Mini-PC comparison benchmarks - MK802 vs. UG802 vs. GK802

Openbenchmarking.org has a nice comparison of the standard HDMI Mini-PC's running Ubuntu 12.04 natively.

The comparison is a Phoronix Test Suite of the MK802+, UG802 and GK802. It also has ODROID-U2, MK803, Exynos 5 Chromebook, and Intel Core i7-950 thrown in for good measure. Unfortunately the RK3188 is not in the comparison. Obviously the Core i7 comes out way on top in all tasks with the Exynos 5 far behind. Then comes the Exynos 4 ODROID and the RK3066-powered UG802. The iMX6q-powered GK802 is slightly slower than the UG802 in most tasks. The AmLogic-powered MK803 and the A10-based MK802 are at the bottom of the performance scale.

Using existing Antutu results as a comparison baseline I would expect the A31 to finish the cpu-related tasks somewhere below the performance of the GK802. The RK3188 should be significantly faster than the Exynos 4 ODROID in most tasks.

Therefore the conclusion of the benchmarks is that RK3066 currently offers the best value for the money in a HDMI Mini-PC. The RK3188 would be the fastest currently if it was available. Thus the RK3188-powered UG007B is a much better value than the GK802 as the UG007B already has a much lower price in presale then the GK802. The GK802 offers better Linux support, however, and if you need SATA it is currrently the best option provided you are willing to pay the "developer board" premium for that feature.

I think the person who did the testing should have included the Acer C7 Chromebook. It would by far be the winner.

Unfortunately I don't own an Acer C7 Chromebook to include in my tests! For the rationale behind the tests please see my original post on Google+ (https://plus.google.com/109451178006683865932/posts/DPrg7Anp1xU) or the excellent article on cnx-software (http://www.cnx-software.com/2013/03/27/mini-pcs-mk802-ug802-gk802-istick-a200-linux-performance-comparison/).

I'd like this to be taken as a starting point and to be build upon with other's results for mini pcs running different software or new mini pcs as they become available. Someone has already done this with these results http://openbenchmarking.org/result/1303271-FO-1303250AR79.

gnexus Wrote:
-------------------------------------------------------
> [url=http://openbenchmarking.org/result/1303250-AR
> -MINIPCPCU05]Openbenchmarking.org[/url] has a nice
> comparison of the standard HDMI Mini-PC's running
> Ubuntu 12.04 natively.
>
> The comparison is a Phoronix Test Suite of the
> MK802+, UG802 and GK802. It also has ODROID-U2,
> MK803, Exynos 5 Chromebook, and Intel Core i7-950
> thrown in for good measure. Unfortunately the
> RK3188 is not in the comparison. Obviously the
> Core i7 comes out way on top in all tasks with the
> Exynos 5 far behind. Then comes the Exynos 4
> ODROID and the RK3066-powered UG802. The
> iMX6q-powered GK802 is slightly slower than the
> UG802 in most tasks. The AmLogic-powered MK803 and
> the A10-based MK802 are at the bottom of the
> performance scale.
>
> Using existing Antutu results as a comparison
> baseline I would expect the A31 to finish the
> cpu-related tasks somewhere below the performance
> of the GK802. The RK3188 should be significantly
> faster than the Exynos 4 ODROID in most tasks.
>
> Therefore the conclusion of the benchmarks is that
> RK3066 currently offers the best value for the
> money in a HDMI Mini-PC. The RK3188 would be the
> fastest currently if it was available. Thus the
> RK3188-powered UG007B is a much better value than
> the GK802 as
> [url=http://www.aliexpress.com/item/Pre-order-Mini
> -PC-Android-4-2-RK3188-Quad-Core-Cortex-A9-2GB-RAM
> -8GB/809621412.html]the UG007B already has a
> [b]much[/b] lower price in presale then the
> GK802[/url]. The GK802 offers better Linux
> support, however, and if you need SATA it is
> currrently the best option provided you are
> willing to pay the "developer board" premium for
> that feature.
>
> I think the person who did the testing should have
> included the Acer C7 Chromebook. It would by far
> be the winner.

Thank you for your post Ian!

Quote

Unfortunately I don't own an Acer C7 Chromebook to include in my tests!

Well I do have a Acer C7 Chromebook ;)

It is running Fedora 18, however. Therefore running your benchmark would not be an accurate comparison.

When I have some time I will try to duplicate a drive and put Ubuntu 12.04 on it. Then we can get an accurate result. In the meantime I will try to load the Phoronix Test Suite on Fedora to see how it compares with Ubuntu.

Quote

to be build upon with other's results for mini pcs running different software

It would be good to see how Debian Sid compares to Ubuntu. I have noted previously that it performs significantly faster than Ubuntu on the A10-based platforms. I would expect that to also be true on the other devices.

Hey Ian, why are you using a Phoronix Test Suite Release from two years ago?

mk802plus-ubuntu-complete:Phoronix Test Suite 3.6.1 - 2013-03-20 15:29:21

Phoronix Test Suite 3.6 "Arendal"
Release Date: December 13, 2011

There have been five releases since 3.6 was released. I'm going to d/l 4.4 since that is the only one that seems available. But it is going to be difficult to get accurate comparisons with people using such different versions of the test.

I just completed Ian's tests on my C2D laptop which is fairly close to the performance of the C7. Looking closer at Ian's Chromebook results I have noticed that the Samsung Exynos 5 Chromebook is a really bad value price vs. performance wise. Epic fail! You can clearly see the massive performance difference between ARM and Intel i7 in Ian's tests. My results are far closer to the i7 than to the Exynos 5 even with a 7 year old cpu. Obviously ARM arch, even with Exynos 5, still has a long way to go to catch up to Intel. I'm a big ARM supporter. But the results here do not make me optimistic.

My results confirm the Acer C7 is by far the winner in the value contest. If you have the extra $100, don't need the HDMI dongle form-factor, and just want to run Linux, there is absolutely no point in getting a Mini-PC. Get a C7 Chromebook instead.
Once I get all the results consolidated, and also have the C7 results, I will post a link.

My ideal ARM board is a G**4 :) 1GHz, 1GB RAM, 1Gb Ethernet, >1GB NAND. With USB 3.0 or SATA and Wifi. For less than $150.
Is there such a product out there?

gnexus Wrote:
-------------------------------------------------------
> Hey Ian, why are you using a Phoronix Test Suite
> Release from two years ago?
>
>

> mk802plus-ubuntu-complete:Phoronix Test Suite
> 3.6.1 - 2013-03-20 15:29:21
>

>
>

> Phoronix Test Suite 3.6 "Arendal"
> Release Date: December 13, 2011
>

>
> There have been five releases since 3.6 was
> released. I'm going to d/l 4.4 since that is the
> only one that seems available. But it is going to
> be difficult to get accurate comparisons with
> people using such different versions of the test.

Interesting point. I simply installed PTS as a package using aptitude - see also http://manpages.ubuntu.com/manpages/precise/en/man1/phoronix-test-suite.1.html. Now that I look at the package in Ubuntu Software Center, under Updates is states "Canonical does not provide updates for Phoronix Test Suite. Some updates may be provided by the Ubuntu community". I also read in the March release of MaxiumumPC that they refused to install PTS as Ubuntu Software Center as it comes up with a "package is of bad quality warning" (see page 69). I had a random look at details for some of the latest test results posted and PTS version variations seem to be common. Given the tests are downloaded prior to running, I'm hoping that the tests are version independent.

Quote
bodhi
Is there such a product out there?

I think we've determined in the other thread that our mythical GbE+SATA device for under $100-150 still does not yet exist.
Unless you have something new to add?

Can't find any either :) But I think the iwave is a very interesting one (no price yet known).
http://www.iwavesystems.com/product/single-board-computer/i-mx6-pico-itx-sbc/i-mx6-pico-itx-sbc.html

The iwave is nice, as are a few other imx6 boards in the other thread. But the iwave will still have the low-volume pricing issue. It seems that recently the price of the imx6 has dropped to about $20, which is fairly reasonable, and not that much more than A10. Therefore I fail to see why no manufacturer has stepped up to the plate to fill the demand for a low-cost imx6q device with SATA. Hell, just put an imx6q module on a Wandboard and call it done, or stick an eSATA port on a larger GK802 and call it done FWIW. Or make a Mele device with an imx6q. Or simply drop the price on the Habey BIS 6332B and sell it as an Android HTPC. It's not that difficult. . .

There could be a bit of money to be made on such a dual-use device if it was marketed like the Raspi. Freescale would be the main beneficiary. It would be in their interest to promote such a device. It would be to their benefit to assist a mfr. in developing and marketing such a device. Instead we have a imx6q tablet that fails terribly in the benchmarks, a GK802 that fails to compete with the older Rockchip devices, and a ton of overpriced "development boards."

Freescale made a mistake in trying to market the imx6q against Rockchip in Android devices. It was obvious from the start it cannot compete in power consumption vs. performance - especially GPU performance, with RK3066 in a tablet. But the imx6q is ideally suited for a low-cost Linux server/HTPC. Just promoting the imx6q as a Kirkwood replacement would have worked.

Freescale is obviously incapable of properly promoting their SoCs as mass-market devices - stupid mgmt.

I finally have Ian's benchmarks completed for my Intel Core 2 Duo T7300 Asus F3SV laptop:

http://openbenchmarking.org/result/1303316-FO-GNEXUS25615

The individual tests in PTS version 4.4 seem to be the same as in Ian's older version. Therefore, as Ian noted, I would expect little to no variation in the test results due to the different versions. Fedora 18 would not work with the CLOMP bechmark due to issues in Fedora with the packages that needed to be installed. Therefore I had to recreate the benchmark without that test. I also had issues with the PTS install mucking up my F18 mime-types.

This test is not of an ARM device. This is a full Intel laptop. It seems fair to Intel to note that did not disable any running services while running the test. I was browsing the web while the tests were running. Mythbackend, MySQL, five Apache threads, libvirtd, smbd, cupsd, and KDE4 with all the akonadi crap was running during the test along with Firefox.
Maybe also running Mythfrontend would have made it a fairer test against the mini-pcs :)

The older C2D laptop and the new Acer C7 Chromebook seem to have fairly similar performance with Fedora. I'm installing the Phoronix Test Suite now on the C7 and will try to get the results posted soon.

To duplicate all of Ian's tests, including CLOMP, you can run the following command after installing PTS:

phoronix-test-suite benchmark 1303250-AR-MINIPCPCU05

My benchmarks above included all of Ian's tests except CLOMP. If you also have installation issues with OpenMP as I did then you can use the following command to duplicate the test without the CLOMP bechmark:

phoronix-test-suite benchmark 1303316-FO-GNEXUS25615

If anybody here gets a UG007B or other RK3188-powered please run one of the above benchmarks to help us out.

Suprisingly, I actually got it wrong about the performance of the 1.10GHz Intel Celeron 847 in the Acer C7. It is nowhere near as fast as the 2.0 GHz C2D In fact, it is about half as fast in most tasks. The dual-core Exynos 5 in the Samsung Chromebook actually compares to it evenly in most tasks! My C7 feels as fast, or even snappier in some tasks, as my C2D notebook. But the benchmark results mostly prove otherwise:

http://openbenchmarking.org/result/1304027-FO-1303316FO12

Considering the $199 C7 is $50 cheaper it is certainly still the winner in the value category. But not for long it seems. Newegg now shows it as "discontinued." Acer says they now have a "new model" with more memory (still only 4GB and not really needed) and a bigger battery (which likely means more weight). It is $280 - $30 more than the Samsung. Raising the price to $280 makes the C7 a bad value. The Exynos 5 Samsung still has much better battery life with equivalent performance.
So if you want the original $199 Acer C7 you had better get it fast, and not from Newegg or Amazon USA.

I would score this one as a big win for ARM. An Exynos 5-based device would undoubtedly be perfect for our needs as an up-to-date Kirkwood replacement. I can't wait to see how the RK3188 compares. Unfortunately it does not have SATA.

I suppose now I'm going to have to run this benchmark on my EeePC to prove how badly Intel Atom sucks. My Mele felt faster.

gnexus Wrote:
-------------------------------------------------------

> I would score this one as a big win for
> ARM. An Exynos 5-based device would
> undoubtedly be perfect for our needs as an
> up-to-date Kirkwood replacement. I can't wait to
> see how the RK3188 compares. Unfortunately it does
> not have SATA.
>

I don't think we will ever see the Rockchip with SATA. It's obviously built to sell to the Smart TV market, and the end users would not know or care to plug in a SATA HDD. The better odds is that it will eventually have USB 3.0.

@gnexus - You've pimped up your C7 as I see 6GB RAM and a 500GB drive on the PTS results? Would you recommend replacing your Scorpio Blue with an SSD (albeit smaller in size) to eek out more performance? Also the CPU scaling governor is set at "on demand" rather than "performance" so there is another potential improvement here.

@linuxium

Quote
linuxium
the CPU scaling governor is set at "on demand" rather than "performance" so there is another potential improvement here.

The kernel I'm using for the C7 was compiled using a config that was as closely aligned to my x86_84 Fedora config as possible. Unfortunately it required a few changes to be able to boot on the C7, and I still have not been able to get the touchpad to work. One thing that was able to stay constant was using the "on demand" governor. I could have changed it to "performance" but that would have probably diminshed the already poor battery life compared with the Samsung. The battery life is good enough for me. I would not want to see it go lower by using "performance" governor. Performance is great, and it seems much faster than what the benchmarks suggest. Battery life is currently much better than my old Atom netbook which originally advertised a bogus "6 hour battery life."

Quote

You've pimped up your C7 as I see 6GB RAM and a 500GB drive

Going with the slimline WD 500GB made a huge difference in performance. The original 320GB Seagate drive was crap. If Acer uses the same, but larger, Seagate drive in their new model that will make the C7 a very poor value. I thought about getting a SSD, but I wanted a big HDD the same size as the Asus laptop. Unfortunately Chrome BIOS (EFI firmware, actually) only uses GPT. So the drives still will not exactly match unless I also change my others to GPT.

The only reason I was willing to buy the C7 was the use of replaceable SoDIMM and HDD. I ordered the 4GB DDR3 SoDIMM at the same time I purchased the C7. I never even booted it with the single 2GB. I also ordered the 500GB drive at the same time. I found the C7 uses a slimline 7mm drive so I had to return the 8mm WD.

Rat here. Haven't been active on the forums for a while but this thread caught my eye because I'm also using an Acer C7 Chromebook.

It is, by far, the absolute best value for the price you pay and performance you get. Best complete system you can get out there right now. (Still available, just order it through Google) I opted for the C7 over the Series 3 because it has a memory socket and removable battery. x86 compatibility helps (since I can run WINE if I need to) but it wouldn't have been the dealbreaker with the S3's ARM SOC. Memory starvation is a real issue, even with 2GB stock, so I could not accept a fixed memory config.

My "gut feeling" benchmark places the C7's Celeron (Sandy Bridge based) at roughly equivalent to a 1.4 to 1.6Ghz Core2Duo. Benchmarks generally tend to agree. It is extremely ahead of the miserable Atom and that was a large part of what enticed me to give the C7 a shot when my original Aspire One died.

I got a cheap stick of 8GB of ram on sale, so yes, I'm using my C7 with an admittedly overkill amount of 10GB of total RAM. No regrets. ;) (The extra ram helps when I play Team Fortress 2, which runs great on the C7 provided you turn down the eyecandy)

I swapped out the hard drive for an older OCZ 60GB Vertex + SSD. No major difference in speeds though less lag in file access. Also gained about a half hour of battery life.

9mm drives will fit in the C7 though it makes for a tight fit. I'd only recommend doing this for an SSD as it won't have moving parts that may be sensitive to compression from trying to put the panel back on.

Between the Dockstar and the C7, I have my two most favourite Debian hack systems ever. Amused and happy to see the C7 mentioned here. ;)