'Doom Eternal' Runs at 1,000 FPS, If You've Got Liquid Nitrogen

The engine that runs Doom Eternal, iD Tech 7, is so powerful that it can run the game at 1,000 frames per second. To achieve levels of such intense gaming requires an RTX 2080TI, a Intel i9 9700K, a 1200w power supply, and a bucket of liquid nitrogen to keep the whole thing cool. Doom Eternal hitting massive FPS heats up the system so much that only the -196 degrees Celsius liquid nitrogen could keep the whole thing cool.

This is a companion discussion topic for the original entry at https://www.vice.com/en_us/article/bv8a58/doom-eternal-runs-at-1000-fps-if-youve-got-liquid-nitrogen

This is really cool, but I do find it a little funny that the aspect that people keep pointing to is the liquid nitrogen, which is the cheapest and most easily obtainable component of this rig.


I wonder if the Doom soundtrack will ever surpass “BFG Division”


I don’t know much about game programming, but executing any advanced loop in under 1 ms seems extremely impressive to me. That’s a lot of pieces which have to fit together almost perfectly, even with a massively overclocked system.

Already did :smiley:

Seems like they are looking at a wall to get the number that high. The ~500 they were getting during combat is impressive. Got me wondering what the max FPS monitors currently support even is.

now I’m imagining overpriced branded gamer liquid nitrogen cylinders with RGB lighting


Is it just me or is the predicate for this article kinda meaningless?
Lots of much less “powerful” game engines can run at hundreds or thousands of rendering frames per second, if they’re on a sufficiently powerful system [and, in fact, one of the problems with running very old games on modern systems is precisely that their internal frame rates end up much higher than the designers ever thought would be possible].

I am pretty sure that the original Quake engine, for example, runs at >900fps on modern systems, without max fps limits set… (which is why you do set an fps limit, because running at that frame rate exposes problems in the simple physics engine).

Now, it’s possible that they mean “optimised” rather than “powerful”, but this article is horrible at conflating unrelated ideas of what a well performing game is [and how that links to hardware requirements].


I agree with you, but I think it would be very difficult for a general-audience article to dive into why this is an impressive feat. I’d love to read that article, though.

From my experience with academic high-performance codes (caveat: I’m not particularly knowledgeable, and game codes would be different in many key ways, so please correct me for whatever things I get wrong), the difficult part of getting the main loop done as fast as possible is that clock speeds have more or less plateaued since a decade ago. We have more CPU cores, threads, SIMD extensions and of course more powerful GPU’s to work with, but a lot of this mostly helps with doing more stuff at once – not the individual elements faster (in academia this is referred to as weak scaling, versus strong scaling). To that, we need to add the cost of communicating between CPU cores, transferring memory back and forth from the GPU, and even setting up data in structures which are more efficient for the different computation elements.

This is very different from earlier PC games like Quake, where in two years your new CPU would run single-threaded work multitudes faster just on account of your clock speed increase (although even this is simplified). Modern games do a lot more processing and on top of that have to figure out extremely efficient ways of setting up and distributing the work load to all available computation elements. This is why it seems to me like an impressive feat to run a modern engine at extremely high framerates, and getting that delta-t for a single frame down to one or two milliseconds.

But it is of course true that framerates themselves don’t tell the entire story. If you do nothing in a loop, you can of course push through millions of iterations per second. Different engines have different needs and objectives. Which is why, while I’m (amateurishly) impressed by this feat, it doesn’t mean the engine itself is better than others out there. Just that it does what it sets out to do very well.

I’d love a technological deep dive into the structure of the engine and what it does to maximize its use of resources, but of course that can never really happen with modern, proprietary engines.


So what you’re saying is that “hell freezing over” are the required conditions.