Nvidia GeForce RTX 4090 Reviews

In heavy PT/RT scenarios it pushes the power envelope to the max, ~430w/450w, so it definitely loses maybe 15% performance when you limit power below the default.
And even then, only if you run it at a very high internal resolution and have a CPU/memory setup that won't bottleneck the card. In other words, it is difficult to make the 4090 sweat.
 
In heavy PT/RT scenarios it pushes the power envelope to the max, ~430w/450w, so it definitely loses maybe 15% performance when you limit power below the default.
Don't think it's that much, in Portal RTX I lose about 7% if I limit it from 600W to 360W (80%) and that's a worst case scenario, most games won't push past 400W even with OC.
 
But, one part of the answer is that a lower TDP would not allow future, more demanding games or software to make use of the 4090 to its fullest. There's no way around it, you need a bunch of power to light-up all those transistors at once. As CPUs and system memory get faster and games evolve to make more use of the 4090, I expect you'll see its power consumption trend upward--but so will performance or, at least, performance won't buckle as it would if the card became power limited at 300W or 350W.

Great point.
 
In heavy PT/RT scenarios it pushes the power envelope to the max, ~430w/450w, so it definitely loses maybe 15% performance when you limit power below the default.
I played/benchmarked God of War, Marvels Guardians, Far Cry 6, Watch Dogs Legion, Plague tale requiem, AC Valhalla, Cyberpunk, Horizon ZD, Metro Exodus EE, Deathloop, Latest Tomb raider and Quake 2 RTX. Quake 2 RTX is by far the most heaviest game on the GPU. At 3840*1620 res 2745MHz@950mV GPU clock (which is considered somewhat undervolted), it hovers around 420-450W. While rest of the games are mostly between 220-320W. Metro is also heavier which can be close to 400W sometimes.
 
I ran a quick benchmark but for some reason screenshots appear darker than it appears on the screen.

Quake II RTX, 3840*1620, settings, savegame file
CCLKMCLKPWRFPSNORMALIZEDProfile
25353000~348104102%CURVE METHOD 2, 2535MHz@875mV, +1500MHz MCLK, 111% PL (500W)
27453000~421109107%CURVE METHOD 2, 2745MHz@950mV, +1500MHz MCLK, 111% PL (500W)
~27152625~438102100%DEFAULT, Palit GameRock OC, 100% PL (450W)
~29403000~487112110%+200MHz CCLK, +1500MHz MCLK, 111% PL (500W)
 

Attachments

  • save4.zip
    23.4 KB · Views: 1
Great UV settings, guys! I often run F@H, but I typically change the power limit to 50% without touching any other setting to minimize heat/power usage. On my Founder's Edition, I tried 2715 @ 950mv (without touching any other setting, power limit at 100%), and saw great results.

At 50% power limit, I averaged 2700MHz and ~350W-360W on GPU. The GPU maintained a constant 1050mv.
Using the UV settings, I averaged 2715MHz and 265-290W on GPU. The GPU maintained a constant 950mv.

Both settings saw a steady 95% GPU utilization.

It'll be fun to tinker with the undervolt settings from here!
 
Placed here as Ada spec/review thread is locked.
While the NVIDIA RTX 6000 Ada Generation launched in early 2023, it remains a very performant graphics card for workstation use. It was only a few weeks ago I finally received a NVIDIA RTX 6000 Ada Generation graphics card review sample from NVIDIA and thus putting it through its paces now under Linux. This does allow for a nice look at how the mature software support is competing for both the NVIDIA RTX Ada Generation graphics cards as well as the AMD Radeon PRO W7000 series on their respective drivers.
 
Back
Top