Qualcomm's MSM8660: Does Dual-Core Hype Translate Into Practical-Benefit Reality?
Here’s the Quadrant Advanced benchmark results screenshot of the MDP with Vsync on:
Next comes my Google Nexus One:
And Motorola Droid:
It’s now time to turn to the tablets. Starting with the smaller 7″ system, the NOOKcolor’s seemingly stunning performance belies its comparably conservative CPU and other system resources:
But look closer and its I/O benchmark result dominates the total, along with vastly overshooting the I/O performance of all other systems I tested save the MDP. Apparently, it seems, Quadrant’s code has a glitch with some hardware configurations…until I have a better understanding of what Aurora Softwork’s program is doing, I’m treating both platforms’ numbers as highly suspect. By the way, I was (for some unknown reason) unable to convince DDMS to ’see’ the NOOKcolor, so I instead used ShootMe to snag the screenshot.
Next comes the ViewSonic gTablet, whose 10″ widescreen LCD supports a 1024 x 600-pixel resolution:
Once again, I struggled to snag a screenshot, in this case because Android 2.x-based systems using the Nvidia Tegra 2 SoC’s GPU don’t apparently support the fb0 frame buffer which both DDMS and third-party utilities attempt to tap into. So I snapped a photo using the Nexus One’s built-in camera instead.
And finally we have the Motorola Xoom outfitted with a 1280×800 pixel 10″ widescreen LCD. Since I was worried that my now-obsolete Quadrant Advanced version might not adequately support Android 3.1 (a concerned that gained credibility when, after completing my testing, I read CrunchGear’s coverage), I ran the Xoom through both Quadrant Advanced 1.1.1:
And newer, but defeatured, Quadrant Standard 1.1.7 from the Android Market:
Here’s the summary data table:
|Qualcomm MDP MSM8660 (Android 2.3.3 subset) (vsync on)||1.5 GHz dual-core Qualcomm MSM8660||3023||5646||2680||5541||380||870|
|Qualcomm MDP MSM8660 (Android 2.3.3 subset) (vsync off)||1.5 GHz dual-core Qualcomm MSM8660||2877||5796||2632||4773||267||916|
|My Google Nexus One (Android 2.3.4)||1 GHz Qualcomm QSD 8250||1384||4602||1110||659||225||324|
|Google Nexus One (Android 2.2)||1 GHz Qualcomm QSD 8250||1340||4598||1040||539||174||351|
|Google Nexus One (Android 2.1)||1 GHz Qualcomm QSD 8250||~500|
|My Motorola Droid (Android 2.2.2)||550 MHz Texas Instruments OMAP 3430||842||2726||697||387||158||240|
|Motorola Droid (Android 2.x)||550 MHz Texas Instruments OMAP 3430||353||398||681||355||74||257|
|My B&N NOOKcolor (CyanogenMod 7.0.3 aka Android 2.3.3)||800 MHz Texas Instruments OMAP 3621||1630||3451||862||3380||172||284|
|My ViewSonic gTablet (CyanogenMod 7.0.3 aka Android 2.3.3)||1 GHz dual-core Nvidia Tegra 2||1855||5292||2302||897||198||585|
|My Motorola Xoom (Android 3.1, Quadrant Standard 1.1.7)||1 GHz dual-core Nvidia Tegra 2||1813|
|My Motorola Xoom (Android 3.1, Quadrant Advanced 1.1.1)||1 GHz dual-core Nvidia Tegra 2||1737||4497||2442||969||84||692|
A few other comments; first off, you might wonder why the Motorola Droid scored so much higher on Android 2.2.2 than it did on whatever prior Android 2.x release was in the Quadrant Advanced database (the handset originally shipped with 2.0.1, and was subsequently upgraded to 2.1 and then 2.2). I suspect this is a result of the Dalvik Virtual Machine performance optimizations made in Android 2.2.
When running the 3D benchmarks, which involve rendering three consecutive animation sequences (migration through several rooms and a set of stairs, a moon rotating around a planet, and a rotating DNA snippet), I noticed a visual glitch. Here’s what the images are supposed to look like, from the Motorola Droid (which matches what the Motorola Xoom delivered):
Now take a look at the MDP (whose graphics output matched that of the NOOKcolor and gTablet)
Again, as previously discussed, disregard the pinkish hue which is a DDMS screen capture bug artifact. Texture-deficient Sequence #3 rendered fine, but the superimposed texture images which should be visible on Sequence 1 and 2’s polygons’ surfaces are absent. Here’s another should-be shot from Sequence 1 on the Droid:
While I don’t think this output glitch invalidates the rendering-focused benchmark results, it suggests that the graphics hardware, software driver, benchmark utility, or some combination is currently suboptimal in its implementation. An even worse fate befell the Nexus One; its Sequence 1 and 2 outputs were completely blank (i.e. black screens, save for the frame count/rate display at bottom).
So do I believe there’s justifiable value in a dual-core CPU and advanced graphics processing subsystem? My answer is a ‘yes’, albeit a somewhat qualified one. Perhaps obviously, should such a leading-edge SoC dramatically increase the system bill-of-materials costs versus a mainstream alternative, the resulting boutique-priced system would have limited market appeal. And were the SoC’s advanced processing capabilities accompanied by prodigious power consumption, the resultant limited battery life (or, alternatively, large and heavy battery requirements) would hamper the system design, therefore dampening the market embrace. Assuming that neither of these potential issues come to pass, however, a multi-core CPU and advanced GPU hold substantial appeal, especially as smartphones become increasingly functional approximations of their computer precursors, and double-especially with large-screen tablets versus smaller-screen handsets.
Admittedly, Nvidia’s Tegra Zone utility for Android (which showcases Tegra 2-optimized apps, but whose conclusions are equally apt for the other advanced ARM-based SoCs such as the MSM8660) focuses the bulk of its attention on games, which not only push abundant high-quality polygons and pixels to the screen but also deliver accompanying physics, sound and other processing-intensive realism metrics. But just as the architectural and clock speed discrepancies in my two Android handsets, the Motorola Droid and Nexus One, are abundantly evident whenever I use them, a dual-core SoC with leading-edge graphics (and, let’s not forget, still and video imaging) processing capabilities will be better still.
If nothing else, such a SoC will enable operating system and application developers, along with their hardware partners, to move beyond today’s neutered pseudo-multitasking capabilities into the full multitasking future. And after having put Qualcomm’s dual-core Scorpion architecture through its paces, I frankly can’t wait for Krait.