Stop wasting money on GPUs with more than 16GB of VRAM

Are you trying to figure out if you need an RTX 5070, 5080, or 5090? Let me help you decide—don't let VRAM be a factor in your decision-making process. You really don't need more than 16GB of VRAM for most use cases, especially gaming.

1080p and 1440p gaming don't take advantage of high VRAM amounts

Gaming at 4K is overrated

I've been playing PC games for a long time. The first graphics card I purchased for myself was the GTX 970 back in 2016, almost a decade ago. That graphics card only had 3GB of VRAM , and it was seen as a pretty solid mid-range card at the time. Today, some consumer graphics cards can have up to 32GB of VRAM—more than many computers.

The thing is, most gaming is still done at 1080p and 1440p— not 4K . In fact, 74% of Steam users are either at 1080p or 1440p , with over 50% of all Steam users gaming at 1080p. Playing games at lower resolutions simply can't take advantage of high VRAM amounts.

I currently game on an RTX 3080 12GB, and had an RTX 3060 12GB before that. When I had my 1440p monitor, I could max out all settings and still come nowhere close to maxing out my VRAM. On either card, I'd get pretty solid frame rates, but never come close to maxing out the VRAM. I think the most I was ever able to push a card was to around 8GB of VRAM in Ghost Recon Wildlands and Breakpoint .

Now, if you had a card with more than 16GB of VRAM and gamed at 1080p or 1440p? It'd be a complete waste. That extra VRAM wouldn't be used at all. If you're still gaming at 1080p or 1440p, stick with more affordable cards and only upgrade your graphics card when you're ready to get a new monitor.

Upscaling tech produces decent quality with lower specs

Sometimes AI can be good

I'm not a huge fan of frame generation or upscaling when it comes to AI-assisted game rendering, but I can't deny how effective the technology is. These capabilities can help lower-end cards to punch way above their weight class —all without extra hardware.

If you have a newer NVIDIA or AMD graphics card, then you already have access to technology like DLSS, DLAA, FSR , and all the other things that can help generate both frames and upscaled graphics assets. I'm honestly becoming impressed with the technology, and I can barely tell a difference of whether it's on or not when gaming.

Regardless of whether I like it or not, though, being able to upscale a game's quality without having to render higher-quality assets means you can play at higher resolutions without having to have more VRAM. You see, it's the 4K or 8K graphics assets that take up your VRAM, and if you're upscaling lower-quality assets to a higher-resolution, it just doesn't take up the same amount of resources.

Even at 4K, it's hard to max out more than 16GB of VRAM

Spend your money elsewhere

On my RTX 3080 12GB, I still find it hard to max out the VRAM when gaming even at 4K. I've definitely come close in some titles, but I still have yet to fully max out the VRAM in most games. If I'm not maxing out the VRAM on a 12GB card, you likely won't be maxing out the VRAM in a 16GB card, either.

I know, I know, if you're anything like me, then you want to make sure you have enough hardware so that way you never max it out—and I get it. The problem is, going with much more than 16GB of VRAM won't give you a solid return on your investment, even if you game at 4K. There are plenty of ways to improve your FPS or gaming experience using software without buying an expensive new graphics card.

I'd recommend saving the cash you were planning on spending on a higher-end graphics card and putting it toward another part of your computer. That could be more RAM, additional storage, or a better processor (though you probably don't need a high-end processor like you think you do).

The main use case for more than 16GB of VRAM is AI, not gaming

If you aren't running massive local AI models, save the cash

There is one use case where having more than 16GB of VRAM does come in handy—AI workloads. If you're running local large-language models (LLMs) or other types of AI workers, then you're going to want all the VRAM you can get . In fact, 32GB might not be enough, and you might need multiple graphics cards to effectively carry the weight of those heavy AI workloads.

However, you can still get by with 16GB or less of VRAM even when using AI locally . I use AI all the time on my RTX 3080, and it works perfectly fine. Is it super fast? No, but it does work, and that's enough for me.

---

At the end of the day, buy a graphics card with as much VRAM as you want—just know that buying a RTX 5090 with 32GB of VRAM to game at 1440p will be a gigantic waste of money.