A Preview of the Krait-Based S4 Vs Tegra 3 and Exynos 4210

February 21, 2012
9 36 5 5

    While it’s not completely surprising if you understand where Krait is coming from as an architectural overhaul for Qualcomm’s CPU’s that finally has out of order execution, but also many performance improvements and bigger memory bandwidth, the Krait architecture and the dual core 1.5 Ghz S4 chip that is based on it manage to show some impressive numbers in benchmarks, making it the most advanced ARM chip that will arrive to market in the coming months.

    S4 vs Tegra 3

    The dual core 1.5 Ghz should beat Tegra 3 in most benchmarks, but also in real world use for most apps, especially those that are not heavily optimized for multi-cores. Tegra 3 has a great GPU, but the truth is that it still uses only 1 or 2 cores for most applications, and those cores only have an advantage of 100-200 Mhz in this case (Tegra 3′s first processor being 1.4 Ghz and the other 3 being 1.3 Ghz) compared to other dual core 1.2 Ghz processors in the market today.

    The S4 is also clocked a bit higher at 1.5 Ghz, but its new architecture it’s what’s really driving the performance, with an estimated 3.3 DMIPS/Mhz compared to the Cortex A9′s 2.5 DMIPS/Mhz and Qualcomm’s S3′s 2.1 DMIPS/Mhz (which would explain why a dual core 1.5 Ghz S3 was only as powerful or even slightly weaker than a dual core 1.2 Ghz Cortex A9 chip).

    This simple math shows that the S4 should be around 60% faster than a Cortex A9-based chip, at the same clock speed, but considering the S4 will come out at a slightly higher 1.5 Ghz per core compared to other 1.2 Ghz processors today, the speed improvement should be about double, even for single-threaded applications.

    Some benchmark scores taken by someone on a Reddit thread:

    Krait Scores:

    • Linpack (single) – 106.794
    • Linpack (multi) – 218.197
    • SunSpider – 1532*
    • BrowserMark – 110345
    • Vellamo – 2581
    • GL Benchmark 2.1 Egypt Offscreen 720p – 52.5
    • GL Benchmark 2.1 Pro Offscreen 720p – 62.2

    Tegra 3 Scores:

    • Linpack (single) – 47.2
    • Linpack (multi) – 135.9
    • SunSpider – 1695*
    • BrowserMark – 103768
    • Vellamo – 953†
    • GL Benchmark 2.1 Egypt Offscreen 720p – 64.4
    • GL Benchmark 2.1 Pro Offscreen 720p – 78.4

    S4 vs Exynos 4210

    It might not seem that the Galaxy S2 was launched too long ago, but it was actually launched in May last year for the first time, and that’s also when we first saw the Exynos 4210 chip. So that’s quite a bit of time, and we’re also expecting to see the Galaxy S3 drop off probably not too long after the first S4 processors arrive in products on the market.

    I would expect Samsung to use at least a quad core Exynos 4412 with a higher clock, or maybe a dual core Cortex A9 chip that we don’t know about with an even higher clock of 1.8 Ghz like the OMAP 4470, or maybe even an Exynos 5250 with a dual core 2 Ghz Cortex A15 chip. But we don’t have any numbers for those yet, so let’s see some benchmarks of how S4 compares to the current Exynos CPU (and others):

    One thing that’s kind of disappointing, and has always been about the Qualcomm chips are their GPU’s. The dual core Krait S4 still uses the old Adreno 2xx architecture, in this case an improved Adreno 225, but it still barely manages to beat the Mali 400 in some cases, while it gets beaten in others. I would expect the Adreno 3xx to be a much better when it arrives in Krait chips at the end of the year, but there’s isn’t any way to know if it will be better than Mali T600, Mali T650 or whatever Nvidia comes up with by then. Another thing I’ve noticed is that it seems to only support DDR2 RAM, and I’m expecting to see more DDR3 RAM in mobile devices by the end of the year, but again, perhaps that’s coming in the same time with the new GPU’s.

    If the S4 shows us anything is that the end is approaching for the Cortex A9 generation in high-end mobile devices, and sometime close to summer and later this year we should start seeing chips that push the performance forward significantly.

    Comments

    • http://www.facebook.com/boonesimpson Boone Simpson

      the razr and Droid 4 have roughly the same specs, can the minor chipset difference really account for the performance difference, or is the issue with all the bloatware MOTO put on it?

      • http://www.AndroidAuthority.com/ Darcy Alexander LaCouvee

        It’s most likely that the small difference is because of the bloat.

    • Graham Laight

      The Krait’s Linpack multi-processor score of 218 MFlops is roughly where the top supercomputers were 30 years ago. From what I know, I think that at that time, with good algorithms, it was possible to simulate the airflow around an aircraft adequately in a few hours. Although in practice, phones would be an unlikely choice of platform for large scale floating point arithmetic, you still need to think about this: at a party, would you want to be the one whose phone was unable to solve a large system of linear equations?

    • Mahsoom Moosa42

      i think the Mali gained an advantage in some tests since it uses Discrete cores not an unified shader architecture. The Adreno 225 MAY(some people think i agree with that) have underperformed because there is some kind of optimization.

    • GettCouped

      Krait needed to use the weaker Adreno to get out in time for Q2 2012. The next iteration will have a very competitive GPU. Besides the big feature is the integrated LTE modem. That should save a good bit of battery life. Not to mention dual channel memory and OOO execution.

    Popular

    Latest