Starfield PC Performance: How Much GPU Do You Need?

Starfield screenshots
(Image credit: Bethesda)

Starfield — Introduction

Starfield hardly needs an introduction. The latest massive RPG from Bethesda, the expectations are that it will basically deliver "Fallout in space," or alternatively, "Elder Scrolls in space." Even if that's all it provided, people would likely flock to it, but it's supposed to be far more than that. We'll leave the reviewing of the game to others, but we want to see how it runs on some of the best graphics cards, plus some other GPUs for good measure.

We have the Starfield Premium edition, courtesy of AMD, which gives us early access for the next several days — the full release will be on September 6. We're running benchmarks similar to what we did with Diablo IV, and we'll update this article with additional performance results over the coming days. Benchmarking takes time. What sort of hardware will you need, and how does the game look and run on a PC? Let's find out.

Latest Videos FromTom's Hardware
Jarred Walton
Senior Editor

Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.

  • kiniku
    AMD holds 3-6% of the PC GPU market compared to Nvidia? Smart move.

    /s
    Reply
  • mhmarefat
    This is a complete disgrace for both Bethesda and AMD. I hope all the money they agreed on was worth it (spoiler: it won't be). Unless some last second patch before the game is released is applied for nvidia and intel users, people are not going to forget this for years to come. GJ AMD and Bethesda for disgracing yourselves.
    You have a something better than Fallout 4 graphics running 30 fps 1080p on 2080?! What is PC gaming industry reduced to?
    Reply
  • Dr3ams
    As I have written in other threads, my mid-range specs (see my signature) play Starfield on ultra at 1440p smoothly. No glitches or stuttering whatsoever. I did have to dial the Render Resolution Scale down to high (which is 62%), because of some shadow issues. In New Atlantis, wandering outdoors, the FPS averaged between 45 and 50. The FPS is higher when you enter a building that requires a load screen. I have tested the game in cities, combat and space combat...again, no glitches or stuttering.

    There are some clipping issues that should be addressed by Bethesda. I also have some observations about things that should be changed in future updates.

    Third person character movement looks dated. I play a lot of Division 2 (released 2019) and the third person character movement is very good.
    Taking cover behind objects in Starfield is poorly thought out.
    Ship cargo/storage needs to be reworked. In Fallout 4, you can store stuff in just about every container that belongs to you. Inside the starter ship, Frontier, all the storage spaces can't be opened. The only way to store stuff is by accessing the ship's cargo hold through a menu. No visual interaction whatsoever. I haven't built my own ship yet, so maybe this will change later.
    Ship combat needs some tweaking. It would be nice to have a companion fly the ship while the player mounts a side gun.
    Movement. The same issue that Fallout has is showing up in Starfield. Walking speed is OK, but running in both the speeds provided is way to fast. This is especially frustrating when trying to collect loot without slow walking through a large area. In settings, there should be a movement speed adjustment. The only way I know to adjust this, is through console commands.I do love this game though and will be spending a lot of hours playing it. :p
    Reply
  • Makaveli
    lol its nice to have a game optimized for your hardware at launch we usually have the opposite on the red team.

    People need to relax it will improve with game patches and driver updates.

    as per TPU forum.

    "DefaultGlobalGraphicsSettings.json" you get the following predetermined graphics list.

    {"GPUs":









































































































































































































    }
    Reply
  • -Fran-
    Thanks a lot Jarred for this set of tests!

    If I may: would you please test CPU core scaling and memory scaling (latency and bandwidth) for Starfield? I have a suspicion you may find interesting data in there to share.

    Also, a special look at the nVidia CPU driver overhead vs AMD. I have a suspicion some of the AMD favouring comes from that? I'd love to see if that hunch is on the money or not :D

    Regards.
    Reply
  • JarredWaltonGPU
    -Fran- said:
    Thanks a lot Jarred for this set of tests!

    If I may: would you please test CPU core scaling and memory scaling (latency and bandwidth) for Starfield? I have a suspicion you may find interesting data in there to share.

    Also, a special look at the nVidia CPU driver overhead vs AMD. I have a suspicion some of the AMD favouring comes from that? I'd love to see if that hunch is on the money or not :D

    Regards.
    The problem is that I only have a limited number of CPUs on hand to test with. I could "simulate" other chips by disabling cores, but that's not always an accurate representation because of differences in cache sizes. Here's what I could theoretically test:

    Core i7-4770K
    Core i9-10980XE (LOL)
    Core i9-9900K
    Core i9-12900K
    Ryzen 9 7950X

    I might have some other chips around that I've forgotten about, but if so I don't know if I have the appropriate platforms for them. The 9900K is probably a reasonable stand-in for modern midrange chips, and I might look at doing that one, but otherwise the only really interesting data point I can provide would be the 7950X.

    You should harass @PaulAlcorn and tell him to run some benchmarks. 🙃
    Reply
  • steve15180
    I find this article interesting for a bit of a different reason. I think the common belief is that Microsoft has written DX12 ( or 11, etc.) as a universal abstraction layer for any GPU to use. As was pointed out Nvidia has the lion's share of the market. So, my question is, is DX12 written so generically that neither card would work well without modifications on the part of the game developer? Since AMD is a game sponsor and assisted with development, it's only fair that the work with AMD to ensure their cards run well. Nvidia does the same with their sponsored titles.

    Here's what I'd really like to know. Do we really know who's camp has the better card from an unbiased point of view? One of the complaints I read was complaining that the developer optimized for 3 to 6% of the market. Well they helped the developers through an agreement. Just like the Nvida programs of the same type. So, when we benchmark for which card is best, is one truly better than the other? Outside of "the way it's meant to be played", and AMD's program, do developers develop on mostly Nvidia hardware because that's what's more popular? Do game wind up running better on Nvidia hardware because their used more often for development? If both companies were to put the same effort into both cards to maximize for best performance, how would that look on a benchmark? (Also excluding and deliberate programming which hnders one side vs the other.)

    I don't know the answers, but it would be interesting to find out.
    Reply
  • -Fran-
    JarredWaltonGPU said:
    The problem is that I only have a limited number of CPUs on hand to test with. I could "simulate" other chips by disabling cores, but that's not always an accurate representation because of differences in cache sizes. Here's what I could theoretically test:

    Core i7-4770K
    Core i9-10980XE (LOL)
    Core i9-9900K
    Core i9-12900K
    Ryzen 9 7950X

    I might have some other chips around that I've forgotten about, but if so I don't know if I have the appropriate platforms for them. The 9900K is probably a reasonable stand-in for modern midrange chips, and I might look at doing that one, but otherwise the only really interesting data point I can provide would be the 7950X.

    You should harass @PaulAlcorn and tell him to run some benchmarks. 🙃
    I would love to see how the 12900K behaves with different RAM if I had to pick out of that list. Please do consider it? :D

    I hope Paul humours our data hunger and gives us some of his time for testing the CPU front on a more complete way.

    Still, whatever comes out of this, I'm still thankful for the data thus far.

    Regards.
    Reply
  • mhmarefat
    steve15180 said:
    I find this article interesting for a bit of a different reason. I think the common belief is that Microsoft has written DX12 ( or 11, etc.) as a universal abstraction layer for any GPU to use. As was pointed out Nvidia has the lion's share of the market. So, my question is, is DX12 written so generically that neither card would work well without modifications on the part of the game developer? Since AMD is a game sponsor and assisted with development, it's only fair that the work with AMD to ensure their cards run well. Nvidia does the same with their sponsored titles.

    Here's what I'd really like to know. Do we really know who's camp has the better card from an unbiased point of view? One of the complaints I read was complaining that the developer optimized for 3 to 6% of the market. Well they helped the developers through an agreement. Just like the Nvida programs of the same type. So, when we benchmark for which card is best, is one truly better than the other? Outside of "the way it's meant to be played", and AMD's program, do developers develop on mostly Nvidia hardware because that's what's more popular? Do game wind up running better on Nvidia hardware because their used more often for development? If both companies were to put the same effort into both cards to maximize for best performance, how would that look on a benchmark? (Also excluding and deliberate programming which hnders one side vs the other.)

    I don't know the answers, but it would be interesting to find out.
    DX12 is not universal, it is for Windows OS only. It is (or used to! welcome to 2023 gaming industry where games are released half-finished and have ZERO optimization unless you pay the optimization tax) Game Developer's job to make sure their game runs well on majority of GPUs. It is true that AMD sponsored this game but nowhere "fair" that it should run completely BROKEN on other GPU brands. It is truly shameful. We do not see nvidia sponsored games run trash on AMD GPUs do we? What makes sense is that It should run well on all GPUs but better on the sponsored brand. Current state is a greed-fest predatory behavior.
    Developers mostly optimize for AMD (RDNA 2) because consoles are RDNA 2 based too.
    Reply
  • Dr3ams
    mhmarefat said:
    DX12 is not universal, it is for Windows OS only. It is (or used to! welcome to 2023 gaming industry where games are released half-finished and have ZERO optimization unless you pay the optimization tax) Game Developer's job to make sure their game runs well on majority of GPUs. It is true that AMD sponsored this game but nowhere "fair" that it should run completely BROKEN on other GPU brands. It is truly shameful. We do not see nvidia sponsored games run trash on AMD GPUs do we? What makes sense is that It should run well on all GPUs but better on the sponsored brand. Current state is a greed-fest predatory behavior.
    Developers mostly optimize for AMD (RDNA 2) because consoles are RDNA 2 based too.
    FSR will run on Nvidia GPUs. And Starfield doesn't run "completely broken" on other non AMD GPUs, with the exception of Intel. But Intel will be releasing a patch for that.
    Reply