Gaming Performance

So there’s going to be a lot of interest as to how this performs in our gaming tests, given the heritage of the processor. However, as previously mentioned, there are three things that are going to be against us here.

First is the driver stack. On a console the top to bottom software stack is optimized for both performance and ease of use. Game engine creators and game developers can both work to a fixed set of hardware, and take advantage of how close to the metal that software stack can be; this is why we get such great looking games as the lifecycle of a console continues. By contrast, our system has a straight forward version of Windows 10. It is as generic as it gets, which means optimizations will be on a much lower scale.

Second are the drivers themselves. There is no up-to-date solution here; our system shipped with beta versions of Adrenaline 17.12, which indicates we have December 2017 drivers. None of AMD’s regular driver packages will recognize this system as it uses a custom embedded processor. Some games will refuse to run because the drivers are so old. As a result we’re stuck in the services with a flat tire and no rescue in sight.

Third is the lack of additional eSRAM to help with memory bandwidth. The Xbox One and One S consoles had 32 MB of SRAM plus DDR3, while the Xbox One X had no SRAM but GDDR5. The A9-9820 APU has neither, instead relying on DDR3, and slow DDR3 at that. Memory bandwidth would appear to be a very obvious bottleneck in this regard, assuming that the graphics cores have plenty to work on.

Gaming Results

With all that being said, here are our numbers, and we’re putting them up against some of the very basic competition from our database. Perhaps the best modern comparison point will be to the Ryzen 5 2400G, however we also have a Ryzen V1605B here, which is a 12 W embedded Zen processor with Vega 8 graphics. On the Intel side, I have the Core i5-6500U, a mid-range Skylake mobile processor used in many mini-PCs. 

All of our games here are running at 720p minimum settings or lower, and the numbers will show you why.

Benchmark Results
AnandTech   Chuwi
Aerobox
Ryzen 5
2400G
Ryzen
V1605B
Core i5
6500U
Frames Per Second Averages
Civilization 6 480p Min 24.4 91.2 52.9 35.7
Final Fantasy XV 720p Med 20.1 26.8 14.2 35.4
World of Tanks 768p Min 144.7 223.8 141.1 165.8
Borderlands 3 360p VLow 31.3 70.8 42.9 29.0
Far Cry 5 360p Low 31.5 58.0 25.5 19.0
GTA 5 720p Low 37.8 83.0 52.9 32.8
95th Frame Time Percentiles (shown as FPS)
Civilization 6 480p Min 17.1 57.6 34.8 26.8
Final Fantasy XV 720p Med 17.1 22.6 11.3 6.8
World of Tanks 768p Min 40.2 130.7 84.5 115.2
Borderlands 3 360p VLow 24.2 55.2 32.7 22.3
Far Cry 5 360p Low 26.0 49.0 21.5 16.0
GTA 5 720p Low 25.4 56.6 38.3 23.3

In games like Civilization where the CPU matters, and in some of the other numbers, the poor performing Jaguar cores show how bad it can get – that low World of Tanks percentile comes into playm scoring only 40 FPS. If it weren’t for the CPU, the A9-9820 would be comfortably ahead of the i5-6500U in all of the tests. Games that didn’t run due to driver issues included F1 2019, Gears Tactics, and Red Dead Redemption.

From a personal experience perspective, I set myself up with a wired Xbox controller, and I very comfortably played several hours of Borderlands 3 single player at 720p Ultra Low settings. Frame rates hovered around the 30s, dipping into the 20s during firefights, or up in the 40s when walking through open spaces or in the towns.

Chuwi Aerobox: Under The Hood CPU Benchmarks, Power, Temperature, Noise
Comments Locked

101 Comments

View All Comments

  • azfacea - Thursday, December 24, 2020 - link

    sounds like a solution w/- a problem. much more expensive than raspberry pi and other SBC for pretty much all things CPU . I guess it has more GPU but also way higher power and the GPU is nothing amazing in 2020. so whats the point ??
  • Fulljack - Thursday, December 24, 2020 - link

    it's literally for a niche market in Japan as the article has been stated
  • brucethemoose - Thursday, December 24, 2020 - link

    Can y'all run some GPU compute benches?

    This could potentially be used as a nice budget renderfarm board, especially if the framebuffer can be bumped up. Also, I bet CPU <-> GPU transfers are pretty fast.
  • Glock24 - Thursday, December 24, 2020 - link

    Can it use higher clocked DDR3 RAM? For example DDR3-1866?
  • MrCommunistGen - Friday, December 25, 2020 - link

    Yeah, I'd be really interested to see if performance increases appreciably using some faster RAM.

    I only have experience with two AMD APUs from this era, but both are in prebuilt system scavenged from my company's eWaste pile:

    1. A cheapo Acer tower with an A10-7800 (Kaveri) which heavily throttles the CPU cores if there is any 3D load -- causing a CPU bottleneck in many cases. It shipped with DDR3-1600. I bought some Crucial JDEC spec'd DDR3L-1866 (no XMP profile required) to see if that'd help at all, but the system seems to be capped to 1600MHz despite the A10-7800 supporting DDR3-2133.

    2. An even more cheapo HP "tower" which ended up being an empty box with an external power brick and a mobile A8-6410 (Beema) on an ITX motherboard. Despite Beema officially supporting DDR3L-1866, this machine is also capped to 1600MHz.

    All of this is to say, I wouldn't at all be surprised that even if you paired the A9-9820 with faster DDR3 sticks it might not actually run at higher memory speeds.
  • boozed - Thursday, December 24, 2020 - link

    Apologies in advance...

    RRRAARRWHHGWWR!
  • Bigos - Thursday, December 24, 2020 - link

    Does it run Linux?
  • abufrejoval - Thursday, December 24, 2020 - link

    Android x86 should be fun to try!
  • Oxford Guy - Thursday, December 24, 2020 - link

    'Why console processors have never made it into the PC market before'?

    Because Jaguar has worse IPC than even Piledriver, right?

    'At the time AMD had two CPU core designs that were worth some merit: Bulldozer cores, with its 1 core/2 backend design that has since been branded a large dumpster fire for the company, or Jaguar cores, aimed more for the low power/high efficiency market. Weighing in the performance target of this generation of consoles, both companies decided on having eight Jaguar cores.'

    Or Jaguar cores, aimed at making more money for AMD because the die size is smaller -- not because the CPU is any better than Piledriver. On the contrary.

    The parasitic 'consoles' drained life out of PC gaming with these terrible cores. There is no good reason to have two artificial x86 walled gardens in addition to the PC gaming platform. Yes, there are reasons but none of them outweigh the cost, from the point of view of the consumer rather than companies like Sony, MS, and AMD.

    These "consoles" actually put AMD in the position of competing against the PC gaming platform. So much for the common unjustified belief that AMD is some kind of white knight on a horse, fighting the good fight for the PC gamer. Instead, the company was content to feed us Polaris, a weak design from the start designed to make money by keeping die size small, and lame GPUs like Radeon VI. The transistor count in that one is nothing to write home about. All the marveling about how much performance AMD has been able to get on 7nm since then generally neglects to mention that Radeon VI was not a high-performance design in the first place.
  • Alistair - Thursday, December 24, 2020 - link

    Polaris is one of the best GPUs ever made. Relax. The CPU might have been bad but the GPUs in the PS4 and later Polaris were both great.

Log in

Don't have an account? Sign up now