Apple Mac Studio (M5 Ultra) review: Local model citizen outpaces DGX Spark and Threadripper

Taking on the Mac Pro's mantle

Apple Mac Studio (M5 Ultra)
Editor's Choice
(Image credit: © Tom's Hardware)

Tom's Hardware Verdict

The Mac Studio with M5 Ultra is a small powerhouse PC with strong performance, fast memory throughput, and the ability to run AI models locally (if that's your thing).

Pros

  • +

    M5 Ultra delivers powerful performance

  • +

    1.2 TB/s memory throughput is excellent for local AI models

  • +

    Small, quiet design

  • +

    Plenty of ports

Cons

  • -

    Pricey memory and SSD upgrades

  • -

    Not upgradeable after purchase

Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.

Earlier this year, Apple discontinued the Mac Pro, leaving the Mac Studio as its top-end desktop. With the new M5 Ultra chip, Apple's relatively diminutive system is more than ready to take on the mantle.

While you might not have a ton of PCIe connections and both RAM and storage are set, the Studio offers a ton of ports for accessories, monitors, and peripherals, memory options up to 256GB (with 512GB coming later), and a clean, quiet design.

Latest Videos FromTom's Hardware
TOPICS
Andrew E. Freedman
Senior Editor

Andrew E. Freedman is a senior editor at Tom's Hardware focusing on laptops, desktops and gaming. He also keeps up with the latest news. A lover of all things gaming and tech, his previous work has shown up in Tom's Guide, Laptop Mag, Kotaku, PCMag and Complex, among others. Follow him on Threads @FreedmanAE and BlueSky @andrewfreedman.net. You can send him tips on Signal: andrewfreedman.01

  • Pierce2623
    Was CP2077 run through a translation layer or is there an ARM native version?
    Reply
  • LordVile
    Pierce2623 said:
    Was CP2077 run through a translation layer or is there an ARM native version?
    There is a native ARM version however I’m not sure on apples RT performance which was a strange setting to turn on. I’m also assuming it’s using upscaling on the windows machines based on the 5090
    Reply
  • Bebydis
    Yay, good showing. But it is too expensive, we are reacing 1990s prices. 3d performance is still worse than a 5 year old amd 6900xt card. Not bad at all, but how much is the bad design decisions in macos holding rhe hardware back? The ipc is still a huge bottleneck, but apple utterly outperforms ntfs with the newer Dominic Giampalo designs.
    Reply
  • DS426
    Not even limited to proprietary or limited upgrades but purely non-upgradable!?... nahhhhh. Might as well shape it like the trash can Macs, lol.
    Reply
  • artk2219
    DS426 said:
    Not even limited to proprietary or limited upgrades but purely non-upgradable!?... nahhhhh. Might as well shape it like the trash can Macs, lol.
    Ironically the trashcan macs WERE upgradable, even if they are a pain to work on. Unfortunately that is not the case with this expensive and put out to pasture before its time future ewaste :-/ . Sadly that seems to be the trend nowadays, lack of serviceability is now unfortunately a common theme from phones to cars to farm equipment.
    Reply
  • artk2219
    LordVile said:
    There is a native ARM version however I’m not sure on apples RT performance which was a strange setting to turn on. I’m also assuming it’s using upscaling on the windows machines based on the 5090
    Nope, thats native on a 5090, but its a 5090 and costs like 6 grand on its own currently. Its not a terrible showing if you don't take price into account, its around RTX 5070 \ RX 9070 territory. Intel still doesn't have a card with that sort of performance available yet.

    https://www.techpowerup.com/review/pny-rtx-5080-slim-oc/33.html
    https://tpucdn.com/review/pny-rtx-5080-slim-oc/images/rt-cyberpunk-2077-1920-1080.png
    https://cdn.mos.cms.futurecdn.net/2R8VeJcDuDaZSLNPebdWzV-970-80.png.webp
    Arc B70
    Reply
  • LordVile
    artk2219 said:
    Nope, thats native on a 5090, but its a 5090 and costs like 6 grand on its own currently. Its not a terrible showing if you don't take price into account, its around RTX 5070 \ RX 9070 territory. Intel still doesn't have a card with that sort of performance available yet.

    https://www.techpowerup.com/review/pny-rtx-5080-slim-oc/33.html
    https://tpucdn.com/review/pny-rtx-5080-slim-oc/images/rt-cyberpunk-2077-1920-1080.png
    https://cdn.mos.cms.futurecdn.net/2R8VeJcDuDaZSLNPebdWzV-970-80.png.webp
    Arc B70
    Techspot has the 5090 lower. Your first chart also doesn’t include what settings they’re on
    Reply
  • artk2219
    LordVile said:
    Techspot has the 5090 lower. Your first chart also doesn’t include what settings they’re on
    That is why i included the link for the article above the picture. Though they're not all going to be the same results, different setups and such, regardless, the performance stated out of the 5090 in the review is not strange, it really is just that beefy.

    https://www.techpowerup.com/review/pny-rtx-5080-slim-oc/4.html
    Test SystemTest System - GPU 2025.2Processor:Motherboard:Resizable BAR:Memory:Cooling:Thermal Paste:Storage:Power Supply:Case:Operating System:Drivers:
    AMD Ryzen 7 9800X3DZen 5, 5.2 GHz, 8 cores/16 threads
    MSI X870E Carbon Wi-FiBIOS 7E49v1A3
    Enabled on all supported cards
    (NVIDIA, AMD & Intel)
    Thermaltake TOUGHRAM XG2× 16 GB DDR5-6200 28-36-36-76 UCLK 1:1
    Arctic Liquid Freezer III280 mm AIO
    Arctic MX-64 TB M.2 NVMe SSD
    Seasonic Focus GX 1000 WATX 3.0 / 16-pin 12VHPWR
    darkFlash DY470Windows 11 Professional 64-bit 24H2
    VBS enabled (Windows 11 default)
    RX 9060 XT, RX 9070 & RX 9070 XT: 25.11.1 WHQL
    Other AMD cards: 25.3.2 Beta
    NVIDIA RTX 50 cards: 581.15 WHQL
    RTX 5050: 576.88 WHQL
    Other NVIDIA cards: 572.83 WHQL
    Intel: 101.6732 Beta
    Date of Retest

    Benchmark scores in other reviews are only comparable when this exact same configuration is used.








    All games and cards are tested with the drivers listed above—no performance results were recycled between test systems. Only this exact system with exactly the same configuration is used for all results in this review.
    All graphics cards are tested using the same game version.
    All games are set to their highest quality setting unless indicated otherwise.
    All games are running at their native resolution without upscaling (no DLSS or FSR), except for the results on the "DLSS 4" page where we specifically look into that feature's performance.
    All games have ray tracing disabled, except for testing on the "Ray Tracing" and "Path Tracing" pages.
    AA and AF are applied via in-game settings, not via the driver's control panel.
    Before starting measurements, we heat up the card for each test to ensure a steady state is tested. This ensures that the card won't boost to unrealistically high clocks for only a few seconds until it heats up, as that doesn't represent prolonged gameplay.
    For better real-life applicability, all game tests use custom in-game test scenes, not the integrated benchmarks.
    All cards used for comparison are reference designs. When a reference design does not exist, we go the extra mile and buy the closest possible match, which has reference clocks and default power limit.Each game is tested at these screen resolutions:
    1920×1080: Most popular monitor resolution.
    2560×1440: Intermediary resolution between Full HD and 4K, with reasonable performance requirements.
    3840×2160: 4K Ultra HD resolution, available on high-end monitors.Date of RetestWe retest all our comparison cards every few months, with the latest drivers, new games, patches and OS updates.
    2025-04-01: Arc A770 16 GB
    2025-04-01: Arc B580 16 GB
    2025-04-01: RTX 3060 12 GB
    2025-04-02: RTX 3070 8 GB
    2025-04-02: RTX 3080 10 GB
    2025-04-02: RTX 3090 Ti 24 GB
    2025-04-03: RTX 4060 8 GB
    2025-04-03: RTX 4060 Ti 8 GB
    2025-04-03: RTX 4060 Ti 16 GB
    2025-04-04: RTX 4070 12 GB
    2025-04-04: RTX 4070 Super 12 GB
    2025-04-04: RTX 4070 Ti 12 GB
    2025-04-05: RTX 4070 Ti Super 16 GB
    2025-04-05: RTX 4080 16 GB
    2025-04-05: RTX 4080 Super 16 GB
    2025-04-05: RTX 4090 24 GB
    2025-08-20: RTX 5060 8 GB
    2025-08-19: RTX 5060 Ti 8 GB
    2025-08-19: RTX 5060 Ti 16 GB
    2025-08-18: RTX 5070 12 GB
    2025-08-18: RTX 5070 Ti 16 GB
    2025-08-17: RTX 5080 16 GB
    2025-08-17: RTX 5090 32 GB
    2025-04-07: RX 6800 XT 16 GB
    2025-04-07: RX 6900 XT 16 GB
    2025-04-08: RX 7600 8 GB
    2025-04-08: RX 7600 XT 16 GB
    2025-04-08: RX 7700 XT 12 GB
    2025-04-09: RX 7800 XT 16 GB
    2025-04-09: RX 7900 GRE 16 GB
    2025-04-09: RX 7900 XT 20 GB
    2025-04-09: RX 7900 XTX 24 GB
    2025-11-17: RX 9060 XT 8 GB
    2025-11-17: RX 9060 XT 16 GB
    2025-11-18: RX 9070 16 GB
    2025-11-18: RX 9070 XT 16 GB
    Reply
  • LordVile
    artk2219 said:
    That is why i included the link for the article above the picture. Though they're not all going to be the same results, different setups and such, regardless, the performance stated out of the 5090 in the review is not strange, it really is just that beefy.

    https://www.techpowerup.com/review/pny-rtx-5080-slim-oc/4.html
    Test SystemTest System - GPU 2025.2Processor:Motherboard:Resizable BAR:Memory:Cooling:Thermal Paste:Storage:Power Supply:Case:Operating System:Drivers:
    AMD Ryzen 7 9800X3DZen 5, 5.2 GHz, 8 cores/16 threads
    MSI X870E Carbon Wi-FiBIOS 7E49v1A3
    Enabled on all supported cards
    (NVIDIA, AMD & Intel)
    Thermaltake TOUGHRAM XG2× 16 GB DDR5-6200 28-36-36-76 UCLK 1:1
    Arctic Liquid Freezer III280 mm AIO
    Arctic MX-64 TB M.2 NVMe SSD
    Seasonic Focus GX 1000 WATX 3.0 / 16-pin 12VHPWR
    darkFlash DY470Windows 11 Professional 64-bit 24H2
    VBS enabled (Windows 11 default)
    RX 9060 XT, RX 9070 & RX 9070 XT: 25.11.1 WHQL
    Other AMD cards: 25.3.2 Beta
    NVIDIA RTX 50 cards: 581.15 WHQL
    RTX 5050: 576.88 WHQL
    Other NVIDIA cards: 572.83 WHQL
    Intel: 101.6732 Beta
    Date of Retest

    Benchmark scores in other reviews are only comparable when this exact same configuration is used.








    All games and cards are tested with the drivers listed above—no performance results were recycled between test systems. Only this exact system with exactly the same configuration is used for all results in this review.
    All graphics cards are tested using the same game version.
    All games are set to their highest quality setting unless indicated otherwise.
    All games are running at their native resolution without upscaling (no DLSS or FSR), except for the results on the "DLSS 4" page where we specifically look into that feature's performance.
    All games have ray tracing disabled, except for testing on the "Ray Tracing" and "Path Tracing" pages.
    AA and AF are applied via in-game settings, not via the driver's control panel.
    Before starting measurements, we heat up the card for each test to ensure a steady state is tested. This ensures that the card won't boost to unrealistically high clocks for only a few seconds until it heats up, as that doesn't represent prolonged gameplay.
    For better real-life applicability, all game tests use custom in-game test scenes, not the integrated benchmarks.
    All cards used for comparison are reference designs. When a reference design does not exist, we go the extra mile and buy the closest possible match, which has reference clocks and default power limit.Each game is tested at these screen resolutions:
    1920×1080: Most popular monitor resolution.
    2560×1440: Intermediary resolution between Full HD and 4K, with reasonable performance requirements.
    3840×2160: 4K Ultra HD resolution, available on high-end monitors.Date of RetestWe retest all our comparison cards every few months, with the latest drivers, new games, patches and OS updates.
    2025-04-01: Arc A770 16 GB
    2025-04-01: Arc B580 16 GB
    2025-04-01: RTX 3060 12 GB
    2025-04-02: RTX 3070 8 GB
    2025-04-02: RTX 3080 10 GB
    2025-04-02: RTX 3090 Ti 24 GB
    2025-04-03: RTX 4060 8 GB
    2025-04-03: RTX 4060 Ti 8 GB
    2025-04-03: RTX 4060 Ti 16 GB
    2025-04-04: RTX 4070 12 GB
    2025-04-04: RTX 4070 Super 12 GB
    2025-04-04: RTX 4070 Ti 12 GB
    2025-04-05: RTX 4070 Ti Super 16 GB
    2025-04-05: RTX 4080 16 GB
    2025-04-05: RTX 4080 Super 16 GB
    2025-04-05: RTX 4090 24 GB
    2025-08-20: RTX 5060 8 GB
    2025-08-19: RTX 5060 Ti 8 GB
    2025-08-19: RTX 5060 Ti 16 GB
    2025-08-18: RTX 5070 12 GB
    2025-08-18: RTX 5070 Ti 16 GB
    2025-08-17: RTX 5080 16 GB
    2025-08-17: RTX 5090 32 GB
    2025-04-07: RX 6800 XT 16 GB
    2025-04-07: RX 6900 XT 16 GB
    2025-04-08: RX 7600 8 GB
    2025-04-08: RX 7600 XT 16 GB
    2025-04-08: RX 7700 XT 12 GB
    2025-04-09: RX 7800 XT 16 GB
    2025-04-09: RX 7900 GRE 16 GB
    2025-04-09: RX 7900 XT 20 GB
    2025-04-09: RX 7900 XTX 24 GB
    2025-11-17: RX 9060 XT 8 GB
    2025-11-17: RX 9060 XT 16 GB
    2025-11-18: RX 9070 16 GB
    2025-11-18: RX 9070 XT 16 GB
    Cyberpunks highest quality setting enables DLSS in quality mode. They likely missed it. Techspots results were 152FPS with RT and DLSS enabled. Turning off RT and disabling DLSS put it to over 200 at 1440p. Thats with. near identical system that should be within 5%. So they’re using RT numbers with DLSS set to quality as per the default settings. If they say they’re not then it’s likely they’ve missed the fact that it’s on be default.
    Reply
  • salgado18
    artk2219 said:
    Ironically the trashcan macs WERE upgradable, even if they are a pain to work on. Unfortunately that is not the case with this expensive and put out to pasture before its time future ewaste :-/ . Sadly that seems to be the trend nowadays, lack of serviceability is now unfortunately a common theme from phones to cars to farm equipment.
    Not that I want to defend Apple (I really don't), but to be fair, they only managed to get all that RAM bandwidth because memory is tightly integrated. There's no DDR or GDDR tech that can match those numbers.

    That said, my prediction is that this is the future of PCs: large RAM pools with big bandwith numbers. AI is going nowhere, and will go local as soon as software integrates and hardware makes it viable on high-end (not niche like Mac Studio or DGX, but top-end desktop Intel or AMD platforms).
    Reply