News GeForce RTX 40-Series Flagship GPU May Hit 800W Power Limit

[Kahless01]

pathetic. they should be getting more efficient, not less. guess im still not upgrading from my 1080ti. i just want 4k60 at under 200W.

 

[InvalidError]

pathetic. they should be getting more efficient, not less.

If you get 100% more performance for 90% more power, it is still 5% more efficient.

 

[JWNoctis]

pathetic. they should be getting more efficient, not less. guess im still not upgrading from my 1080ti. i just want 4k60 at under 200W.

Wait for the marketing blurb; chances are they'll claim they are.

Technically, a 1080Ti is not under 200W either, and a 3070 mobile (for instance) would consume roughly half that much power for just about the same performance, or even a bit better.

...Though I've seen the latter spiking ~700W at times, if HWiNFO is to be believed and there's no sensor glitch.

I do still wonder what they'd come up for a cooling solution, for something with a constant power density closer to what must be 100W/cm^2 than not. That kind of power density would fill a tankless water heater with jealousy.

 

[Math Geek]

that kind of power would be too much for even a 360mm set-up. no clue how much it really would take, but no way air would be an option.

 

[Phaaze88]

View: https://www.youtube.com/watch?v=wnRyyCsuHFQ


I'd blame Nvidia.

 

[alceryes]

View: https://www.youtube.com/watch?v=wnRyyCsuHFQ


I'd blame Nvidia.

Yea! A new Gamers Nexus video!

It's definitely a GPU manufacturer problem. Steve spells it out nicely too. These spikes have been around for the last 5+ years but the issue now is that both the baseline power draw AND the transient spike multiplier are BOTH increasing substantially, while the GPU manufacturer-reported power draw (and recommended PSUs wattage, especially) aren't increasing enough to match.

 

[InvalidError]

It's definitely a GPU manufacturer problem.

Intel's updated ATX / 12VO specs say it is the PSU manufacturers' problem:
https://www.tomshardware.com/news/intel-atx-v3-psu-standard

The updated specs require that compliant PSUs support 100% overload for 100us at a 10% duty cycle. An ATX 3.0 / 12VO 2.0 compliant 800W PSU would need to be able to provide up to 100us long bursts up to 1600W per 1ms interval.

Higher-quality PSUs already have plenty of plenty good enough output capacitors to handle the transients and only need their OCP tuned to not trip on 2X current peaks under 100us long.

 

[alceryes]

Intel's updated ATX / 12VO specs say it is the PSU manufacturers' problem:
https://www.tomshardware.com/news/intel-atx-v3-psu-standard

The updated specs require that compliant PSUs support 100% overload for 100us at a 10% duty cycle. An ATX 3.0 / 12VO 2.0 compliant 800W PSU would need to be able to provide up to 100us long bursts up to 1600W per 1ms interval.

Higher-quality PSUs already have plenty of plenty good enough output capacitors to handle the transients and only need their OCP tuned to not trip on 2X current peaks under 100us long.

Incorrect.

300μs measurements shows the 3090 alone drawing over 590W. NVIDIA says that the card's max draw is 350W. 300μs is 3 times longer than the 100μs spec you just posted so it doesn't even apply in this instance. Not to mention that this spec was made official less than 5 months ago and the 3090 is 1 year and 9 months old.

My biggest issue is that the 3000 series cards (specifically anything in the 3080 or 3090 line) were horribly underrated in both max board power AND required power supply, by manufacturers, when they knew that these spikes went 2-2.5x their ratings.


Edit - So I see that these spikes are defined in 1ms, 10ms, and 100ms times in the new standard as well. It would have been great if this spec existed, in its final form, before the 3080s and 3090s were released and PSU mfgs. released PSUs back then to adhere to this new spec. The power ratings released by NVIDIA, when the 3080s and 3090s were released, was to adhere to the last ATX 2.5.2 or 2.5.3 spec not the new ATX 3.0 spec.

Unfortunately, putting the cart before the horse does nothing for users whose PSUs are powering off/rebooting because of NVIDIA's misrepresentation.

 

[ingtar33]

i can assure everyone right now, that i WILL NOT entertain a single thread on this forum asking "how the heck do i cool my 400w cpu + 800w gpu??" and the obvious "what 2000w psu do i need to run this??"

this, so much this... the only way to cool a beast like this will be to take heatexchange/water loops to the next level. like those old mods back before water cooling was something you could buy from a beautique, where people used copper plumbing and geothermic cooling by running the pipes out into their yard 20 feet deep or something stupid like that.

Could you imagine a 1200W system (cpu/gpu) in a room smaller then 200sqft? it would be like a server room, you'd need dedicated AC in the room, the noise from the fans could make you go deaf... and you'd be locked into test bench style setups lest you overheat your overpriced $1200 cpu, $600 mb, $1000 ddr5 ram, and $2000 gpu.

 

[TrantaLocked]

Power usage was already stupid with Ampere which is on average only 30% more efficient than Pascal. At times barely higher:

https://www.techpowerup.com/review/nvidia-geforce-rtx-3060-ti-founders-edition/36.html

MSI GeForce RTX 3070 Ti Suprim X Review

MSI's GeForce RTX 3070 Ti Suprim X is built around a massive cooler that achieves impressive temperature levels. What's even better is that the card runs much quieter than any other RTX 3070 Ti tested today: whisper-quiet, even when fully loaded.

www.techpowerup.com

I bought a 3060 Ti to undervolt it and at least that went well. It hits 150w max, but at 200W stock it's already a stupid card, let alone a 800w Lovelace. I won't upgrade again until Nvidia stops overvolting and overclocking their cards. Where is the mid range 150w card that they had for decades before Ampere? It wasn't just that Pascal was efficient, previous generations were never as ridiculous as Ampere and Ada Lovelace.