You Asked, We Answer – AOSP Build Times on AMD Processors
A few months ago I kicked off a series of articles aimed at examining what affects build times when building an Android ROM from open source options such as CyanogenMod or OMNIRom.
You showed us you were very interested and then did something else — started to ask questions. Does Intel or AMD make a difference? What about mobile processors versus desktop? I had many of the same questions and very little proof to back them up at the time. Today’s article will help answer some of the questions as far as AMD processors go to see how they fare in AOSP build times.
Now, before we get started it’s a good chance to explain something – what you can expect to see here on XDA regarding PC hardware coverage. We will continue a focus on things that affect build times as this seems to be the largest interest that can’t be found elsewhere. But when we shift away from that focus to look at other aspects, such as virtual/augmented reality or providing a recommended build configuration, we wouldn’t be doing a good job if we didn’t look at the complete system including graphics solutions. If there’s something you would like to see for development or other reasons feel free to reach out and let us know!
After the previous article we contacted AMD and they were happy to provide 3 processors in their current lineup for testing: The Athlon X4 845; the FX-6350 and FX-8350, both with the new Wraith coolers. This unfortunately requires two separate motherboards for testing, but all other components will be kept the same.
Test Bench: Lian Li Pit Stop T60B
CPU: AMD Athlon X4 845; FX-6350; FX-8350 (all using provided air cooling solution)
Motherboard: ASRock A88M‑G/3.1 (X4 845); ASUS A5R97 R2.0 (FX-6350, FX-8350)
RAM: Micron DDR3-1600 2x4GB
SSD: 2 x OCZ Trion 150 240GB configured in RAID 0 (except /boot and swap)
Power: Corsair CX750M
O/S: Ubuntu 16.04 LTS with latest updates installed on June 2nd.
Sourcing: All CPUs and ASRock motherboard provided by AMD. All other parts were either purchased or readily available to me.
Both CyanogenMod (cm-13.0 branch) and OmniROM (android-6.0 branch) will be synced and necessary additional device/kernel/vendor for the Nexus 6P (angler). A total of 12 builds will be performed in each configuration: 3 with ccache cleared, 3 using pre-populated ccache. An average will be taken from the 3 builds and used as the configuration’s “non-ccache” and “ccache” build times. I’m sure that there are better results that can be achieved on AMD hardware with other options, and we’ll explore those in a separate test. But for this one it’s all about consistency – minimizing the variables and maximizing the control group.
Before getting to the results I’d like to give a shout out to AMD for the Wraith cooler. I opened up the one that came with the FX-6350 and then reused it for the FX-8350 after cleaning and reapplying thermal paste. And even during the entire build process and in a server room I could hear plenty of fans going, but not the Wraith. The cooler also comes with an LED that didn’t appear to light up on the one I opened, but that could have been an issue with the board since it was second-hand.
Results & Interpretation
OK, enough chatting, you’re here for build times. So here we go, the averages of 3 builds under each:
Times are provided in hours and minutes. The first processor, the X4-845, is based on Carizzo and therefore the most recent architecture. The other two, the FX-6350 and FX-8350, are based on Vishera designs and have been on the market for 3 and 4 years respectively. We can’t take much away from the Carizzo performance versus Vishera because the CPU bottleneck is just way too high and causes the builds to suffer dramatically on a full build without ccache. That starts to open up and we see a better picture of performance with Vishera, but even in the case of the FX-8350 all CPUs were pinging at 100% for a good portion of both the non-ccache and ccache builds. This unfortunately means that there is a bottleneck here.
When we look at this from a different perspective though – price – we can see that for around $500 it’s possible to build the test configuration I have here. By comparison the Intel i7-4790K would cost about $150 more for that faster build performance, and builds might go to half of the time of the FX-8350. For most that could be just over the limit – I’d even suggest anyone consider the FX-6350 based on the slightly longer build time but $30 less. Kernel builders or app developers would probably still experience very little issue with either of the FX processors tested in this scenario. Add a video card such as AMD’s upcoming Radeon RX 480 that is supposed to come in at around $199 and you could have a build machine that can easily double as a VR capable system for under $700.
And then there’s Zen, the processor that AMD has been working on for years to better combat the performance that Intel has been delivering with its Core desktop lineup. Last year slides surfaced that indicated Zen would offer a 40% increase on instructions per clock (IPC) over Excavator, two revisions beyond the Vishera processors tested. While no solid numbers have been provided estimates have anywhere from a 60% to 90% IPC increase from Vishera to Zen. We’ll be looking forward to testing those processors as they make their way out and comparing them to these build times to see just how much they improve the situation
If you’re one of those users who has to crank out a build in very short order, the AMD FX line does seem to struggle to match the performance from the higher ends of the Intel Core lineup. But where it’s lacking in build times it makes up for in price, making the ability to build one of these systems on a budget much more likely – especially if reuse some existing parts, such as the power supply or case. Those who are focused only on a full AOSP ROM from source may have issues even then – kernel and app developers should be able to develop on AMD processors with little to fear in the way of performance. And with new processors on the way as early as this fall for Zen, this has been a great opportunity to review what AMD’s current lineup offers and set the stage to properly evaluate Zen’s performance when it releases.
For $30 less and almost similar build times the FX-6350 seems to be the way to go for most out of these three. Those that can afford the extra $30 may be better off putting that in faster RAM or an improved GPU if they are gaming. Even though these processors are longer in the tooth, they’ve also proven they can keep up with most average applications, especially when paired with the right GPU for gaming and are a great value for someone looking to do an upgrade on a budget. But what if you’re considering on upgrading away from any of this lineup currently?
My advice would be to wait just a little bit longer for Zen to come out. Go ahead and upgrade that GPU now while you wait, since you can take that over to a new motherboard. Zen has the potential to dramatically change the game and it’s definitely worth the wait to see how it performs before deciding.
Again our thanks to AMD for the supplying the CPUs for these tests! Did you find this information valuable? What would you like to see tested next? Let us know in the comments below or feel free to carry on the discussion in our forums, Twitter, Facebook or Google+!