GeForce RTX 40-Series Flagship GPU May Hit 800W Power Limit
But the laptop parts remain limited to 175W or less.
Rumors claim that Nvidia's next-generation graphics cards may be power hogs. In a recent 'everything we know' article about Nvidia Ada Lovelace and the GeForce RTX 40 Series, we discussed power rumors, possibilities, and likelihoods in depth. Earlier today, Kopite7kimi shared some figures that may mean some of our higher estimates were too modest.
Before we go further, it is essential to distinguish between graphics card power limits and TDPs. The power limit will be the maximum wattage the GPU can run at by design and defined in the card's BIOS. High-end cards with the best coolers might be able to run at somewhere near to the power limit; they might offer higher power limits and be pushed quite a distance further by the tinkering of extreme overclockers. On the other hand, mainstream consumer cards will have TDPs that are probably significantly lower than the maximum possible power limit for the GPU.
AD102 is likely to appear in an upcoming GeForce RTX 4090 graphics card with somewhere between 132 and 140 SMs enabled (from a maximum of 144), according to our best knowledge. The weight of previous rumors suggests RTX 4090 cards will ship with TDPs of under 600W. However, Kopite7kimi reckons that the underlying GPU will have a power limit of an astronomical 800W.
Thankfully the AD102 GPU's power limit, hinting at huge TDPs, is an outlier in the Ada Lovelace generation. Moving down a peg to AD103, this second rung GPU will probably see service in an upcoming GeForce RTX 4080 and others. The power limit shifts down a gear dramatically to 450W, according to today's leak. For reference purposes, current gen GeForce RTX 3090 and RTX 3080 reference cards have BIOS-defined max power limits of 400W and 370W, respectively.
We have tabulated Kopite7kimi's latest aspersions, alongside some other essential stats. Please note that these are a collection of rumors, best guesses, etc., so add some salt.
GeForce RTX 40-Series Silicon*
GPU |
Desktop power limit |
Mobile power limit |
Max CUDA cores |
RT/Tensor cores |
---|---|---|---|---|
AD102 |
800W |
NA | 18,432 |
144/576 |
AD103 |
450W |
175W | 10,752 |
84/336 |
AD104 |
400W |
175W | 7,680 |
60/240 |
AD106 |
260W |
140W | 4,608 |
36/144 |
*Specifications are unconfirmed.
One of the more interesting revelations in the data today concerns the Ada Lovelace mobile GPUs. No Nvidia laptop partner wants to tame a GPU pulling more than 175W, so this stat anchors the top RTX 40 mobile GPUs. With the massive difference between desktop and laptop power limits, we should expect the platform performance gulf to get even wider in the upcoming generation.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
We just mentioned the power gulf between Ada desktop and mobile, but what might be most interesting to PC enthusiasts is what AMD can manage with RDNA 3, with much more sober power limits being rumored. A recent leak suggested that RDNA 3 graphics cards, i.e., the Radeon RX 7000 series, won't have much higher TDPs than their predecessors.
Mark Tyson is a news editor at Tom's Hardware. He enjoys covering the full breadth of PC tech; from business and semiconductor design to products approaching the edge of reason.
-
Phaaze88 "However, one must remember that power limits don't equate to mainstream TDPs."Reply
Yes, the actual product usually pulls more than the specified TDP.
RTX 3080 has a specified TDP of 320w, but a crap ton of models have 400w or more total board power limit. Plus, for very brief periods power use can spike higher than the BPL.
800w BPL, if it turns out to be real: ~dang... -
Sleepy_Hollowed So... are dual 1000w PSUs going to be a thing now? One on top and one on the bottom?Reply
This is pretty wild. With that amount of power requirements, that's what I'd do, so that the rest of the computer does not get destroyed by a power spike or a failing PSU. -
Neilbob Only a 175W limit for the mobile variants, eh?Reply
<sigh>
I remember when 175W might have been considered a bit excessive for desktop graphics cards. -
watzupken I seriously think Nvidia is out of their mind to release a retail class GPU @ 800W power limit. Power consumption aside, it will dump a lot of heat into the atmosphere and raise temps really quickly in any room. When you consider the power limit of the RTX 3090 at launch, this is more than double. Nvidia must be really desperate to fend off competition that they forgot they are making graphic card and not a bread toaster.Reply -
InvalidError
We'll see when the stuff actually becomes available. Maybe a good chunk of the power increases is simply due to AMD and Nvidia weighing their power calculations toward transient peak power so people don't run into as many issues with PSUs that should be large enough to handle a given GPU but cannot because short-lived transients trip protections.watzupken said:When you consider the power limit of the RTX 3090 at launch, this is more than double. -
Kamen Rider Blade
That might be necessary to make sure that you're not constantly overloading the cables or maxing out the Maximum wattage of the cables.InvalidError said:Time to bring out PSUs with 2x HPWR cables.
Wouldn't want your 12V HWPR cables to melt by constantly running them near the edge of their 600 watt per cable upper end specs.
Isn't the newly minted ATX 3.0 PSU spec designed to address the high transient power spikes?InvalidError said:We'll see when the stuff actually becomes available. Maybe a good chunk of the power increases is simply due to AMD and Nvidia weighing their power calculations toward transient peak power so people don't run into as many issues with PSUs that should be large enough to handle a given GPU but cannot because short-lived transients trip protections.