RGB-triangle.svg

<svg xmlns="http://www.w3.org/2000/svg" version="1.1" width="600" height="600" viewBox="-10 -10 120 120">

<style>
#rgb-triangle { isolation: isolate; }
#rgb-triangle>* { mix-blend-mode: screen; }
</style>

<defs>
    <polygon id="triangle" points="0,0 100,0 50,86.6" stroke="none"/>
    
    <linearGradient id="red" x1="50" y1="86.6" x2="50" y2="0" gradientUnits="userSpaceOnUse">
        <stop offset="0%"  stop-color="#f00" />
        <stop offset="100%" stop-color="#000" />
    </linearGradient>

    <linearGradient id="green" x1="0" y1="0" x2="75" y2="43.3" gradientUnits="userSpaceOnUse">
        <stop offset="0%"  stop-color="#0f0" />
        <stop offset="100%" stop-color="#000" />
    </linearGradient>

    <linearGradient id="blue" x1="100" y1="0" x2="25" y2="43.3" gradientUnits="userSpaceOnUse">
        <stop offset="0%"  stop-color="#00f" />
        <stop offset="100%" stop-color="#000" />
    </linearGradient>
</defs>

<g id="rgb-triangle" transform="matrix(1 0 0 -1 0 90)">
    <use href="#triangle" fill="url('#green')"/>
    <use href="#triangle" fill="url('#blue')"/>
    <use href="#triangle" fill="url('#red')"/>
</g>

</svg>

Nice triangle! I think it'll look even better (and it'll interact with subsequent filter operations such as feColorMatrix better) if you add to the style:

linearGradient { color-interpolation: linearRGB; }

@donhatch Now that I look at it again I think it actually needs some fixing. The rotations of gradients don't work as I expected.

As for your suggestion, it's quite possible that it will look better. But I think the purpose of this triangle was to be linear in the numbers, e.g. to have the color (255/3, 255/3, 255/3) in the center, which works with sRGB I guess (?).

"(?)" is right! I do not claim to understand what it's thinking-- multiple spooky layers of gamma correction and anticorrection everywhere, I suppose. But you're right, in the end sRGB seems to make it come out with #555555 in the center (and e.g. #800080 in the middle of the magenta edge), which seems like what we want, whereas linearRGB seems to make them come out gamma-adjusted to be somewhat brighter than that in the end.

I have at least fixed the gradients.