How well does Intel’s new phone work as a phone?

Review: Hands on with the Lava Xolo X900

Intel Atom LogoIntel is now making phone chips, and the first device to bear one, the Lava Xolo X900, is on the market. With that in mind, we tested one for a week to see how well it worked as a phone.

Normally we review things with benchmarks, numbers, and objective measurements, but in the case of the First Ever Intel Phone (TM)(R)(C)(BBQ) that seemed a little pointless. In our minds, the best comparison was not FP number crunching or other technical benchmarks, but a more human standard, does it work as just another Android phone? Would the Intel ARM code emulation for x86 slow it down to a crawl? Would it eat batteries at a horrendous rate? Would it work with random ARM code? Most of all, would it crash a lot, or just be unobtrusive as a phone? Those questions are far more important than things like millions of polygons per second, far more.

We should note that our brilliant plan to use it as a phone did not come about because of the broken state of the US cellular phone system. There are four main carriers, and only two of them use the same GSM signaling of the Xolo X900. Only one of those, AT&T, has the right 3G frequencies to test that with. T-Mobile might work, but since they screwed us (twice actually), we wouldn’t go near them with a radiation suit and firehose.

That left AT&T, so off to the local house of high prices to get a temporary SIM card to test with. Once there, our plan quickly fell apart. AT&T would sell us a pre-paid SIM to use voice, and only wanted merely absurd fees to do it. Unfortunately, they would not sell us anything with a data plan despite listing one prominently in their literature, on their computers, and through their staff. The only way to actually get what they offered was to swear fealty under their stock ticker for a two year indentured servitude program. Instead, we stuck to Sprint and left voice and 3G testing out of the picture.

The Phone:

When you pick up the Xolo X900, you immediately notice two things, that it is very light, and it has concave sides. The front is glass, the back a rubberized plastic, both in fashionable black. Those concave sides are aluminum, and are deep enough to mostly prevent errant presses of the three buttons, volume, power and camera. Feelings on this were mixed when the famed SPORP-WRIWHTPIOP (SemiAccurate Panel Of Random People We Ran Into When We Had The Phone In Our Pocket) played with it, no one expressed strong objections or attractions, but all had opinions. One thing we can say is that it was very easy to differentiate this from the other phone in our pocket via touch.

Intel Medfield Lava Xolo X900 Front

Front of the phone

The front of the phone has the normal four android buttons, a mic, speaker, camera, light sensor, and status LED. In addition to the buttons, the top side has a 3.5mm headphone jack, the bottom a micro-USB plug, and the sides each have a speaker, and one has a micro-SIM slot with obligatory ejector hole, the other a micro-HDMI port. Nothing added, nothing removed, it is about what you would expect, vanilla features.

Intel Medfield Lava Xolo X900 phone read

Back of the phone

The rear of the phone is a little more controversial with a raised bump out for the camera lens and flash. Here the SPORP-WRIWHTPIOP was a little more negative, with no one really liking the solution, but most being neutral. A few would have preferred a larger raised area like the new Razr, a lower bumpout, or even a slightly thicker phone. None expressed an interest in giving up the camera for a sleeker phone when we asked, but most would trade a much larger battery for a thicker phone. Raising half the phone 2mm or so would easily allow for 2x the battery volume, something we hope a phone maker tries soon.

Overall, the phone is pretty solidly built for a first effort. There is nothing to rave about, nothing to complain about, and the SPORP-WRIWHTPIOP were mostly split on general aesthetics with overall feelings being somewhat neutral. It is thinner, lighter, and better built than the prototypes we played with prior to launch, and this is only the first of many designs to come out. Motorola, Lenovo, and others will undoubtedly come out with more polished devices later this year.

For raw specs, the phone is about 123mm x 63mm x 11mm, with roughly 2mm of the rear height being the camera bumpout. The Xolo X900 weighs a mere 127g, less than the 140g of the iPhone 4S, but more than the Samsung Galaxy SII’s 115g. The battery is 5.4Whr but unfortunately is non-removable. There are some things you should not copy from Apple, this being a prime example. No, there are a lot of them, but non-removable batteries are a blight no matter who ‘inspired’ them.

One last thing to note, if you pull off the back to see what you can poke at, some of the buttons on the side of the phone seem glued to the cover, making it impossible to pry the plastic off fully. This is annoying, and doesn’t seem to serve any real purpose other that to be the one hardware bit we vehemently dislike. Luckily, most users will never run in to this ‘feature’.

Hardware:

When Intel announced their 32nm Atom for phones, the Medfield core and Penwell SoC last January, we looked at them all in detail, you can read the three parts of the article here, here, and here. Since then, nothing has really changed, so we won’t go over the silicon aspects again, but just sum it up by saying that the chips look good, but we held off on more commentary until we had an actual phone in hand to test. That is what we are doing now.

The Xolo X900 sports a pretty standard Intel reference platform as far as the silicon is concerned. The single core 1.6GHz Penwell has two threads, and is backed by 1GB of LPDDR2 at 400MHz and 16GB of, subjectively speaking, pretty fast flash storage. On the visual side, the 4″ 1024 * 600 backlit LCD screen is bright and crisp, with no real problems noted.

This screen is powered by a 400MHz Imagination PowerVR SGX 540 GPU, and several video encoders, decoders, and image processors all discussed at length in the earlier articles. They provide support for the front mounted 1.3MP still and video camera, and the 8MP rear facing main camera. The rear camera has a decent flash, and will do 1080p30 video as well. As far as paper specs go, the Xolo X900 is near the top of the heap for modern phones, ahead of the ARM A9 based competition, but looking a little anemic when compared to the imminent crop of A15 based phones that are in the process of launching.

Software:

This is where things get a little interesting. The prototype Medfield phones SemiAccurate saw earlier all had Android 2.3.x on them, and a build of 4.0 had just made its way to a few devices. We were expecting the shipping products to come with 4.0, but instead the X900 has Android 2.3.7, not bad, but a little disappointing. The Lava page about the Xolo says it is upgradeable to 4.0, and we have no doubt that it is. Whether it is actually updated or not is an open question, but Intel seemed to indicate that it will be in short order, but no hard promises yet.

The first thing you notice about the X900 is the rather sparse software load, the usual raft of unremovable bloatware and popup-inducing garbage is missing. This initially off-putting feeling is quickly replaced by a big smile when you realize that you don’t have to spend hours figuring out how to turn that garbage off and dig up utilities to remove as much as humanly possible. This lack of crapware is a blessing, but is unlikely to carry over to other Medfield phones once the carriers have more time to impose their annoyances on unwitting buyers. This, however, is not Intel’s fault.

Overall, the Xolo X900 can be described a a fairly basic Android build, no frills, and very few extras. On the plus side, most extras are negatives, but a few defy carrier wishes and transcend to usefulness. All in all, there were a few bits that took head-scratching to figure out, but nothing that was unusually opaque if you have any familiarity with a modern Android phone. Any Intel features that were added under the hood were transparent or silent, nothing other than the logo on the back made the x86-ness apparent on the software front.

Objective tests:

OK, we lied, we ran one numerical test, AnTuTu v2.7.3. We tried to run GeekBench, but without a SIM, we couldn’t actually buy anything from the Android store. With more time, we probably could have worked around this, but AnTuTu answered most of our questions. As a comparison, we have put up the numbers for a Motorola Photon, a 1GHz dual core Tegra 2 based Android phone.

Medfield benchmarks

Antutu scores

The first thing we noticed is that a 1GHz Tegra 2 loses out to the 1.6GHz Penwell overall by more than 10%, but the total is not telling us anything really important. The memory score of the Intel chip is about 50% higher than the Tegra’s, no doubt due to the greater width of Intel’s memory interface. As we have noted earlier, this carries over to Tegra 3, and will only add to the pain for devices based on that part. Moving down a bit, the SD numbers also show a similar story. AnTuTu seems to multiply the read and write speeds of the flash by 10 to get the scores for storage scores, the Photon had a rating of 6.5MBps while the Intel SoC was clocked at 6.9MBps. These numbers seem very high for flash, we expected about the same numbers, but with Mb, not MB. That likely means both are cached heavily, and the benchmark is reading that info.

Caching like this is a good thing though, it gives a much clearer picture of the SoC properties rather than an add-on flash chip’s stats. More importantly, the read numbers are very telling. The Photon scores 128, and the read rate is listed as 12.8MBps. The Xolo scores a 206 so it reads storage more than 50% faster, right? AnTuTu lists the read rate not at 20.6MBps but at >50MBps, something that seems, well, unlikely.

Unfortunately for gut feelings, our subjective testing completely back up this massive bandwidth number. Apps load really really fast on the Xolo X900, loading the same thing on the Photon takes notably longer in most cases. The speed difference is quite notable to the layman, you really can notice the difference. That said, it isn’t universally faster, some things, notably native ARM games, do take a bit longer to load than on the Photon.

The CPU scores tell a similar tale, the 1.6GHz Medfield should be notably faster than the 1GHz Tegra 2 as shown by the SunSpider tests in the earlier articles. SunSpider is a best case for Intel, and somewhat of a trivial test for a modern SoC. Even based on theoretical MIPS, FLOPS, or specs, Medfield should be head and shoulders above the Tegra.

We ascribe the lower than expected AnTuTu performance to the ARM emulation overhead, basically it takes a chip that it ARM A15 competitor on paper and slows it down to the level of a lower clocked A9. That said, the Medfield core still scores higher than the Tegra 2 on Int, but loses by a bit on FP. If you are running Dalvik (Android VM based) code, the Xolo is very fast, notably faster than the Photon. On native ARM code, it is hobbled quite a bit, but still faster than many current ARM A9 parts most of the time.

Graphics are roughly a wash, something that is reflected in our subjective testing of games and apps, and the numbers show that. A 400MHz SVX540 is not exactly a world beater, but it is more than enough to run just about all current games without bogging. Nothing to see here numerically speaking, move along people.

The last bit, Database IO, is about what we expected, and mirrors the memory and storage numbers nicely. IO is reading and writing, so if it is not bottlenecked by CPU performance, it should be a pretty accurate reflection of memory and storage speeds. The ratio of the results here are about equal to that of the RAM tests, so it mainly backs that data up nicely.

For objective testing, the Xolo X900 confirms our initial subjective impressions, it is fast, very fast actually, when doing some things, but when the ARM emulation steps in, it takes the edge off that advantage. Even handicapped like this, the Penwell SoC still posts numbers better than many current generation phones. So far, so good, and the ARM code emulation is a bit better than we expected, and more importantly, results in a real world usable device.

Subjective Tests:

For me, this is the most important part of the test, subjective things that don’t fall in to easy spreadsheet cells. In a nutshell, it is the “How does it work as an everyday phone?” factor, and that is not easy to pigeonhole. SIM difficulties aside, the test was basically carrying the Xolo around for a week and using it everywhere I could instead of my normal Android phone.

A successful result would be that an x86 Android phone did everything an ARM Android phone did, and nothing went awry. We didn’t do much testing of features around the phone construction itself because everything from the case and speakers to the cameras will change according to vendor preferences. Some may be better, some worse, but none are really indicative of the silicon and SoC underlying the X900, and that is the object of the test.

All this said, the sound quality was pretty good from the speakers, pulling up some soothing melodies from YouTube put the sound quality a bit above OK, but far from great. Anthrax’s soothing melody “Antisocial” was a bit tinny, and other similar ditties lacked tonal depth, but nothing was really objectionable. That said, each Intel based phone will likely have a different speaker set, so try before you buy.

The cameras were again solid, above average even. Pictures were as clear as any we take with a camera phone, (Editor’s note:  Not just Charlie was allowed to take pictures and evaluate them as he sucks at photography.) and the 1080p videos were of good quality and clear. They even played back on an HDTV without a hitch, something that few modern phones can pull off if the bit rates approach acceptable. This goes to show the power of the Intel image processors and video en/decoders that Penwell sports. Intel was rightly proud of these, and it does add to the overall experience on the resulting devices. Similarly, the 8MP camera on the back of the Xolo X900 is good, but not standout. 1080p30 is uncommon too, but not unique anymore.

One thing that is unique is the burst mode of the still camera, it can take up to 10 pictures in rapid fire succession, something that no other phone camera that I know of can come close to. Even some DSLRs have trouble with this type of speed, but the Penwell SoC can do it, and do it well. This speaks well of the image processor, but also of the I/O capabilities of the phone. Both are needed to do a burst mode well, and Intel has delivered here. It will be interesting to see how well the near future ARM A15 SoC can pull this off, this is one thing they might not come close to doing.

On the user interface front, Intel spent a lot of time optimizing the human interaction bits, and were not shy about talking up their progress. The X900 is really fast and smooth when sliding screens back and forth, and never seemed to bog down. Most Android devices stutter, pause, and are anything but smooth, especially after months of cruft builds up and background apps take a bite out of CPU time. We didn’t have the X900 long enough to clog the silicon arteries with Android borne digital cholesterol, but our initial impression is the smoothness of this phone are better than any we have tried recently, new or clogged.

Software:

This is by far the most important aspect of Intel’s plans to enter the phone market, can they match the performance of the incumbent with their upstart ISA, x86? Can they run the ARM software out there not just seamlessly but fast? Intel emulates ARM code, and claims a 90%+ compatibility rate, something that should be good enough for most users. On paper, they have what it takes, but how well does it work in the real world?

For a week or so, we took the phone out and used it whenever we needed to do something on our phone. As long as we could get on the net, the Xolo was used preferentially, and that was almost everything not call related. When free time allowed, we played the games we normally play, and tried a few others based on whim more than anything else. Nothing unusual, nothing scripted, just normal use.

The short story is that it worked almost perfectly. Almost was because of the one hitch we hinted at earlier, a single app would not load, Opera mobile, it gave an invalid package error every time we tried, something that may not be the fault of the phone itself. Luckily, Opera Mini, the browser we intended to load in the first place, worked like a charm.

For browsing, the X900 was fast, notably faster than a Tegra 2. Pages popped up quicker, transfers over Wi-Fi felt much faster, and large pictures rendered very quickly. More impressively, the pages scrolled noticeably smoother and zoomed in and out with much more fluidity. The horsepower of the Intel CPU was quite evident here, and while notable, the choke point was still the network . A Tegra 2 is just about fast enough to render more data than the bandwidth can bring in, a Penwell is fast enough, but then bandwidth becomes the major problem.

One interesting bit to note, the speed of the rendering made the Chrome browser baked in to Android a little annoying. Chrome loads and renders data, then display the page. Opera on the other hand progressively renders the images and displays text while loading the page. While it can be a little annoying to watch on a slow connection, but I much prefer to be aware of the progress and see the text while large images trickle in. It is basically a useful progress bar. With the added speed, Chrome on the Xolo waits and waits, then pops up in a jarring way, enough to be annoying. Stick with Opera on the Xolo.

Flash worked just fine, the phone came with 10.3, and will likely be upgraded to 11.x if and when Intel pushes out Android 4.0. With luck, this will be sooner rather than later. That said, everything that worked on Flash 10.3 worked just fine on the Xolo X900, games played fast enough, video played too, and things worked at decent speeds.

All of the above is something that I would consider ‘basic functionality’, and would have been very surprised if it didn’t work. Random apps were loaded and tried, and somewhat unexpectedly, they all just worked. While we didn’t try and differentiate between Dalvik and native ARM apps, nothing we tried caused any speed problems, and no lag was noted. In fact, apps tended to install very fast once downloaded, fast enough to be noticeable.

One app that we use a lot, Wunderground’s weather app, was quite a bit smoother when rendering maps, zooming and scrolling. Some of this was likely due to using it exclusively over Wi-Fi instead of 3/4G, but the rendering of maps once loaded was definitely better on the Xolo.

Nothing we tried broke on the apps side, other than the Opera bit noted above. This lack of problems was actually unexpected, and almost enough to say that you won’t have any problems running apps on an x86 phone. That said, if something is critical to your life, test it in the store before you shell out several hundred dollars for one of these. No app we tested had any notable speed problems either, stuff just worked.

Games:

Like apps, all the games we tried actually worked, something we were not expecting. Performance on the other hand was a much more mixed bag, and decidedly did not reflect the numbers in AnTuTu. If they had been 100% representative, the Xolo should have run everything out there more than fast enough, but as you will see, that is not the case.

Everything loaded and ran, we didn’t find anything that did not work, had glitches, bugs, or outright crashed. The compatibility on games seemed to be 100%, be the games native ARM or Dalvik code based. Unlike apps though, you could quite easily tell the difference between code types when gaming. How? Load times and game speed.

We will focus on two games that we play often, GRave Defense HD and Dungeon Defenders: Second Wave, various others reflected the performance of one of these two. GRave defense is in my opinion the best tower defense game out there, and DD:SW is a fun little time killer once you get your brain wrapped around the new paradigm of the game and the somewhat obtuse controls. Both ran without bugs, glitches, or obvious compatibility problems, but showed extreme speed differences.

GRave Defense ran absolutely perfectly, nothing was out of place, and nothing lagged, it just worked. On a Tegra 2 phone, it acts the same way, but on the highest of levels and settings, you can get the game to lag a bit. Unfortunately, we did not spend the hours it takes to get to that point on the Xolo, but unless you beat the game, you will never see lag on any modern phone. The Xolo worked perfectly, and ran smoothly at full speed on every level we got to.

Dungeon Defenders (DD) didn’t fare as well. It loaded and ran perfectly, the graphics were flawless, but the speed was notably lacking. Unlike GRave, it was obvious that the Xolo was running emulated code, and it took enough of a speed hit to make the game basically unplayable. That said, it technically worked perfectly.

Frame rates were so low that the game skipped and jerked around, and input lag made it unworkable. Setting the graphics options to the lowest level and turning off sound made the game look miserable, but didn’t bump the frame rates in to playable territory. In this case, working perfectly on a technical level didn’t include playable frame rates.

For many games, the loading seemed to be slower than it should have been, something we attribute to unpacking the code in native ARM mode. Games that didn’t show speed differences during play, a likely sign of Dalvik code, loaded fast in similar way to apps. When things ran in to ARM code, the loading speed took a second place to the Photon. Despite the slowdown, the loading speeds were never objectionable, it wasn’t seconds to minutes, but it did make slow loading games like DD much slower.

Subjectively speaking, we didn’t notice the slowdown until we put the phones side by side and tapped the icons at the same time. Only in the worst cases would any game take more than a handful of seconds more on the Xolo, but the load speed advantage of Penwell is definitely erased when you move to ARM code. It just goes to show the old adage that “What Intel giveth, Microsoft taketh away” has moved to a new era, are we now seeing, “What Intel giveth, Google taketh away”.

Overall, games were mixed. Everything loaded, everything ran, and there was never anything that we would chalk up to being problematic as far as bugs go. Unlike apps though, many games, especially 3D games, tax both the CPU and GPU, and are very input and output sensitive. Minor speed lags can make something useless even if it renders beautifully at high rez. In these cases, the Xolo X900 hits its first speed bump. Games are hit and miss, and mainly depend on code type. If you have a favorite game or series, try it out before you buy an x86 based phone, it will probably be fine, but if anything isn’t, this is where issues will crop up.

Conclusion:

We concluded our initial three stories about the Intel reference platform by saying that it looked good enough to do the job, nothing was missing, but software would be the key issue. Intel needed to put out a solid phone, not overpromise, not overreach, and just deliver something that worked. We didn’t think the hardware was lacking in any way, but would it work with today’s native ARM code base, and work at speed?

After living with the Intel x86 powered Lava Xolo X900, we can definitely say that it has met our expectations as “Just another ARM phone”, a make or break bar for Intel’s phone ambitions. It unquestionably made it. App compatibility was nearly 100%, only one thing we tried didn’t work, and that wasn’t necessarily the phones fault, it could have been a bad package on the store. This was much better than we hoped, and if Intel keeps playing whack-a-mole with incompatibilities and pushes updates like they promised, this simply won’t be an issue.

Performance on the other hand was, well perfect other than games. If playing the more advanced 3D games are something you are in to, well, you might want to avoid x86 based phones until they get enough of a market share for companies to compile their apps for x86, or just write Dalvik code. For the moment, native ARM games, especially CPU and GPU heavy ones, take too much of a speed hit to be playable. This problem cropping up is somewhat hit and miss, it may or may not hit the app you care about, and in a way, that is more of a problem than the speed. We expected all apps to see a similar performance degradation, and the fact that only a few games were ‘bad’ is much better than we hoped for.

Overall, Intel has done what they promised and exceeded our expectations. If you buy an x86 based phone, be it the Xolo X900 or any of the successors, it should just work the way you expect an Android phone to. The speed ranges between really fast for most things to really slow on some games, but that is liveable unless you are a die-hard gamer. It does just work.

I will end this with the most obvious question, would I buy one if I had to spend my own money? Unfortunately the answer is no only because of the game speed. That aside, everything else would put a Medfield phone on the short list for my next purchase, but for me personally, I like games. If you can live without bleeding edge 3D mobile games, I wouldn’t hesitate to cross-shop an x86 phone. I think Intel has hit the mark with it’s first phone, something we weren’t entirely sure of a week ago.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate