Years ago, before we hit the power wall, CPU innovation happened at a slow but steady pace. Every five years or so we’d get a new microprocessor architecture and every couple of years we’d get smaller transistors.

The smaller transistors made chips run cooler and at higher clock speeds. The shrink in die area also paved the way for new features, but between major architectural shifts those features normally came in the form of larger caches.

Texas Instruments’ move from the OMAP 3430, used in phones like the Palm Pre, to the OMAP 3630 used in the Droid X is reminiscent of this sort of steady progress I mentioned above.

The OMAP 3430 was built on a 65nm process (like Qualcomm's Snapdragon), while the 3630 is a 45nm shrink (like Apple's A4). Architecturally the two SoCs are very similar. They both use a standard ARM Cortex A8 CPU paired with an Imagination Technologies PowerVR SGX 530 GPU. The two SoCs fit in the same size package (12mm x 12mm BGA) and are ball compatible. If a customer wanted to, it could simply drop in a 3630 into an existing 3430 design with minimal engineering efforts.

Note that the most direct competitor to the 3630 is Qualcomm’s Snapdragon. While TI uses a standard Cortex A8 core from ARM, Qualcomm designed its own low power ARMv7 based core that is similar, but not identical to the Cortex A8. Both are dual-issue, in-order architectures - they’re like the original Pentium, but in your phone. Qualcomm also integrated the cellular modem into the Snapdragon SoC while TI’s OMAP 3 is a strict application processor - the modem is housed in a separate chip.

On the CPU side TI doubled the L1 cache of the 3430 to 64KB (32KB instruction, 32KB data). The L2 cache remains unchanged at 256KB. We won’t get a larger L2 until the OMAP 4, which will ship with a 1MB L2 shared among its two cores. There are the usual tweaks and bug fixes which may improve performance per clock a little bit over the 3430, but overall the 3630 just gets a larger L1 as a result of the die shrink - oh and a much higher clock speed.

The Cortex A8 now runs at up to 1GHz. The OMAP 3430 topped out at 800MHz in shipping configurations but most vendors ran it at sub-600MHz speeds to save power. The 3630 in the Droid X runs at a full 1GHz. It's worth pointing out that Qualcomm was able to hit 1GHz on a similar architecture at 65nm by designing its core from the ground up. There's clearly value in these custom designs from a performance and time to market standpoint. These advantages will only become more critical as the SoC performance wars heat up.

Power Consumption

The OMAP 3 as well as Qualcomm’s Snapdragon SoC support dynamic voltage and frequency scaling. Based on application demand and cues from the OS the CPU clock speed and voltage can vary. The Cortex A8 in the OMAP 3630 will drop down to the low 300MHz range when idle at the Android home screen and ramp all the way up to 1GHz when needed.


TI's OMAP 3630 running at full tilt


...and automatically underclocked to 446MHz under lighter load

This is nothing unique to the OMAP 3630, but it shows that frequency and voltage scaling is alive and well in Android.

TI supports full power and clock gating. All major IP blocks are placed on their own power islands, so based on input from the OS the CPU, GPU, video decoder, etc... can ramp down depending on the application needs at the time. The power and clock gating is no more granular in the 3630 than it was in the 3430. What has changed however are operating voltages.

While the 3430 needed 1.35V to hit 720MHz, the 3630 can reach 1GHz at around 1.26V. That still sounds a bit high to me but at the same clock speed, thanks to voltage scaling, you can drop power by around 30% compared to the 3430.

Coupled with the dynamic voltage/frequency scaling of all OMAP 3 parts this means that overall power efficiency should be better on the 3630 vs. the 3430. The added CPU performance in the form of larger L1 caches and a higher clock speed should make tasks complete quicker and allow the 3630 to get to a lower voltage state than the 3430 was ever able to reach.

Samsung has already shown off 1.2GHz+ versions of its Cortex A8 based SoC at 45nm, so I would expect to see higher clocked versions of the 45nm OMAP 3 family to follow at some point in time.

Comparisons and What's In the Box The GPU Performance Showdown: Snapdragon vs. OMAP 3630
Comments Locked

89 Comments

View All Comments

  • TareX - Thursday, July 22, 2010 - link

    Why are all the Samsung Galaxy S phones missing from your charts and comparisons? I wanted to see how Hummingbird compared and how the Super AMOLED fared against their Droid X counterparts...
  • strikeback03 - Tuesday, July 27, 2010 - link

    Maybe because the first US version launched a day before the review went up, and they weren't sent a review sample.
  • enealDC - Thursday, July 22, 2010 - link

    Great job!
  • Juniper Research - Friday, July 23, 2010 - link

    Very interesting article... we have this week published a new report on smartphones and a free whitepaper is available to download here... http://www.juniperresearch.com/reports/next_genera...

    John Levett
    Marketing Executive at Juniper Research
  • Homefries - Saturday, July 24, 2010 - link

    First off, great review Brian.

    However, while you did a very thorough comparison of the Droid X to other Android devices, you barely mentioned the real competitor the Droid X has to stand up against: the iPhone 4.

    Readers want to know if the Droid X is the best phone on the market - the whole market - not just the market subset dedicated to Android devices.

    Like the majority of the tech media, your review furthers the notion readers belong exclusively to either the iOS camp or the Andriod camp. This is simply not true. Informed readers, like the ones that peruse Anandtech, want to buy one smartphone, regardless of any marketing slants, that is the best.

    Your review of the Droid X should have helped us answer the question, "Should I buy the Droid X or the iPhone 4?" But, it did not.
  • strikeback03 - Tuesday, July 27, 2010 - link

    Well, I think the conclusion in the iPhone article and some earlier Android articles applies, there is no "best" for everyone. Some people love Apple and the Apple way of life, some people won't touch it. As the iPhone is currently limited to AT&T, that is going to deter a lot of people. How large a pocket/bag you plan on carrying the phone in might make size differences more important to some than others. So while I am not one of the readers calling for no subjective opinions (it is interesting that FroYo feels significantly faster/different) I think it is still up to each buyer to decide what is most important to them.
  • Electrofreak - Saturday, July 24, 2010 - link

    Brian and Anand, are you sure you're using the correct information regarding the SoC in the Droid X? I believe it is a OMAP 3640, not a 3630, as the maximum recommended clock speed of the 3630 is described in numerous places across the net as being 720 MHz, while the max clock speed of the 3640 is described as 1 GHz. In addition, the max recommended clock speed of the 3430 in the Droid was 600 MHz, not 800.

    The information I cite above is widely available across the web... if you've got inside information the rest of us don't have, by all means let us know. But as someone who has written articles of my own on ARM SoCs and follows ARM industry news closely, I suspect that your data may not be 100% correct.

    Regardless, I do have to thank you for writing some of the most informative hardware articles on the net. I appreciate it!
  • Electrofreak - Saturday, July 24, 2010 - link

    ...and I just found this: http://e2e.ti.com/blogs_/b/mobile_momentum/archive...

    So, per TI's blog it is the 3630... now we just need an explanation of the other info on the web that describes the 3630 as maxing out at 720 MHz.
  • Brian Klug - Monday, July 26, 2010 - link

    Yeah, TI's documentation is a bit outdated. Anand tackled the SoC part, but the 3630 is indeed a 1 GHz part, it isn't the 3640 guaranteed. There was a lot of confusion online about it, but Anand got the official word. ;)

    I agree, back when I did my other OMAP 3 piece it was 720 MHz.

    -Brian
  • Electrofreak - Saturday, July 24, 2010 - link

    Looking forward to that Hummingbird review Brian. I hope you're able to dig up some info that I wasn't able to when I wrote my article (http://alienbabeltech.com/main/?p=17125) back in April.

Log in

Don't have an account? Sign up now