Before I start: I don’t know in what sort of category this question fits, but I thought rendering was the closest one.
My current computer can barely launch UE5 so I have started looking at computer parts for a new rig. But I am a very poor student and is therefore buying every single component used (except PSU). Right now I am looking at an used rtx 4080 and rtx 3090 both priced at 600 dollars. The 4080 has a fan defect but I can deshroud it and use some noctua fans so that wont be a problem. I already have the needed fans so it wont add any cost either.
The most common recommendation for buying GPUs on the UE forums is to get as much VRAM as possible. Sure, but how important is core clock and memory clock frequencies in comparison? I could not really find any information on this. The reason I am wondering is because the rtx 4080 and rtx 3090 differ in these ways. The 3090 usually has a avg. clock frequency in the range 1800-1900MHz and a memory frequency of around 1200MHz. The rtx 4080 usually has a avg. core clock frequency in the range 2800-2900MHz and a memory clock of around 1400MHz. Data is taken from 3DMark time spy online benchmark stats. But at the same time the rtx 3090 has 24GB of VRAM compared to the 16GB of the rtx 4080 model I am looking at. So if the GPUs are priced the same, what specs do you prioritize? I dont care about the in-game framerates/performance, just what will help me the most when making games in UE5.
Even though I am asking for buying advice between the two cards. I am still very interested in a technical explanation of why one spec should be prioritized over the other.
You should not look at the core clock and memory clocks alone. This will be very misleading. Instead look at the newer architectures as this provides you much better performance improvements that core and memory clocks could ever hope to. A RTX 3070 might perform much better than a GTX 1080ti (don’t quote me on this, its just an example) even though the core clocks are relatively the same. So essentially, your number one priority should be VRAM as you already know, and then comes the GPU Architecture, the newer and the higher tier, the better. If you could provide a budget, I could probably help you select a good card.
If you cannot, I would personally recommend the 4080 (or any lower tier card you wish) as you have mentioned. I say this because of the DLSS 3.0 technology available with them. You can double your framerate with it (which is supported by UE5) but some people do not like how it looks, so this will be dependent on you, if you like it, or wish to sacrifice a little visual fidelity for roughly double the framerate, I’d say this is the best option for you.
clocks don’t matter cause they multiply by the amount of shader cores and memory channels.
the raw specs of the 3090 and 4080
the 4080 has 130% of the general performance of the 3090. teraflops of compute and rop. less but faster rt cores. it has lower memory bandwidth that is compensated with a larger L2 cache. means less frequent memory fetches. not that important for gamedev, but a down to the metal optimization fact.
when it comes to performance and memory capacity you should take the audience into consideration. how many teraflops and how much memory does the average consumer have to run your game.
that’s all the gamedev math you gotta do, really.
dlss and fsr will increase fps or get you to something playable if you target high fidelity and raytracing enabled features. frame gen is an overrated feature tho. it’s just temporal and “ai” generated inbetweens. not a clean render. aka a gimmick to push the fps numbers.
Ok, thank you the help. Dont know if I understood this correctly:
“when it comes to performance and memory capacity you should take the audience into consideration. how many teraflops and how much memory does the average consumer have to run your game.”
So you think having 24GB of VRAM can a bad thing because actually using this much VRAM during development would result in a compiled game that uses very large amounts of VRAM and therefore has very high system requirements?
exactly. you can use it during development. but the average consumer stat should be taken into account. steam hardware stats are a very good helper there. for console deployment the limit for vram is 10-12 GB, depending how you balance cpu and gpu workloads.
Thank you for the help. I cant spend more than 600 dollars on the GPU (but want to save money where I can). I am buying from the used market in Norway which has very different prices than other places, so trying to help me find a GPU would be hard. But thanks for the offer
Ah, okay. I just keep reading about people saying that even their 16GB of VRAM is not enough for their projects, but those might just be extreme cases of people developing for high end platforms. Do you know if there is like a general ratio of editor VRAM usage compared to compiled game VRAM usage? For example that the compiled game will generally use about half the VRAM of the editor. So a 10GB VRAM usage in the editor will become 5GB in the finished game.
i dunno about nanite. i don’t use it yet for that very reason that it’s not widely hardware supported on older amd cards. for the core engine in a template map i did a test for a question reply a couple days ago. the mesh geometry, texture and shadow load should be the same. the core engine memory load with editor and lumen is ~3.2 GB, while sitting at ~2.2 GB in packaged 1080. not much of a concern, but the overall memory should be within a certain budget, so users don’t have to drop scalabilty levels.
Having a GPU with a lot of VRAM does not automatically make the game resource hungry. You can still have a game use 6gb VRAM while having 24gb total. It all depends on what optimizations you do, how big your textures are, and various other factors. But saying that having extra VRAM will make your game resource intensive is just plain wrong.
Ok, thank you for the reply. I did not mean having 24GB of VRAM will automatically make the game more resource intensive. But that it possibly could be pointless having 24GB vs 16GB because having the game require 24GB in the editor would possibly lead to a compiled game that requires a lot more VRAM than most GPUs support.
You will definitely have more VRAM usage in the editor and that is a good way to test it. You need to package your game, turn your editor off, and any resource intensive applications to get the best case scenario. Seeing your VRAM usage or even FPS in the Editor will be misleading.
