GPU boost - gimmick?

[notty22]

Yeah, that's referring to Adaptive Vsync which we all know about, it's done in the driver and not a fixed function hardware unit as you described. As for 'Dynamic framerate function when played in surround' that could be anything, it could perhaps refer to single GPU running 3 monitors in surround which is a launch feature.

Here's TH take on it. Nvidia is not revealing the patent, for lack of a better word.

When headroom exists in the card’s thermal budget, GPU Boost dynamically increases the chip’s clock rate and voltage to improve performance. The mechanism is actually somewhat similar to Intel’s Turbo Boost technology in that a number of variables are monitored (power use, temperature, utilization, and more—Nvidia won’t divulge the full list) by on-board hardware and software. Data is fed through a software algorithm, which then alters the GPU’s clock within roughly 100 ms.

IMO, the magic has to happen within those ms, or would not be effective.

 

[evilspoons]

Let's try to straighten this out a bit for those who don't quite understand the point of GPU boost...

A GPU or a CPU have various functional units that they can be divided in to. Most of them can do one of many different tasks. Some tasks take more power to do than other tasks, but each functional unit can only do ONE task at a time.

So. Let's say we have an imaginary (very very small) GPU with 4 functional units. All of them are doing task X, which takes 5 mW. This GPU would be drawing 20 mW plus overhead for, say, 10 clock cycles until the task completes. So, 20 mW for 10 cycles.

Now, let's say we have a task Y. This takes 10 mW, and also takes 10 cycles to complete. Now we fully load the GPU with 4 of these tasks - the GPU is now using 40 mW for 10 cycles.

In both situations, the GPU is 100% jammed up, it can't do anything more, but when running task X, it uses half as much power as when running task Y.

So the "boost" situation is to overclock the CPU when doing task X. Double the clock speed. We still take 10 clock cycles to complete, but since they go by twice as fast we are now drawing 2x20 mW = 40 mW (our peak usage on task Y).

We have now doubled the GPU's efficiency at doing task X by maximizing the usage of our thermal envelope (40 mW peak).

If you decided to run two of task X and two of task Y, you're using 30 mW. This still gives you 10 mW of overhead, so you could overclock by 33%, hit your 40 mW envelope again, and get task X and task Y finished in less than the time for the base clock.

This is what GPU Boost is doing, only over an entire graphics card with thousands of stream processors, hundreds of texture units, and so on.

 

[railven]

I'll test this out myself, but I asked this during the initial revela of this GPU Boost.

If I were using V-sync, would GPU boost even matter to me?

I'd figure in games that I'm rendering under 60 FPS that aren't CPU bottlenecked, my GPU should be at 100% load and thus not have much overhead? Anything above 60 FPS wouldn't matter due to v-sync?

With adaptive V-sync now not sounding all that, and if normal v-sync essentially counters GPU-Boost - the feature set isn't as exciting anymore.

 

[Dribble]

While this exists on desktop cards and is useful I think it was really designed to help laptops.