Hacker News Reader

Comment by adamsmith143 | Hacker News Reader

Why can't we get efficiency leaps like this in GPUs? Tired of my 3080 heating my entire office up...

Undervolt/downclock it? Afterburner or GPU Tweak, etc, will do this relatively easily. (I've undervolted AMD cards that come with ridiculous core voltages without having to drop clocks at all in past)

Almost everything is stock clocked past the point of diminishing returns right now. This Intel part looks to mostly be a downclocked version of existing 13900.

Look at the recent AMD 7900 vs 7900X. You can get 90-95% of the performance for far less power by just backing off the voltage and clocks a bit. (In their TDP terms going from 115w to 65w TDP, loses less than 5-10%)

Everyone's fighting for chart king/significant generational improvement number they can point to and missing the sweet spot on the efficiency curve, but you can bring it back yourself. I bet the 3080 still runs great at 50-100w less power limit/TDP depending on your use case and I doubt that will result in anywhere near a 1:1 perf/power reduction.

Yeah, exactly. I was able to get a decent ~50W drop and stayed at a higher than base clock rate (1920 MHz @ 900 mV). It's still a space heater, but it's better.

I also frame rate limit myself in a lot of games - I don't need my MMOs running at 160 fps, so many games I'm at 150-200 W. I still wish it was less, but that's much more reasonable than 400 W.

Which makes you wonder why GPU companies don't proactively give you control over these parameters.

AMD's GPU drivers give you a decent range of control over board power limits, fan curves, GPU and memory clocks and voltages, and on at least some models the ability to fine-tune the shape of the voltage/frequency curve used. Really the only thing missing is an automated tool to explore the V/f parameter space to find the limits of stable operation for your particular chip.

Even on nVidia cards you can literally edit the V/f curve.

I'm using the software that came with my card to do that. It makes sense that the graphic card manufacturer's would ensure stability with the cost of it running hot - by undervolting I'm playing with potential crashes and that won't do for a card defaults settings.

> (I've undervolted AMD cards that come with ridiculous core voltages without having to drop clocks at all in past)

It's often even worse than that. There are plenty of cases where you can undervolt so far that you now have enough headroom in the power delivery and cooling to allow you to run at substantially higher clock speeds.

Performance/watt increases more every GPU generation than it does every CPU generation.

Arguably Intel is now just catching up with the times after their long manufacturing quagmire, I suspect large part of the "leaps" are due that. As GPUs have been managing more steady progress, such leaps do not happen. Also there is lot of doubt around these sort of benchmarks, especially for P+E setups

The 4090 is a lot better in terms of heat.

Depends what you mean. It generates more heat, it’s just more effective at getting that heat off the die and into your room.

Edit: since it’s performance per watt is higher, if you’re capping frame rates then you can get less heat out of the 4090. Like I said, depends what you mean.

At half of power usage it only loses 10% performance.

https://videocardz.com/newz/nvidia-geforce-rtx-4090-power-li...

It's hard to frame that as 'better' though.

It's more badly tuned, so it's more fixable.

Reply on news.ycombinator.com