Insiders Predict Introduction of NVIDIA "Blackwell Ultra" GB300 AI Series at GTC, with Fully Liquid-cooled Clusters

[T0@st]

Supply chain insiders believe that NVIDIA's "Blackwell Ultra" GB300 AI chip design will get a formal introduction at next week's GTC 2025 conference. Jensen Huang's keynote presentation is scheduled—the company's calendar is marked with a very important date: Tuesday, March 18. Team Green's chief has already revealed a couple of Blackwell B300 series details to investors; a recent earnings call touched upon the subject of a second half (of 2025) launch window. Industry moles have put spotlights on the GB300 GPU's alleged energy hungry nature. According to inside tracks, power consumption has "significantly" increased when compared to a slightly older equivalent; NVIDIA's less refined "Blackwell" GB200 design.

A Taiwan Economic Daily news article predicts an upcoming "second cooling revolution," due to reports of "Blackwell Ultra" parts demanding greater heat dissipation solutions. Supply chain leakers have suggested effective countermeasures—in the form of fully liquid-cooled systems: "not only will more water cooling plates be introduced, but the use of water cooling quick connectors will increase four times compared to GB200." The pre-Christmas 2024 news cycle proposed a 1400 W TDP rating. Involved "Taiwanese cooling giants" are expected to pull in tidy sums of money from the supply of optimal heat dissipating gear, with local "water-cooling quick-connector" manufacturers also tipped to benefit greatly. The UDN report pulled quotes from a variety of regional cooling specialists; the consensus being that involved partners are struggling to keep up with demand across GB200 and GB300 product lines.



View at TechPowerUp Main Site | Source

 

[Ominence]

The 5090 Ultra Titan cometh! Bow before your Supreme Overlord, lol!

 

[furia]

Better would be if they cut this wasted pile of good silicon into many actual GPUs

 

[Philaphlous]

Hello Fermi! 1400w tdp bring on the space heaters! They're going to have to redesign the 12v power lines because with all the 5090's melting I'm sure 116 amps on your 12V psu is going to cook those wires in no time...
Maybe we'll have 48v psu's now...

 

[Rightness_1]

NV has hit the wall. Nothing left in the tank but overclocking and making the die larger.

 

[Dr. Dro]

This is obviously an exascale AI chip, why are people acting like they'll release a GeForce based on this? It's like looking at a MI300X and moaning that an APU is now a power guzzling fire hazard or something.

 

[Bwaze]

People still want to believe RTX cards are somehow relevant in this generation - despite all the hard numbers from Nvidia itself, where it was shown in financial report for the last quarter that "Gaming and AI PC" only amounts to about 6% of revenue - and we can only guess how much of that is Gaming.

There are even people who dispute this, and somehow think that all corporate purchases of RTX cards for AI, LLM acceleration gets counted into Data Center, that there is no chance Gaming + all the AI related purchases (Chinese smuggling, for instance) brings so little - even less than just Gaming before AI boom, and falling...

 

[Jism]

Correct - you really cant use consumer cards in DC's where if you where to work with software licences, for example Nvidia drivers would simply gimp the consumer cards in regards of compute.

AMD is making far more through enterprise right now then it ever will on consumer GPU's or CPU's even combined.

 

[igormp]

Correct - you really cant use consumer cards in DC's where if you where to work with software licences, for example Nvidia drivers would simply gimp the consumer cards in regards of compute.

There's nothing really technical that blocks you from slapping 4090/5090s into a 4U unit and using those to do whatever you want, just take a look at vast.ai and see the amount of such systems available for rent.
However Nvidia's terms do prohibit using consumer products this way (not that most people care), and the geforce lineup is unable to use some features such as vGPU or full rate tensor FP16 with FP32 acc, that's not related to the driver but rather firmware.