18
Dec
1[4]

With CES coming up next month, we're starting to see more and more leaks that will likely materialize on the tradeshow floor. It looks like Chinese site Chip Hell has gotten its hands on one of NVIDIA's little secrets: the Tegra 4, which is codenamed Wayne.

1

The T4 looks to trump its predecessor in every way - much like the T3 did the T2. It uses the same 4-PLUS-1 core configuration as the Tegra 3, but it does so in a 28nm package - making it not only faster, but more energy efficient, as well. It also boasts six times the GPU cores as the Tegra 3, with a staggering 72 cores on board. As such, it supports resolutions up to 2560x1600 and full 1080p output at 120Hz.

I'm sure the T4 will have some more tricks up its sleeve, as well - and we suspect to hear the full skinny at CES.

Can't. Wait.

[Chip Hell via Engadget]

Cameron Summerson
Cameron is a self-made geek, Android enthusiast, horror movie fanatic, and musician. When he's not pounding keys here at AP, you can find him spending time with his wife and kids, plucking away on the 6- or 7-string, or watching The Texas Chainsaw Massacre on repeat.

  • Bas

    Holy droid!

  • deltatux

    Now to hope that NVIDIA has had some common sense and put a dual channel memory configuration. The Tegra 2 and 3 were really slow because it lacked dual channel memory. Both SoC's GPU were nothing to write home about when compared to the ARM Mali 400MP. I'm hoping NVIDIA actually steps up their game for once.

    • http://www.androidpolice.com/ Cameron Summerson

      Looks like it'll support dual channel memory.

      • deltatux

        Ah, excellent, that means Tegra chips will stop being bottlenecked in memory performance. Which means less stuttering when multitasking or better framerates while gaming.

    • defred34

      How is dual channel helpful? Does anyone out there use dual channel?

      • DCMAKER

        every computer in the world uses dual channel.....dual channel is paramount on low power ram because they have low bandwidth compared to regular ram hence how dual channel helps

      • deltatux

        Almost every single modern SoC out there except the Tegra 3 supports dual channel memory.

    • PINJ

      I Resend That. Any Arm Mali 400MP In Any St Nova Is Nowehere Near As Good As The Tegra 3 GPU.

      • deltatux

        Please re-read my post, you clearly didn't lol.

  • DCMAKER

    Anything on transistor count or die size or even TDP?

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

      The slide's all there is.

      • DCMAKER

        any rumors?....errr reputable rumors

  • Skitshin

    Party on.

  • http://jordanhotmann.com/ Jordan Hotmann

    I'm hoping the cpu is based on the ARM a15 architecture, anyone know if it's a9 or a15?

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

      It's A15, almost certainly - if the low-power core is 28nm, I highly doubt the high-powered ones are A9.

      • http://jordanhotmann.com/ Jordan Hotmann

        Good deductive reasoning skills :)

      • ssj4Gogeta

        Sorry but I didn't get the reasoning? What does 28nm have to do with microarchitecture?

        • DCMAKER

          i thought i read something before about A9 are made on 32nm HK and A15 are made on 28nm for some odd reason. At least thats what i remember reading on it. Was an article on Samsung chips but i only grazed(SP?) it. I don't devote a huge amount of my memory to cell phone tech....got better uses for it right now

          • ssj4Gogeta

            They're made on what's available at the time. Samsung, for example, fabricated A9's on 45nm first (Exynos 4210) and then moved to 32nm (Exynos 4212). They also have A15's on 32nm (Exynos 5250).
            Heck, you could make the original Pentiums on 22nm.

      • DCMAKER

        almost never do they make part of a CPU two different fabrication sizes. I can ohnly recall Intel did this once maybe twice with there CPUs where CPU was one size and GPU was another but i still don't get the reasoning behind that because it just seems a waste

        • ssj4Gogeta

          When they used two different processes, the GPU and CPU were different dies, on the same package. They just did that because they had a lot of 45nm fab capacity which they wanted to use.

      • PINJ

        Eh... They'll Be Releasing An Updated Tegra 3 Chipset With Quad Arm 9s On A 28nm Die.

    • http://www.androidpolice.com/ Cameron Summerson

      A15. :)

      • http://jordanhotmann.com/ Jordan Hotmann

        Excellent, thanks!

  • http://profiles.google.com/mechanizedapathy Shawn Brandel

    Mother of cores.

  • http://twitter.com/GitMuny Ciroc Obama

    Insert generic comment about efficiency over performance, and Tegra being first to market but not as good as its competitors.

  • KRS_Won

    Great, at this rate, my OG Droid will be out of date soon.

  • defred34

    Holy hell! That means the Tegra 4 is over three times better GPU wise than the Tegra 3?!

    I'm all for this if (and only IF) games can began running without lag and what-not, at least comparable to the latest iPhone and iPad in terms of fluidity.

  • ssj4Gogeta

    It also boasts six times the GPU cores as the Tegra 3, with a staggering 72 cores on board.

    That by itself doesn't mean much. It's probably a different microarchitecture.
    Any word on whether they've moved on to a unified shader model?

    • DCMAKER

      if anything the cores would be better but who knows....it is weird that there is 6 times more. 40nm(if i recall) to 28nm isn't a monumentally drop. to magically squeeze in 6 times more.

      • Ryan Wallace

        It has a lot to do with the arrangement of the die, and how stuff it put onto it

        • DCMAKER

          there is only ~42% increase in actual space from the die shrink (if i did the math right 40/28 right? or is it 28/40? (30%)) so either they are smaller cores by transistor count or they made a vastly larger SOC.

          • ssj4Gogeta

            Actually, density (over area) increase would be (40/28)^2 times = 2 times, in an ideal world. But different kinds of transistors scale differently. Correct me if I'm wrong.

          • DCMAKER

            your right it is squared but still a 2x increase still no matter how they work it can't account for 6 times the cores. Either each core has less transistors or they increased the die size.....or better yet both.

    • Yann Lacour

      Yup, 72 unified shader

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

    So here's some good news. The bottom portion of the I/O section next to USB 3 that's obscured says "eMMC 4.5." That's the internal memory I/O standard, and wouldn't you know:

    "e-MMC v4.5 offers performance enhancement features, including an interface bandwidth increase from 104 Mbyte/sec to 200 MByte/sec." (http://www.jedec.org/news/pressreleases/jedec-announces-publication-e-mmc-standard-update-v45)

    Hopefully this will finally solve Tegra's R/W speed woes, along with the addition of dual-channel memory.

    • DCMAKER

      good god thats probably my first complaint is the horrible slow read and rights in all cell phones. When will they get some real flash going in these things. I haven't used any new smart phones in the last year but i haven't heard anything specially about them with load times and stuff. I remember a friend of mine telling me about android instant on and so on....never experience that. My damn phone takes like 4 mins to turn on.

      • Ryan Wallace

        that's because you need a newer phone... my droid dna takes maybe 45sec from full power off to being usable. thats not slow at all

        • DCMAKER

          my computer is faster still with a freakin hardrive. he was talking about this like 4 years ago with 2.2 and 2.3 and 3.0. Before the xoom and during the xoom. 45s isn't instant on. Even a Windows computer with hardrive is faster. Throw in SSD and it is like 6-10s with a dozen programs loading at boot.

          • ssj4Gogeta

            Of course it's faster. Even a modern computer HDD is faster than the flash on phones. An SSD isn't faster only 0because it uses flash, but because it uses multiple channels in parallel. And I guess they also use faster flash itself and controllers.

          • DCMAKER

            the fact that phones use flash that have worse IOPS than a disk hard drive is horrible and make them slow as molasses. It is hard to believe that they use such slow flash for an actual OS experience. Disk drives have any where from 80-400 IOPs and most are only 100+. I don't know how you think its common sense that they use horrible flash

        • fixxmyhead

          my nexus 4 took about 30 secs. from full off to start up. its fast . hardly any of that start up lag when moving homescreen to homescreen. after like 1 second its all good

    • ssj4Gogeta

      Are you sure the interface is the bottleneck in this case? I don't think they even saturate the 100MB/s capacity of the older interface, on any phone.

      • DCMAKER

        not sure how the tegra has read or write woes due to bandwidth limitations. They can't possible using the whole bandwidth. Modern day mSATA cards are only in the 200-300MBps range and cellphone flash doesn't even come close to that nor do i even believe they reach 100MBps.....if they did you would have any bottlenecks. I can't think of anything on a phone that would use 100MBps transfer speeds for flash except for a pure file transfer. Hell most games are just barely breaking 1GB which would take 10s tops to trasnfer

        • ssj4Gogeta

          And hence, this will do nothing to change the state of NAND IO on Tegra. Which is what my point was.

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

            The controller has been the issue in the past, a new revision signals a new controller. I'm not saying it will go 200MB/s - merely that that is what it's quoted at, so hopefully the real-world speeds will be quicker as well. Much like a car's horsepower rating, a peak transfer rate is indicative of little in and of itself if other bottlenecks are present - as you've pointed out - but people have been pointing the finger at NVIDIA's choice of NAND controller since the TF Prime came out.

  • http://www.bradley.zxq.net/gallery Brad Biederman

    Is anyone else worried about the "secure boot" bit? Is that supposed to be like secure boot on Windows 8, preventing the loading of an unauthorized/unsigned OS, or in this case, custom roms?

    • DCMAKER

      good question not sure what that does

    • http://www.facebook.com/leo.davidson Leo Davidson

      NVidia will be aiming Tegra 4 at Windows RT devices as well as Android ones (same as they've done with Tegra 3, I believe), and Windows RT requires Secure Boot, so they'd need to have it in there.

      Presumably it will be up to the OEMs whether they enable and/or mandate it in any particular device.

  • DrMacinyasha

    I can't wait until Tim Cook tries to explain why the "New iPad 4S 6G SLTEA Gold Super Platinum Steve Jobs Honorary Edition with Electron Microscope Display and Unicorn Piss Tri-Core CPU" is better.

    • http://codytoombs.wordpress.com/ Cody Toombs

      Respectfully, it's the Android products that get those horrible names. Apple's naming conventions don't always make complete sense, but they are always simple enough that even the octo-mom can keep up with them.

  • http://www.facebook.com/profile.php?id=1745689461 Hal Motley

    But will it snap the mighty Qualcomm Snapdragon?

  • mma173

    Will the OS get every possible benefit out of it?! I doubt it