We had a chance this evening to take a closer look at NVIDIA's Tegra 4 and Tegra 4i chips, and with Tegra 4, a chance to run some benchmarks. We also took a quick look at NVIDIA's reference design phone for Tegra 4i, the Phoenix (though we were only allowed to look - not touch).


Left to right: Phoenix, Tegra 4 board, Tegra 4i board

We'll start with the Phoenix reference phone, because there isn't much to say. The phone is in a very early state, and the non-working samples we were allowed to hold (the powered on device was eyes-only) felt like engineering samples... because they are. It's really not worth getting into much about Phoenix, since it's something consumers won't see - the platform will end up as more refined derivations from various OEMs. If you want the skinny on what is inside Phoenix, check out our Tegra 4i announcement post.

What we saw of Phoenix, basically, was a demo of Real Boxing, and it ran pretty smoothly. Tegra 4i is using a slightly toned-down version of Tegra 4's GPU, so we'd expect nothing less.

Of more interest to you, perhaps, were the benchmarks we were allowed to take using NVIDIA's Tegra 4 tablet reference platform.


wm__DSC0317 wm__DSC0319

Top: Tegra 4 tablet reference platform, Bottom: Tegra 4 / 4i, Phoenix

The results were what you would hope for: pretty amazing. While I'm not exactly convinced about the ability of benchmark results to translate very well in real-world use scenarios, there's no denying that what we saw Tegra 4 do in said benchmarks was impressive. Recently, HTC's new Snapdragon 600-powered One phone was benchmarked in popular suites like Quadrant and AnTuTu. Tegra 4 blew away the 600 in both of them. We ran Quadrant, AnTuTu, Vellamo (HTML5), and SunSpider. Here are the results.

quadrant_result tegra4_sunspidertegra4_vellamo_html5 tegra4_antutu_score

We don't have any side-by-side comparisons here (well, any of value), but for reference, here's how HTC's One phone scored in three of the above benchmarks, versus Tegra 4.

  • AnTuTu (higher is better - holistic benchmark)
    • HTC One: 22678
    • Tegra 4: 36317
  • SunSpider (lower is better - web benchmark)
    • HTC One: 1124ms
    • Tegra 4: 494ms
  • Quadrant (higher is better - holistic benchmark)
    • HTC One: 11746
    • Tegra 4: 16593

So, there you have it. We also saw a slide about Tegra 4's power consumption compared to the HTC DROID DNA, which uses a Qualcomm APQ8064 quad-core chip.


As you can see NVIDIA's numbers (mW) are smaller than Qualcomm's, and that's better! We didn't really hear any other new information, though NVIDIA has promised more demonstrations of Tegra 4's new features on the MWC show floor, starting tomorrow.

David Ruddock
David's phone is whatever is currently sitting on his desk. He is an avid writer, and enjoys playing devil's advocate in editorials, and reviewing the latest phones and gadgets. He also doesn't usually write such boring sentences.

  • Jason Banich

    Jesus, those benchmarks are insane

    • http://wakoopa.com/yo2boy yo2boy

      I have regained faith in NVidia! Holy shit.

    • sri_tech

      Dont get too excited guys.
      Here is the in depth read about tegra 4 performance and power consumption from anandtech.com


      "NVIDIA’s Tegra 4 is a significant step forward in both CPU and GPU performance. Although Tegra 3 was decent in both areas, Tegra 4 really moves things forward. ARM’s Cortex A15 is an excellent performer, although that performance comes at a high power cost. In a tablet, as we’ve already seen with Google’s Nexus 10, the power consumption associated with the Cortex A15 core is manageable. If NVIDIA’s data is to be believed however, Tegra 4 can get into a smartphone just by aggressively controlling frequencies. At reduced frequencies, Tegra 4 can draw less power than Tegra 3 but with no performance advantage. NVIDIA could then scale up performance (and power) to offer an improvement over Tegra 3. The real question at that point is whether or not Qualcomm’s Krait 300/400 designs offer better efficiency at these intermediate points on the performance/power curve. We’ll be able to find out for sure later this year when both Tegra 4 and Snapdragon 600/800 based devices are shipping."

      For same performance and low power than tegra 3, tegra 4 needs to operate at 825Mhz frequency.

      • ickkii

        I respect what you're saying, but I may have been distracted by the cash I was throwing at my computer screen while I was reading this.

      • didibus

        True, but one thing is sure, the next Ouya with Tegra 4 will be able to crank it to the max and it will deliver.

  • HellG

    eh, how i wish that real devices would get such score :/
    reference devices always gets much better results as the whole unit is optimized for the CPU performance, just check the Snapdragon performance back in the days of the HTC sensation, it sucked but the reference unit for the snapdragon that anandtech reviewed (he is a known isheep) gave MUCH better score than the sensation EVER got :/

    • chris125

      That's what happens when they throw skins on top. Hinders performance

      • ssj4Gogeta

        The reference platforms aren't optimized for power savings. The chips throttle down more aggressively in retail versions.

    • Cycleops

      Anand a "known isheep"? Really? If by that you mean giving respect or dissatisfaction where it's due, regardless of vendor, then sure.

      • HellG

        Did you check his site phone reviews? they ALWAYS show benchmarks/ battery tests where the iPhone 5 is on top, they never show others where iPhone 5 is an underdog to give the illusion that the iPhone 5 is ALWAYS on top! not sure what you call that but that's an isheep to me

        • Cheeseball

          Did you ever think that it's because the Apple Swift SoC is Cortex-A15 based and equipped with the SGX543MP3? The only chip that can nearly match it is the APQ8064, but even that has to be throttled due to the evident heat issues.

          • HellG

            you are so wrong on many levels, first its not Cortex A15 its a mix like the Krait core, sec its not on top of many benchmarks just check GSMarena/Phonearena/PocketNow/engadget/theverge reviews, same for battery tests which again has nothing to do with the A15 as it even consume more power, now check the reviews of any phones and i mean ANY phone in anandtech now go and compare the results with the other sites the bias on his site is so disgusting.

          • Cheeseball

            While I'll admit that AnandTech's performance charts show the iPhone 5 to be a bit biased (it *is* topping most of the charts), the other review sites you mentioned all show the iPhone 5 to be quite high as well, showing that the Swift+SGX543MP3 is heavily competitive, in both battery life and performance.

            I didn't say that the Swift uses reference Cortex-A15 cores, but is *based* off it, just like the Snapdragon Krait you mentioned, but having support for more instruction sets and extensions like an A15.

            While I can agree Anand may be a known "iSheep" in the mobile division, the Apple Swift A6 is a powerful SoC that I can only wish we have for use with Android. The last PowerVR chip we had was the SGX 540 in the Galaxy Nexus, and that was quite good even though it wasn't multi-core like the Adrenos or the Mali-400MP.

          • HellG

            and that's my point exactly, i can't deny the power of PowerVR chips, in matter of facts if you went on my profile and checked my other posts you will find one a couple of days ago where i was wishing that other OEMs learn from powerVR to make such amazing graphical chips, i got banned from android authority after calling anand la shimpi an iSheep as well after the OP called him one of the most trusted in the phone industry or something like that lol

  • Kokusho

    So, Qualcomm‘s CMO Anand Chandrasekher ? Care to explain how the Tegra 4 is similar to the S4Pro ? How the s600 is way better and How your futur s800 will totaly obliterate it ?
    Seems like a lot of trash talk without any clue to me.

    • chris125


    • HellG

      I dont know if you mean my post or the OP, if you are talking to me how ever i dont know what trash talking do you mean, all i said that these unites are optimized to show the chip performance and real life units will underperform, and the Sensation was using a MSM8260 not an S4 pro o_O

      • RajivSK

        He's clearly talking about some Qualcomm presentation of their new chips. It wasn't directed at you or anyone else here.

      • A.Noid

        Yeah, Qualcomms CMO said that, not you. Hes referring to a statement he made about Tegra 4. Clearly Tegra 4 is smoking it. I hope they have a 7" tablet running that. Try to keep up. ;)

  • adi19956

    Loving the hard and software buttons of the Phoenix

  • squiddy20

    Anyone find it odd how, in the first screenshot of the reference device, it has software buttons AND capacitive buttons? Wouldn't Android by itself detect the capacitive buttons and automatically hide the software ones?

    • http://www.androidpolice.com/ David Ruddock

      It's an early engineering prototype. I'm guessing they uh, need to fix that.

      • Chronus719

        Or maybe they were trying to show that manufacturers could use either? (A stretch, I know).

        • http://www.youtube.com/kimirPORTALS kimir

          Smaller number of pixels to push around, better benchmarks... hmmm

          • http://www.facebook.com/profile.php?id=1548515355 Damir Delivuk

            agree ;-p

      • http://twitter.com/Defenestratus Defenestratus

        Also if its an early engineering prototype, software is clearly in even more infantile state of development. On-screen buttons would likely be unreliable at best ...

  • jalen venable

    OMG those benchmarks are surreal. I love the look of Phoneix

  • Chronus719

    I would have guessed that it would be faster than the 600, but I'm interested to see how it fares with the 800.

  • GraveUypo

    holy fuck i can say i'm genuinely impressed. i didn't expect anything CLOSE to this.

  • CoreDuo08

    While it's nice that for once they did a power comparison to a device that actually exists and didn't hide the numbers behind an unlabeled Y axis, one area of serious concern is what happens when you wind up those 4 big A15 cores, how well NVIDIA optimized the power management and if they will make sure that OEMs are implementing it correctly. The power consumption figures from the Exynos 5 Dual weren't exactly heartening.

  • http://www.facebook.com/profile.php?id=100001482286220 Karan Ocs

    HTC One Should have been the first TEGRA 4 Phone Just like ONE X did with TEGRA 3 !

    These Benchmarks are INSANE! :O

    HTC Did i Ever Tell you the definition of insanity?!

  • http://twitter.com/yokuyuki Ed Chen

    The benchmarks seem a bit unfair to compare. First, it shows the benchmark results from a phone running Snapdragon 600 vs the Tegra 4 tablet platform which doesn't the clock limits imposed on it like it would on a phone platform. Then, it shows the Tegra 4 using less power than the S4 Pro implying that the Tegra 4 is faster and uses less power than the Snapdragon platform when the Tegra 4 in phone benchmark is clock limited to consume less power.

    • HopelesslyFaithful

      yea that was what i was wondering. Does anyone know how padded those benchmarks are. Also the pro and 600 are two different SOCs right? I am a bit out dated on Snapdragon SOCs. The new scheme lost me.

  • mercado79

    The first phones featuring these chips aren't expected until late 2013 / early 2014. So comparing benchmarks to chips already available (or soon to be) is next to pointless. What will Qualcomm or Samsung or x chip manufacturer have out by then?

    • http://AndroidPolice.com/ Liam Spradlin

      I don't think this post is meant to put down current processors, but rather showcase what it is that NVIDIA is working with right now, and how much better the next generation will be compared to the current. Since we don't have access to Samsung's unreleased processors, there would be no point speculating about them when we can discuss NVIDIA's developments right now.

  • Ben

    Imma ditch HTC One for Tegra 4, sunspider benchmark just blew every shit out of the water.

  • Jeff

    Cool. I'm not convinced until someone benchmarks the I/O :). That's the huge flaw in Tegra 3...

    • Grahaman27

      thats part of the benchmark.

  • Nathan Borup

    I want... right now

  • Stacey

    Yay! Let's drool over benchmarks for an SoC that has a TDP too high to ever be feasible in a real phone (although the 4i should work...unfortunately it's not as mind blowing).

  • Monji

    Watsup Qualcomm Snapdragon 600 is better than tegra 4 and 800 brings the gaps ever farther... Shame

  • guest

    "NVIDIA claims to offer performance competitive with Apple’s iPad 4".. Lol, we'll probably have an iPad 5 before Tegra 4 is available :>. Still, the iPad 4 is pretty impressive.