Amazon Fire TV Cube Benchmarks — Faster than the Fire TV 3 but with a catch

Now that the Amazon Fire TV Cube has been released, a lot of people will be deciding between it and the less expensive Fire TV 3 pendant. While they look very different on the outside, thanks to the Fire TV Cube’s hands-free Alexa capabilities, they’re very similar on the inside. They both have the same video playback capabilities, which include 4K @ 60FPS and HDR support. That’s because they both have the exact same 1.5Ghz quad-core CPU, the same Mali-450 MP3 GPU, and the same 2GB of RAM. With identical chips, you’d expect them to perform the same on benchmarks, but it turns out the Fire TV Cube is faster than the Fire TV 3. However, there’s an important catch that throws things off.

For this benchmark comparison, I’m focusing on only the Fire TV Cube and Fire TV 3, since their specs are so similar. I hope to compare these two devices to the Fire TV Stick and past Fire TV models in a future article. Even though the Fire TV Cube and the Fire TV 3 use the exact same chips, my theory, and the reason for running these benchmarks, was that the Fire TV Cube’s larger size would allow for better cooling and, therefore, result in slightly better performance.

Remember, benchmark scores that push devices to their limits do not translate well into real-world performance, since you’ll rarely be pushing these devices this hard, but it’s still an interesting exercise. Even though the Fire TV Cube is slightly faster, you’re not going to see a difference in day-to-day use between it and the Fire TV 3, especially when it comes to watching videos.

I started off by running GFXBench’s suite of GPU tests. This is a good measure of each device’s gaming capabilities. Right off the bat, you can see that the Fire TV Cube scored 601 frames on the offscreen 1080p T-Rex test, while the Fire TV 3 scored 570 frames. A 5% higher score might not seem like much, but it’s fairly significant and a much bigger difference than I was expecting, considering these two devices have the exact same CPU and GPU.

Next, I ran the 3DMark Icestorm benchmark. This mostly tests the GPU but also measures the CPU’s performance using a physics test at the end of the benchmark. The score of 4,173 for the Fire TV Cube and 4,069 is an average of several GPU tests plus the one CPU test. At about 3% higher, this benchmark confirms that the Fire TV Cube does have a slight advantage over the Fire TV 3, as far as raw power goes.

Taking a look at 3DMark’s benchmark chart reveals exactly why the Fire TV Cube is scoring higher than the Fire TV 3. This chart shows CPU clock speed, CPU usage, and FPS values throughout the benchmark test. As you can see, the CPU clock speed (blue line) for the Fire TV Cube is locked to its full 1.5Ghz speed the entire time, while the Fire TV 3’s CPU clock speed is fluctuating drastically during the test.

Investigating further reveals that the Fire TV Cube’s CPU governor is set to “Performance” mode, while the Fire TV 3’s CPU governor is set to “Interactive” mode. The CPU governor configuration of a device is what determines how the operating system scales the frequency of the CPU. Performance mode means the CPU frequency is set to a specific value and will never change, so the Fire TV Cube always runs at 1.5Ghz, regardless of whether the CPU is being used or sitting idle. Interactive mode means the CPU frequency dynamically scales in response to the workload placed on the CPU. When the Fire TV 3 is idle, it lowers the CPU frequency and ramps it up when needed.

The Fire TV 3’s less than ideal CPU governor configuration is likely due to heat management. By keeping the CPU frequency low when it’s not in use, the Fire TV 3 will produce less heat. Since it is in a smaller package, there’s probably less room to dissipate heat, so it’s a bigger concern than with the Fire TV Cube, which probably has better overall cooling. Even though ramping up the CPU frequency on the Fire TV 3 happens nearly instantaneously, the tiny delay is likely why the Fire TV Cube performs better in these benchmarks.

After determining why the Fire TV Cube’s GPU benchmark scores outperform the Fire TV 3’s scores, I moved on to run benchmarks that focus only on the CPU. To my surprise, the Fire TV 3 does notably better when it comes to CPU benchmarks. In GFXBench’s Driver Overhead test, which taxes the CPU, the Fire TV 3 scored 2,614 frames compared to the Fire TV Cube’s score of 2,468. I expected to see similar results for the CPU tests as the GPU tests from early, but instead, there’s about a 6% drop with the Fire TV Cube.

Running the PCMark benchmark confirms the results seen with GFXBench. PCMark stresses the CPU by running various productivity tasks like photo editing and data manipulation. It’s not very relevant to Fire TV usage, but the Fire TV Cube’s score of 2,692 and the Fire TV 3’s score of 2,890 shows the same 6% drop for the newer Fire TV model.

After re-running all the benchmarks and re-examining the 3Dmark charts from above, I finally noticed the cause of the unexpected CPU performance drop with the Fire TV Cube. Looking at the CPU usage lines (purple) in the 3DMark charts, you can see that both devices go to around 100% CPU usage during the physics portion of the benchmark. This is expected, since that test stresses the CPU, but when looking at the CPU usage during the GPU tests you can see the Fire TV Cube’s CPU usage is notably higher than the Fire TV 3’s CPU usage.

I enabled the System X-Ray utility to view the CPU usage in real-time and noticed that one of the 4 cores on the Fire TV Cube was always at around 50% usage while idle. All 3 other cores on the Fire TV Cube, as well as all 4 cores on the Fire TV 3, hover around 25% usage while idle.

Checking the list of running processes on the Fire TV Cube revealed the culprit to be a package called “amazon.speech.wakewordservice,” which I assume is what the Fire TV Cube uses to detect the Alexa wake word. This means that the Fire TV Cube uses approximately 6% of its processing power (25% of one CPU core) to listen for hands-free commands. That 6% extra load correlates perfectly with the 6% lower CPU benchmark scores.

While the Fire TV Cube is capable of performing better than the Fire TV 3, despite having identical hardware, thanks to its better CPU governor and heat dissipation, you’ll only see those advantages when utilizing the GPU because it needs to use about 6% of its CPU for its hands-free Alexa feature. That overhead is obviously not found on the Fire TV 3 or any other Fire TV device. For those wondering, no, muting the microphones does not clear up that 6% usage. I wouldn’t expect it to since the Fire TV Cube still needs to be ready to respond instantly the moment you unmute its mics.

Ultimately, these benchmark scores aren’t different enough in either direction to translate into real-world benefits or handicaps for either the Fire TV Cube or the Fire TV 3, and so shouldn’t really be considered when making a buying decision.

ShareTweetShare+1

34 comments
  1. Steve says:

    Many thanks for these, looking forward to the analysis with the other models (especially the Gen 2 FTV).

    And benchmarks are a good place to start when evaluating what you want to buy, but as you say the raw numbers rarely translate into viewable differences since people use these boxes in different ways.

    I would like to know if the old limitation of a max of 20 Mbps is universal for all the FTV including the new Cube if you know.

    Thanks

    • AFTVnews says:

      You’re welcome! What 20 Mbps limitation are you talking about? If you’re asking about network speeds, a speed test through my Google WiFi router is reporting 348Mbps to the Fire TV Cube over WiFi.

      • Smitty says:

        I think he’s referring to bitrate limitation if your playing local media for example.

        I don’t know what the FTV2 bitrate limit is, but its played all the high bitrate 4K rips I’ve thrown at it.

        • Hegemon says:

          Thats very interesting I’ve noticed on some of my larger videos that the FTV will stutter in emby until I manually set the bitrate lower perhaps this is the cause.

  2. Charlie says:

    Interesting and I assume software adjustments to both devices to change these values are possible and all this isn’t “hard-wired.”

    I imagine there will be a flurry of updates to the cube in the near future. I’m feeling only slightly jealous of you Cube owners.

    • AFTVnews says:

      I can’t say for sure, but I assume the CPU governor can be changed with software like on most Android devices. Amazon would have to do it, which I doubt they would after the fact, or users can do it with root. Although, if the Fire TV 3 gets rooted, I don’t see a 5% performance gain worth the risk of overheating/damaging the device.

  3. OG Charlie says:

    Could you test how it handles games? I’m curious since it has double the internal storage than the Fire TV 2.

    I imagine it has no problem with older games but can it handle more extreme games like Star Wars: KOTOR, Grand Theft Auto: SA, Shovel Knight, Oddworld: Strangers Wrath? I remember the Fire TV 1 having difficulty with Oddworld.

    • AFTVnews says:

      What I really want to do is come up with a real-world set of repeatable tests, to compare all the models together, that reflects real usage, with gaming included. I don’t know how feasible that will be or when I’ll find the time to do it.

      You can see all the games that the Fire TV Cube currently supports here (but I know support does not always mean a good experience): http://bit.ly/firetvcubegames

      Of the games you listed, the Cube does not currently support GTA: SA and Oddworld: Strangers Wrath. But interestingly it does support GTA 3, GTA: VC, and Oddworld: Munch’s Oddysee.

      • OG Charlie says:

        I wonder if it’s the same issue the launch of the Fire TV 2 had where developers had to check a box to show that it was compatible with the new device.

        But maybe it doesn’t have the umpf for those games. GTA: SA is way bigger than the previous games and Oddworld: Stranger’s Wrath brought my Fire TV 1 to its knees.

        Thanks for the reply!

  4. David Scarpa says:

    Please test the Wifi , i had a pendent and when sitting right next to my Apple TV 4k, it would recieve the same wifi network at about half the rate and it would’nt stream anything above 1080p, i eventually returned it

    • AFTVnews says:

      I’m seeing WiFi speeds consistently above 300Mbps, but I didn’t have any issues with the Fire TV 3 pendant either.

      • clocks says:

        At what approx distance are you getting these >300meg speeds? This is with Google mesh wifi?

        • AFTVnews says:

          Yes, the Google Wifi Mesh. THe access points are only a few feet from the Cube because my house is pretty overkill when it comes to WiFi, so it’s impossible to get too far away from an access point. Google says 1 access point is sufficient for a 1,500 sq.ft. house. My house is 1,000 sq. ft. and I have 3 access points that are all hard-wired with gigabit ethernet.

          • clocks says:

            That’s interesting, as I didn’t think that Google mesh generally did much better than 150meg. Maybe I need to try it. You’ll have to update the gear section of your website, since I think you list a LInksys router there. :-)

            BTW – I sent an email link earlier today, though I know you are busy with the cube. But two more Prime Phones came onboard today.

          • AFTVnews says:

            Thanks. I completely forgot about the gear page. I went through and updating things.

            And thanks for the heads up on the phones. I did see the email but just haven’t had time to cover it.

  5. Alex says:

    Kicking myself for missing the preorder sale. Hopefully it drops down to pre sale prices on prime day.

  6. clocks says:

    I love these performance comparisons, done using true benchmarks. I had hope the cube would be able to perform a tiny bit better than FTV3 due to its larger size, but I did not account for the Alexa overhead. I wonder if they can reduce that 6% over time.

    I’d like to see a comparison of wifi between the two devices and also FTV2. I’m hoping with the cubes larger size, that maybe they were able to put more/better antennae in it.

  7. Randy says:

    The CHANNELS – LIVE TV app look awful on the Cube as compared to FTV2
    I loved Channels – Live TV on my FTV2 but it is unwatchable on the new FTVCube
    If you can help me out, that would be great; however, I have not found any software switch to fix it.

  8. Gareth Price says:

    Great analysis as usual Elias. I’m figuring keeping the cooling levels to the sweet spot is the reason Amazon kept with the adaptor rather than built in Ethernet.

  9. BobR says:

    Excellent insights. I was a bit concerned about the processing load due to the microphones, which I don’t intend to use since my cube will be in a cabinet. I think real-world tests need to consider unknown future software/tech changes so benchmarks can be very useful. Comparing them to new major OS versions is also helpful. My first gen stick performed great at first, but now almost 4 years later it takes 20 (twenty) sec to load Netflix from the Home screen; even after factory resets, etc. My FTV1 takes only 2 sec and the new Cube takes only 1.8 sec. I’m not disappointed that the stick didn’t last 4 years at its price point, but I would be very concerned if the FTV1 or the Cube performed that poorly after 4 years.

  10. AFTV Fan says:

    Can you run the benchmark of the stick with some dry ice underneath it to keep the stick cooler? Would that even ale a difference? The Cube has same hardware, but more space inside for heat to dissipate. Using an active temperature lowering measure like using dry ice may even boost the stick to have higher scores than the Cube.

    • AFTVnews says:

      I think you mean to say the pendant, not the stick. The Fire TV Cube and the Fire TV 3 Pendant don’t have the same hardware as the Fire TV Stick.

      Either way, actively cooling probably wouldn’t make a difference. It isn’t slower because it is getting hot. It’s configured to be slightly slower so that it doesn’t get hot. You can’t overcome the CPU governor configuration by cooling it more.

  11. Andy says:

    I was very disappointed and found the wifi really slow… buffering like crazy…I have 50 down and this was barely pulling in 18, yet my trusty Apple TV was getting 45, no problem.
    Plus I kept asking it to play “Goliath” which is an Amazon Prime show, but it brought up just about every other thing that sounded like it, but not what I wanted.
    I do love the way it works with turning on my TV and sound system when I ask Alexa to turn on the TV, but without the other stuff to back it up, it’s a big fail for me.

    • clocks says:

      I like the way Prime Video is on FTVs more than any other device, but I HATE HATE HATE that Amazon MAKES me watch their trailers. Over, and over, and over again. You used to be able to hit back, then go back into the vid, but now Amazon doesn’t even allow you to do that. Ugh just thinking about this has me considering returning my two unopened FTV cubes.

      • BobR says:

        Very annoying as is the YouTube workaround as is the need to re-authenticate apps after firmware upgrades and do so for each Fire TV that you have!
        My workaround for the trailer is to hit fast forward once. BTW my HBO app does that too.
        FYI my Cube still doesn’t consistently turn on my FIOS box so beware

        • clocks says:

          Oh is it as easy as just hitting the FF once?!?! I’ll have to try.

          I’m still waiting for some reviews to show up on Amazon for this cube. I have two unopened, and am trying to decide if I should keep them or not.

          • BobR says:

            Good idea. I’m upgrading from a gen 1 stick so took the plunge. When the equipment controls work, it meets expectations. I wish there were more options to modify the IR cmds as I have figured it out on my universal remote which I still need as backup. I do lose my HDMI switch capability and would now need to manually swap HDMI cables to add my Blu Ray player since I’m constrained on HDMI inputs. I won’t upgrade my other first gen stick for this. I really just need an upgraded FTV2 box as many others had hoped for. I’ll likely get a Roku Ultra on sale instead

  12. David says:

    This is great analysis. This must have taken hours of work. Thank you for that :)

    Can you tell me what the raw performance difference is compared to a old FTV1? Especially the GPU.

  13. Randy Reid says:

    What app should I use to measure the Ethernet vs the WiFi?
    I downloaded and used “Internet Speed Test” but I do not think it is correct. It is saying I’m getting 9 Mbit/s Down and 2 Mbit/s Up.
    Does this sound correct?
    I’m not having any issues. Also, the speed is not much different when I switch between WiFi and Ethernet.

  14. Stuart says:

    If Alexa is muted on the FTV cube, can the 6% cpu overhead be regained?

    Stu

Leave a Reply

Your email address will not be published. Required fields are marked *

*

Get notified of new posts

Enter your email address to receive notifications of new posts by email.