trump administration has revealed plans to implement tariffs on chips made in Taiwan in an attempt to bolster domestic production.

wccftech.com

It would only apply to the US, right?

If these tariffs are only for products coming into the USA with Taiwan-made ICs, then they're only applicable when the GPUs leave the assembly lines in China/Vietnam/etc. and reach USA shores.

That would mean everyone else like us Eurolanders are safe from this crap. We're still paying ~23% VAT, but it's not 100%.

 

[Josh128]

It would only apply to the US, right?

If these tariffs are only for products coming into the USA with Taiwan-made ICs, then they're only applicable when the GPUs leave the assembly lines in China/Vietnam/etc. and reach USA shores.

That would mean everyone else like us Eurolanders are safe from this crap. We're still paying ~23% VAT, but it's not 100%.

I believe, but I dont know if Trump realizes that AMD, Intel, and Nvidia also sell to the rest of the world, lol. This sounds outlandish, tbh.

 

[Golgatha]

It would only apply to the US, right?

If these tariffs are only for products coming into the USA with Taiwan-made ICs, then they're only applicable when the GPUs leave the assembly lines in China/Vietnam/etc. and reach USA shores.

That would mean everyone else like us Eurolanders are safe from this crap. We're still paying ~23% VAT, but it's not 100%.

Don't forget, we get the tariff + sales tax (locally about 10% in Kansas). Can't wait to give my 110% (secretly buys all electronics I think I'll need for the next 4 years)!

 

[jpiniero]

It could mean why Blackwell DIY supply is so low - they are focusing on OEMs first.

 

[Win2012R2]

It could mean why Blackwell DIY supply is so low - they are focusing on OEMs first.

Supply is very low because this is a **** gen so the only way they can get away with launching it is by creating artificial scarcity.

 

[Timmah!]

Just saw the supposed pricing for MSI 5090s here and its downright bizzarre. Like only one model, Ventus, at MSRP, as expected. But the remaining 4 models, all straight up at 3000 EUROs or more (obviously including VAT). So no 200 to 600 EUROs about MSRP options, like in the past, straight up 600 for second cheapest variant. I am speechless.

 

[adamge]

If the AI bubble pops, NV will pivot to robotics..

Can they pivot to rasterization performance? That has been neglected for the last six years.

 

[Heartbreaker]

Can they pivot to rasterization performance? That has been neglected for the last six years.

The AI bubble that might get deflated a bit are the big front runners like OpenAI.

DeepSeek is more like how PCs first decentralized computing away from mainframes.

Computing didn't go away, it just became more pervasive, and NVidia will still be selling HW to the more decentralized AI players, even if OpenAI has to slow it's roll.

 

[amenx]

Deepseek bypasses CUDA to use PTX close to metal programming that makes it far more efficient. Thus needing less HW to do the same thing. Thats what hit Nvidias stock.

DeepSeek's AI breakthrough bypasses industry-standard CUDA, uses Nvidia's assembly-like PTX programming instead

Dramatic optimizations do not come easy.

www.tomshardware.com

 

[branch_suggestion]

Deepseek bypasses CUDA to use PTX close to metal programming that makes it far more efficient. Thus needing less HW to do the same thing. Thats what hit Nvidias stock.

DeepSeek's AI breakthrough bypasses industry-standard CUDA, uses Nvidia's assembly-like PTX programming instead

Dramatic optimizations do not come easy.

www.tomshardware.com

Parallel Thread Execution - Wikipedia

en.wikipedia.org

The exact opposite, Tom's Hardware should be ignored and laughed at.
NV's stock is down because their margins are going to return to the mean, lower barrier of entry and the increasing shift to inference over training means NV faces more competition and it will be harder for large customers to make ROI on such massive CapEx investments.
Not to mention models like this infer nicely in 8 GPU boxes, no need for a more fancy switching setup unless you are trying to break through to a new frontier of model ability.
If the next wave frontier models, which are far more expensive to train and run do not offer a large enough leap in capability, NV will be forced to lower prices on future or existing orders or risk orders being canceled and given to other IHVs or internal Si.
Their moat is in training, inference is a far more open game.

Oh, and they know this is inevitable, Cisco crashed when Broadcom et al offered 90%+ of the capability for half the money.
Robotics is their next big bet to try to chase the eternal boom.
But NV is a boom bust company, they cannot help themselves whenever they see a chance for a quick buck.