Next year is looking to be a very important year for smartphone and tablet performance. Just as we saw widespread migration to the ARM Cortex A8 and Qualcomm Scorpion CPU cores in smartphones in 2010, in the next 12 months we will see the first tablets and smartphones based on dual-core SoCs from TI, Qualcomm and NVIDIA. The long awaited Tegra 2 will start shipping in smartphones and tablets in early 2011, and Qualcomm will have its own 45nm dual-core Snapdragon SoCs featured in devices as well. Today TI is announcing more details on its OMAP 4440, a high performance dual-core Cortex A9 SoC slated for production in the second half of 2011.

The OMAP 4430 is a 45nm SoC expected to replace the current 3630 and 3430 used in high end smartphones today. The 3630 will move further downstream and exist within more mainstream smartphones.

The 4430 features a pair of ARM Cortex A9s running at 1GHz with a PowerVR SGX 540 GPU from Imagination Technologies. The Cortex A9 features a shorter pipeline than the current A8 and Scorpion cores and thus should boast a higher IPC. It also enables some amount of out-of-order execution.

The A9 cores share a 1MB L2 cache and are paired with two 32-bit LPDDR2 memory controllers. The memory bus can support LPDDR2 devices at data rates up to 400MHz. This gives the OMAP 4430 the potential to offer 4x the memory bandwidth of the 3630, which means better CPU and GPU performance as well as more flexibility with video.

The architectural improvements should result in significant performance gains in everything from OS interaction to web browsing. We’ve already seen real world improvements as high as 50% vs. existing A8/Snapdragon platforms.

The PowerVR SGX 540 GPU is used in Samsung’s Hummingbird SoC, which we’ve seen in the Galaxy S line of smartphones. This is already the fastest ultra mobile GPU we’ve tested, so its well suited for use in the 4430.

Today’s announcement isn’t about the 4430 however, it’s about the 4440. A higher performance OMAP4 variant, the 4440 is still built on a 45nm process, it still features a pair of Cortex A9s and a PowerVR SGX 540 GPU - but it just runs faster.

OMAP 4440 vs. 4430 Feature List - Provided by TI
Feature Benefit
Two ARM Cortex A9 MPCores @ 1.5GHz each 50% increase in overall performance, 33% reduction in webpage load time
End-to-end graphics acceleration enhancement (triangles per second, fill rate and shaders) 25% increase in overall graphics performance
Support for HDMI v1.4 3D modes Full 1080p HD S3D playback to 3D TV
1080p60 video format support 2x higher performance video playback
Support for up to two 12-megapixel cameras in parallel Higher stereoscopic resolution encode enabling stereo photography, which meets the same resolution previously experienced only with 2D photography
IVA 3 multimedia hardware accelerator Industry’s highest quality video playback at low bit rates
Complete pin-to-pin hardware and software compatibility Easy migration from the OMAP4430 processors

The A9s run at 1.5GHz and although TI isn’t announcing the SGX 540’s clock speed, it promises a 25% increase in overall graphics performance (triangle rate, fill rate). Presumably the GPU is clocked 25% higher than in the 4430 (which could be as high as 400MHz).

There are more video features on the 4440 as well. You get 60 fps 1080p decode support (bitrate limitations aren't completely fleshed out, but TI has some details in a whitepaper- PDF link), and the SoC supports up to two 12-megapixel camera sensors (potentially for use in high resolution stereoscopic 3D photography). HDMI 1.4 and stereoscopic 3D are also supported by the SoC.

OMAP 4440 Mobile Video Teleconferencing Features - Provided by TI
Mobile video teleconferencing component OMAP4440 processor-enabled feature
High-quality mobile video conferencing Improved video quality in low-light conditions; video stabilization
Chat software (i.e., Skype or Google Talk) Video codec support includes H.264, VP7, H.263, SVC, and more
Peer-to-peer (1 local user with one other user) chat functionality 1080p mobile video conferencing
Multi-chat (1 local user with up to 4 other users) functionality 720p resolution with stereo audio support
Cloud access for simultaneous application support (e.g., browsing the web while chatting or document sharing) Optimized symmetric multiprocessing architecture to deliver low latency and high bandwidth support

The OMAP4440 is pin compatible with the OMAP4430 and thus can be dropped into existing OMAP4 designs if need be. The SoC will be sampling in Q1 2011 and be in production in 2H 2011. I’d expect the 4440 to be used in tablets while the 4430 seems more like a smartphone SKU.

POST A COMMENT

38 Comments

View All Comments

  • blueboy_10 - Thursday, December 23, 2010 - link

    Hopefully the will fix the problem with overheating in smartphones, cause it was reported on one of the tech sites (can't remember which) which stated that they had run test on triple-core processors on a prototype phone and the freaking thing actually caught on fire and melted. I don't know about you, but I wouldn't want to have that burning sensation in my pants. OUCH!! But seriously, all joking aside, if TI and Samsung do get their act together and put this tech in phones hopefully by mid-2010, then we'll really start seeing the competition heat up. It will be just like the PC processor old days. Can't wait!! - BLUEBOY Reply
  • zoubi - Wednesday, December 08, 2010 - link

    So when do we get a 4430 or 440 in a Gumstix?
    I can't wait to make super small video projects that actually can process HD video (pandboard is too big)

    That and a 5MP+ HD able camera going directly on the card with provided drivers and SOC would be so awesome.
    Reply
  • wasnt - Wednesday, December 08, 2010 - link

    Doesn't even matter... with battery life falling with ever update, these should get about 1.5 hour talk time in a smart phone.

    Battery tech updates is what I want to see! This coming from an HTC Evo 4g user.
    Reply
  • beginner99 - Wednesday, December 08, 2010 - link

    lol. That's why I stll have my ancient and simple mobile phone. Weighs like half than all these smartphones, much smaller and battery life is still very good after like 4 years especially compared to these smartphones.
    Ok, and I don't need the functionallity they offer. I mean I have a pc at work and at home and when I'm not I sure don't need to be gaming or surfing too. And if it is unavoidable it can still do email and run opera mini. ;)
    Reply
  • R3MF - Wednesday, December 08, 2010 - link

    is it only the 4440 SKU that is due 2H 2011, or the entire Omap4 family?

    bad news if the latter, because Omap4 was expected to be in production Q1 2011.

    in short, can we expect the 4430 any sooner?
    Reply
  • Brian Klug - Wednesday, December 08, 2010 - link

    So OMAP4430 we'll see well in advance of OMAP4440. Expect OMAP4430 devices right after the new year (early 2011) and OMAP4440 will be the drop-in upgrade from that point (mid 2011).

    -Brian
    Reply
  • o.mar - Wednesday, December 08, 2010 - link

    " It also enables some amount of out-of-order execution. "

    Weren't they already OoO?

    Are there different levels of OoO?
    Reply
  • Tuna-Fish - Wednesday, December 08, 2010 - link

    > Are there different levels of OoO?

    Yes, absolutely. A Nehalem does a lot more reordering in a lot more situations than f.e. a pentium pro, and that is a huge part of what makes it faster. However, that is not the case here.

    > Weren't they already OoO?

    Nope. Cortex A8 is superscalar -- or it can execute multiple instructions per clock, as long as they don't depend on each other. However, if the next instruction in the stream depends on some half-completed result, it didn't bypass it and try to execute the next one. Cortex A9 does this -- this doesn't help well-optimized compute-heavy code that much, but is a *huge* win on branchy code that does a lot of loads. You know, pretty much everything that would be sane to run on a phone. :)

    Even still, OoO is *not* the big deal here. The memory interface is. I've done some profiling on a Cortex A8, and on it, damn near everything is bandwidth limited, even many things you'd really expect not to be. A single pathetic lpddr interface just wasn't that good of an idea, especially as it was shared with the GPU and CPU. OoO is just icing on the cake -- and it also helps hide memory latencies.

    I expect this to be *at least* twice as fast as a similarly clocked A8 in single-threaded games.
    Reply
  • Tuna-Fish - Wednesday, December 08, 2010 - link

    Apparently I really need to learn to proofread -- s/Cortex A8/OMAP3430/ for the last two paragraphs. Reply
  • metafor - Wednesday, December 08, 2010 - link

    It absolutely helps compute-heavy code. Especially since both NEON and ARM core only has access to 16 registers each (only 14 for ARM core, since one is used as the PC and the other the link register).

    OoOE is one of the best ways to get around register starvation. On an architecture with more registers (say, IA-64 or PPC), it's less of an issue.

    Even so, being able to reorder low latency instructions (VADD, VAND, etc.) around high-latency instruction (VMLA) is very advantageous given limited execution resources.
    Reply

Log in

Don't have an account? Sign up now