NVIDIA Blackwell's High Power Consumption Drives Cooling Demands; Liquid Cooling Penetration Expected to Reach 10% by Late 2024

[Klemc]

Why the name's Blackwell, any source ?

 

[Yukikaze]

Why the name's Blackwell, any source ?

David Blackwell - Wikipedia

 

[[crs]]

I really would love the card manufacturers to release a card with no heatsink, lower the price accordingly, and I will buy my own water block.

 

[stimpy88]

Maybe it's time nGreedia focussed on the architecture IPC instead of just adding more of the same building blocks to a die. Their greed has made them lazy.

 

[Jism]

The global climate is getting hotter and hotter...

It's time for PCs to also participate in this phenomenon !

2700W for just compute units looks staggering but if you have a target or a demand for a X compute performance a single unit will be more efficient then using 10+ units to archiving the same.

20 years ago if you would have bought a server or host multiple applications you likely needed 10 servers with 10x the power usage. Today you can get away with just one since the tech has turned so efficient.

So 2700W looks like a lot - its likely more efficient then hosting a machine with 4 separate GPU's.

 

[wolf]

They see nvidia on the title and go berserk.

/thread. As reliable as the sun rising. I'm just surprised nobody wrote the more you buy the more you save, or anything about a leather jacket.... yet.

I live a very energy and waste conscious life overall, it runs through the core of how I operate, yet, power draw ≠ efficiency and performance per watt, it's all relative. I do enjoy my efficient parts, and undervolt / overclock everything i own to get the most out of it per watt.

 

[Minus Infinity]

Just limit it? At some point we need to realize the 5090,just like the 4090, are the most efficient chips, thus the best at "protecting the environment".

Of course people will buy 5090s like crazy. Why wouldn't they? As ive said, even people that care about effieicny and power draw will buy 5090s cause they are the fastest at any power level.

LOL 4080 Super is the most efficient, so even smarter people worried about efficiency will save a pretty penny. Consumes way less power and gets more fps/Watt.

 

[metalslaw]

We should probably start growing more brains, as they say win-win

Probably cheaper to just put people in cryo pods, and hook their brains together in some kind of matrix.....

 

[JustBenching]

LOL 4080 Super is the most efficient, so even smarter people worried about efficiency will save a pretty penny. Consumes way less power and gets more fps/Watt.

No it doesn't. You do realize you can limit the power on all of these cards, right?

 

[yfn_ratchet]

LOL 4080 Super is the most efficient, so even smarter people worried about efficiency will save a pretty penny. Consumes way less power and gets more fps/Watt.

No it doesn't. You do realize you can limit the power on all of these cards, right?

I'll do you both one better and grab some numbers from a prior report: https://www.techpowerup.com/301649/...mit-and-undervolting-loses-just-8-performance

As it is, higher bins and higher tier chips will reliably perform more efficiently and carry a higher performance ceiling; the 4090 isn't even the cream of the crop. Try the Quadro RTX A6000 Ada. That's some top-shelf chips, and they come at a PREMIUM.

Generally though, point is that in the consumer market all players will push their power draw far beyond the zenith of the perf/watt curve to get a few extra (very marketable) frames. Undervolting is the one truth, fite me.

 

[starfals]

If you dont want it, dont buy it? There's clearly plenty of us willing to pay for powerful GPUs. That's ALWAYS been the case. And people have winged and moaned about high power expensive GPUs since the dawn of PCIe.


It's time you DID something about it! Have you abandoned your PCs and returned to an agrarian society? Oh, no, you have not. Better get on it, you got a planet to save!

Indeed, fk the planet. We are alive today and it's all fine today. A few hot days are no prob for any cheap AC. I can't wait for 2000-3000W GPUs, honestly. I want the 240 fps in RT. By the time the planet is dead, ill also be long gone too. At least i died with god rays and reflections all over my face while i was alive!

P.s. The 3000W GPU might be good for the winter too, think about it. It will warm your room. Here winters are freezing!

 

[Dr. Dro]

It’s really going to be up to us as consumers to reject these high power, expensive, obscenely high margin parts.

Consumer GPUs won't reach full-rack, platform power consumption figures, in other words, 5090 should still be 450W or less.

Why the name's Blackwell, any source ?

All NVIDIA GPUs are named after important theoretical physicists, inventors or mathematicians who made groundbreaking or landmark contributions to science. This one is named after David Blackwell, one of the most prominent mathematicians of his time and recipient of countless honors in his field

David Blackwell - Wikipedia

en.wikipedia.org

Ada is named after Ada, Countess of Lovelace (mathematician that is credited with creating the first computer program), Ampere is named after André-Marie Ampère, physicist credited with the discovery of electrodynamics and creation of the electrical telegraph, Turing is named after Alan Turing, computer scientist famous for breaking the Enigma code and father of modern computing, Volta is named after the inventor Alessandro Volta, creator of the battery, Pascal after Blaise Pascal, famous for his work in probability theory and many other areas, Maxwell after James Clerk Maxwell (theory of electromagnetic reaction), Fermi after Enrico Fermi (creator of the first nuclear reactor and member of the Manhattan Project), Tesla after Nikola Tesla, creator of the AC system, etc.

 

[JustBenching]

I'll do you both one better and grab some numbers from a prior report: https://www.techpowerup.com/301649/...mit-and-undervolting-loses-just-8-performance

As it is, higher bins and higher tier chips will reliably perform more efficiently and carry a higher performance ceiling; the 4090 isn't even the cream of the crop. Try the Quadro RTX A6000 Ada. That's some top-shelf chips, and they come at a PREMIUM.

Generally though, point is that in the consumer market all players will push their power draw far beyond the zenith of the perf/watt curve to get a few extra (very marketable) frames. Undervolting is the one truth, fite me.

Isn't that what I said kinda? I mean let's say nvidia releases the 5090 at 1000 watts. Why would I or anyone else actually care? I'd still limit it to 320 watts like i've done with my 4090. Don't see the problem.

 

[yfn_ratchet]

Isn't that what I said kinda? I mean let's say nvidia releases the 5090 at 1000 watts. Why would I or anyone else actually care? I'd still limit it to 320 watts like i've done with my 4090. Don't see the problem.

Was mostly just nabbing some graphs that demonstrate driving down power draw on a 4090/AD102 via either a simple PL adjust or a half-decent undervolt is very much capable of outperforming a 4080S/AD103 at the same power. It backs up your point, I suppose I wasn't all that clear towards that end.

 

[londiste]

Consumers buy perf/$ and get products geared towards that - maximum performance that can be wrung out of a chip.
Datacenters buy perf/W and get products geared towards that - maximum efficiency point that can be found.

The development goals flow and ebb here but usually it is a compromise between max performance and efficiency anyway.

 

[JustBenching]

Was mostly just nabbing some graphs that demonstrate driving down power draw on a 4090/AD102 via either a simple PL adjust or a half-decent undervolt is very much capable of outperforming a 4080S/AD103 at the same power. It backs up your point, I suppose I wasn't all that clear towards that end.

Which basically means the 4090 is indeed the most efficient desktop gpu currently in existence.

In other words, people that complain about ruining the planet with high end gpus, if you don't have 4090s you are the ones ruining the planet

 

[Noyand]

It’s really going to be up to us as consumers to reject these high power, expensive, obscenely high margin parts.

Our world will literally die the moment humans turn on millions of 1000W GPUs in order just to see better virtual water and light reflections in games and CGI films.

I don't think that the CPU-based render farms still used to render movies today are better in that regard...a lot of us here are judging those figures from a consumer running a single GPU POV, but the market that they are talking about has always dealt with power figures that a house wouldn't be able to handle. GPU rendering is actually much faster per watt, but the market leaders for heavy-duty rendering (Renderman and Arnold) still don't have feature parity between the CPU and GPU renderer. Last I've heard CPU rendering is more robust even if it's drastically slower per watt.
What We Can Learn from Pixar's 24,000-Core Supercomputer | No Film School
Wētā 'outgrew the power grid' during rendering of Avatar 2, producer says | Stuff
(Avatar 2 used Renderman which is still CPU based for a production output)