AMD Ryzen 9 3900 Review: a Taste of Eco Mode

Pre-built systems just took a big step forward.

(Image: © Tom's Hardware)

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.

Test Notes 

We list the test system particulars at the bottom of the page. All Ryzen entries with "PBO" indicate an auto-overclocked configuration with DDR4-3600 memory. Intel's overclocked configurations also use DDR4-3600. 

It's notable that widely-available motherboard firmwares enable the Ryzen 9 3900, but they aren't tuned specifically for the processor. OEMs will likely add further system-specific tuning for their respective platforms, but these results represent a good measure of expected performance. 

VRMark and 3DMark

The 3DMark DX11 and DX12 tests indicate the amount of raw horsepower exposed to game engines, but most game engines don't scale as linearly with additional compute resources. 

The Ryzen 9 3900 impresses at stock settings with more performance than the Ryzen 7 3700X, but within a similar power envelope. Uncorking the power limits via the PBO feature exposes performance that is comparable to the 3900X, and even outstrips its counterpart in the DX11 tests. Again, we do expect some differences in performance between the two, as silicon quality varies. AMD's processors also feature a mix of fast and slow cores, with the frequency range of the slower cores being undefined. That means we could see greater variation between samples than we're accustomed to. 

The same trend plays out with the VRMark test sequence, though Intel's chips leverage their per-core performance (a mixture of frequency and IPC throughput) advantage to take a commanding lead. 

Civilization VI AI and Stockfish

Civilization VI prizes per-core performance, so it isn't surprising to see the Intel chips take the lead in this test of AI engine performance in a turn-based strategy game. We can also see that the engine doesn't scale as well with AMD's architecture, as the 105W 3900X provides a slim advantage over the stock 3900 at 65W. Performance between the two chips is similar after tuning. 

Stockfish, an open-source chess engine, is designed to extract the utmost performance from many-core chips by scaling well up to 512 cores. That scalability plays well to AMD's threaded heft and allows the Ryzen 9 3900X to deliver more performance than the 3900 due to its higher power envelope. The stock Ryzen 9 3900 once again proves impressive compared to the stock Ryzen 7 3700X. 

Ashes of the Singularity: Escalation

Ashes of the Singularity: Escalation finds the stock Ryzen 9 3900X with a slight lead over the 3900, and it retains that slight lead after tuning. The Ryzen 7 3700X beats both 39xx models at stock settings, though the 3900X pulls slightly ahead after overclcoking. 

Civilization VI Graphics Test

The tuned 3900 achieves parity with the 3900X, but the benchmark slightly favors the 3900X at stock settings. It's important to remember that these deltas would shrink appreciably with higher resolutions. 

Dawn of War III

The Warhammer 40,000 benchmark responds well to threading, but it's clear that Intel's clock speed advantage has an impact. 

Far Cry 5 

The Ryzen 9 3900 takes a 0.3 fps lead at stock settings over the 3900X in the Far Cry 5 benchmark, but this is essentially a tie. After tuning, the two tie again. 

Final Fantasy XV

We run this test with the standard quality preset to sidestep the impact of a bug that causes the game engine to render off-screen objects with the higher-resolution setting.

Here we see the Ryzen 9 3900 trail the 3900X at stock settings, but it notches a slight win over the 3900X after tuning. 

Grand Theft Auto V

Grand Theft Auto V continues to be popular six long years after its release. This title favors Intel architectures and, more generally, multi-core designs with high clock rates. The Ryzen 9 3900X offers more performance at both stock and overclocked settings than the 3900, proving there are some cases where the more expensive model offers a performance advantage. 

Hitman 2

The same trend emerges during the Hitman 2 benchmark, but it's important to remember that these performance deltas are slim. 

Project Cars 2

Although Project CARS 2 is purportedly optimized for threading, clock rates obviously affect this title's frame rates. Intel's per-core performance advantage pays big dividends in this title and the -9900K and -9700K take a commanding lead over the Ryzen competition, particularly after tuning. 

World of Tanks enCore

The World of Tanks benchmark doesn't hold any surprises, with performance trending within our expectations. 

Test Setup

Swipe to scroll horizontally
AMD Socket AM4 (X570)AMD Ryzen 9 3900 and 3900X, Ryzen 7 3700X
Row 1 - Cell 0 MSI MEG X570 Godlike
Row 2 - Cell 0 2x 8GB G.Skill Flare DDR4-3200
Row 3 - Cell 0 2x 8GB G.Skill FlareX DDR4-3200 - Stock: DDR4-3200, OC: DDR4-3600
Intel LGA 1151 (Z390)Intel Core i9-9900K, i7-9900K
Row 5 - Cell 0 MSI MEG Z390 Godlike
Row 6 - Cell 0 2x 8GB G.Skill FlareX DDR4-3200 - Stock: DDR4-2666, OC: DDR4-3600
All SystemsNvidia GeForce RTX 2080 Ti
Row 8 - Cell 0 2TB Intel DC4510 SSD
Row 9 - Cell 0 EVGA Supernova 1600 T2, 1600W
Row 10 - Cell 0 Windows 10 Pro (1903 - All Updates)
CoolingCorsair H115i

CONTENT

1: Cramming 12 Cores & 24 Threads into a 65W TDP

2: AMD Ryzen 9 3900 Gaming

3: AMD Ryzen 9 3900 Application Testing

4: Conclusion

MORE: Best CPUs

MORE: Intel & AMD Processor Hierarchy

MORE: All CPUs Content

Paul Alcorn
Managing Editor: News and Emerging Tech

Paul Alcorn is the Managing Editor: News and Emerging Tech for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.

  • gggplaya
    Why not publish the IDLE power consumption, both just chip and total system consumption?? That's what really matter for people building a Plexbox or DIY NAS box.
    Reply
  • Gillerer
    I believe an AMD representative has stated that AMD truncates the advertized clock speeds - removing the (mostly irrelevant) +50 MHz; That's why a CPU whose actual boost frequency is 4.35 GHz has a "4.3 GHz" printed on the box and in marketing materials. I think the rationale was along the lines of "it's prettier" or "it's easier to take in", or something.

    Due to the variations in motherboards, the 10 MHz digit isn't significant anyways; just a 100.5 MHz base clock would add +21.5 MHz to CPU speed at a 43 multiplier.
    Reply
  • phiasquiz
    Typo on the first chart? 3700x 8 core/"6" threads
    Reply
  • svan71
    "friendlier 65W TDP " ? I'd like a meaner 140 w version at 5GHZ, you can keep the friendlier versions.
    Reply
  • iam2thecrowe
    gggplaya said:
    Why not publish the IDLE power consumption, both just chip and total system consumption?? That's what really matter for people building a Plexbox or DIY NAS box.
    Agree, publishing idle power is relevant, but anyone building a plexbox/nasbox/HTPC is going to opt for a much lower end CPU.
    Reply
  • gggplaya
    iam2thecrowe said:
    Agree, publishing idle power is relevant, but anyone building a plexbox/nasbox/HTPC is going to opt for a much lower end CPU.

    Not for me, I want more cores for transcoding. It takes about 2000 passmarks for every CPU encoded stream, so the 3900x can do 15 simultaneous streams. The 3900 will presumably do around 12 streams give or take.

    GPU encoding looks terrible in Plex at low bitrates, like 3-6mbps, with tons of macroblocking on fast scenes. CPU encoding looks nice and clean at those bitrates. It might downgrade to 720p at those bitrates, but at least it's a clean image.

    Another thing people do is use these processors as servers for virtual machines, again with the computers running 24/7. Idle power consumption matters again in this case as places like california have expensive electricity.

    Also, some people can only have 1 computer, and not have a dedicated plexbox or NAS server. So they use their computer for gaming, work, video editing etc.... But leave it turned on 24/7 to act as a plex or file server. My coworker does this.

    It would be awesome when reviewing CPU's in the future to have idle power consumption and the passmark score. It's something reviewers have been neglecting lately on most websites, and hard information to find.
    Reply
  • PaulAlcorn
    gggplaya said:
    Not for me, I want more cores for transcoding. It takes about 2000 passmarks for every CPU encoded stream, so the 3900x can do 15 simultaneous streams. The 3900 will presumably do around 12 streams give or take.

    GPU encoding looks terrible in Plex at low bitrates, like 3-6mbps, with tons of macroblocking on fast scenes. CPU encoding looks nice and clean at those bitrates. It might downgrade to 720p at those bitrates, but at least it's a clean image.

    Another thing people do is use these processors as servers for virtual machines, again with the computers running 24/7. Idle power consumption matters again in this case as places like california have expensive electricity.

    Also, some people can only have 1 computer, and not have a dedicated plexbox or NAS server. So they use their computer for gaming, work, video editing etc.... But leave it turned on 24/7 to act as a plex or file server. My coworker does this.

    It would be awesome when reviewing CPU's in the future to have idle power consumption and the passmark score. It's something reviewers have been neglecting lately on most websites, and hard information to find.

    You're right, I feel it is a glaring omission, but hard to measure correctly. We ARE adding it though in some fashion. I solicited some suggestions here:
    1189735229285707776View: https://twitter.com/PaulyAlcorn/status/1189735229285707776

    feel free to weigh in either here or there with ideas of what you would like to see. Sorry for the delay on adding idle power, things have been...busy.
    Reply
  • gggplaya
    PaulAlcorn said:
    You're right, I feel it is a glaring omission, but hard to measure correctly. We ARE adding it though in some fashion. I solicited some suggestions here:
    1189735229285707776View: https://twitter.com/PaulyAlcorn/status/1189735229285707776

    feel free to weigh in either here or there with ideas of what you would like to see. Sorry for the delay on adding idle power, things have been...busy.

    Yea, that would be awesome.

    Personally I would just load windows, install all drivers and updates including motherboard driver updates and bios's. Set it to power savings plan or balanced and not install any programs. Let the machine sit for 45minutes.

    Then measure the average power over say 10 minutes to get the idle power draw.
    Reply
  • bloodroses
    PaulAlcorn said:
    You're right, I feel it is a glaring omission, but hard to measure correctly. We ARE adding it though in some fashion. I solicited some suggestions here:
    1189735229285707776View: https://twitter.com/PaulyAlcorn/status/1189735229285707776

    feel free to weigh in either here or there with ideas of what you would like to see. Sorry for the delay on adding idle power, things have been...busy.


    Would using something like Linux (command line or a bare window manager) be an option for testing idle power consumption since as you said Windows is well, Windows? It'd at least give a ballpark figure relative to (I assume) what Windows would use as well and be more consistent. You'd have to of course mention in the test that Linux is being used instead of Windows.

    Another option could be something like Hiren's BootCD.
    Reply
  • King_V
    svan71 said:
    "friendlier 65W TDP " ? I'd like a meaner 140 w version at 5GHZ, you can keep the friendlier versions.
    Lol, I'll admit I didn't see that coming, and it made me laugh!
    Reply