Comments Locked

36 Comments

Back to Article

  • NextGen_Gamer - Monday, January 6, 2014 - link

    During press conference Jen said it is 128 KB L1-I cache & 64KB L1-D cache. And according to Nvidia's press release on their website, Tegra K1 with 4+1 Cortex A15 cores is available in devices in 1H 2014, while Tegra K1 with dual-core Denver is available in 2H 2014 (both sound optimistic to me).
  • skiboysteve - Monday, January 6, 2014 - link

    In my experience add a full half year to all chip maker "shipping in" quotes because the device makers have to put it in a product before we can use it. So 2H 2014 for the A15 ver and 1H 2015 for the Denver ver.

    that's what's nice about apple... None of this year early tease crap. They talk about it when you can buy it. (and btw... I'm no apple fan.. But I appreciate this aspect)
  • jjj - Monday, January 6, 2014 - link

    Actually no , they showed the A15 silicon in july so chances are we will see devices with it soon (MWC). For Denver it's hard to say since things can go wrong and delays can occur but it's clear that they are planning for products this year.
  • pricedOut - Monday, January 6, 2014 - link

    The A15 one will appear very soon in a Shield and Tegra Note, and they had samples out in July. For Denver, they have silicon back in Santa Clara and if things go well, samples will be available by the end of this month. That's exactly a 6-month separation, so if second half of 2014 is realistic for the first 64-bit Android.
  • pricedOut - Monday, January 6, 2014 - link

    I meant 64-bit ARM-based Android. Intel may be able to get someone to produce a x86-based 64-bit Android device earlier.
  • dylan522p - Monday, January 6, 2014 - link

    They already have. They just haven't written the drivers and such.
  • camelNotation - Monday, January 6, 2014 - link

    Hmm, I wonder if there's an L2 cache. The die shot doesn't show anything that looks like L2.
  • Homeles - Monday, January 6, 2014 - link

    That's because Nvidia doesn't provide die shots of their Tegra products, only censored "artist's representations."
  • extide - Monday, January 6, 2014 - link

    L2 is going to be per-core, and probably on there. It would be in with each core, not an extra block. However, there definitely isnt anything that looks like L3 cache, though; which would be an extra block.
  • Kevin G - Monday, January 6, 2014 - link

    L2 cache per core on the A15 or Denver version?

    From historical precedent, A15 version should have a shared L2 cache.
  • Exophase - Monday, January 6, 2014 - link

    A15 will share L2 cache. nVidia isn't about to make such a massive change to ARM's RTL. I doubt there'd be much of a point anyway.

    On the Denver variant, I only see local L2 happening if there's a shared L3, Nehalem (and forward) style. Everyone else has moved towards shared LLC, except for archaic designs like Wii U's.
  • Exophase - Monday, January 6, 2014 - link

    These L1 caches are gigantic as far as L1 caches go, but that hardly obviates L2 cache entirely. The trend has been to move towards more levels in the cache hierarchy; even in mobile we can see the first signs of L3 caches. I can hardly see nVidia moving in the opposite direction here.
  • Kevin G - Monday, January 6, 2014 - link

    Indeed. I'm curious about the L1 latencies since they're so large. The massive L1 Instruction cache is just massive at 128 KB.
  • jjj - Monday, January 6, 2014 - link

    Nice to see that the core seems big ,hoping for a less mobile chip with it at some point since on the x86 side we aren't getting more cores at sane prices.
  • chizow - Monday, January 6, 2014 - link

    Denver K1 looks to be a monster. All the CPU perf of Apple's A7 with unseen GPU graphics from a full Kepler SMX. Only question is WHEN. Could be a great year for Nvidia in the mobile space.

    Definitely glad to see they are going with 2 beefier cores over 4 or even 8 weaker cores. Also great to see they are pin-compatible, means they should be much more flexible in terms of options on any particular device (entry-level A15, high-end Denver).
  • MikhailT - Monday, January 6, 2014 - link

    At this point, while I'm excited to hear about this, I feel like it doesn't really matter much.

    We've seen this type of progress/innovation from nVidia for the past few years but they're not really winning much design wins. I'm pretty sure I've seen more AMD design wins for their laptop APUs than Tegra 2/3/4 combined.

    With that aside, it would be interesting to see what nVidia can do with K1-based Note 7" project. I'm still waiting for a good alternative to iPad. Asus/Nexus is banned in my household as their products keep failing in months, while my original iPad is still working properly.
  • jameskatt - Monday, January 6, 2014 - link

    By the time an Android Phone is on sale with this chip, we will see Apple's A8 chip on the iPads and iPhones - and probably the iPod Touch also. nVidia is simply playing catch up with Apple.
  • Torrijos - Monday, January 6, 2014 - link

    While I agree with you (about the delay between announcement & release for everybody in the industry except Apple), I still wonder about the implication on the GPU side for Apple...

    A long time ago PowerVR lost the battle for GPU domination in the game console market when they moved towards more polygons, I wonder how the current core of each GPU IP compares?

    AMD seems to offer better OpenCL performances than Nvidia, while Nvidia seems to offer better graphical performances per watts.
    Where does PowerVR Series 6 (rogue) falls on this aspects? Right now it still lags in openGL (3.2) what about the rest (and their implication in mobile)?
  • MikhailT - Monday, January 6, 2014 - link

    ImgTech announced the PowerVR Series 6XT today. It seems to be a refresh of Series 6, quicker and more power efficient. I have a feeling that Apple will be quicker with A8 CPU, but slower with GPU this year, I doubt 6XT will catch up with K1.

    http://www.anandtech.com/show/7629/imagination-tec...
  • winterspan - Monday, January 6, 2014 - link

    There are many reasons for this, most relevant being that Nvidia hasn't had a combined LTE/Applications processor SoC. They are working on one now, but be first one doesn't have high end CPU cores. This requires device makers to have to buy and integrate a separate 4G LTE chip, thus you see Qualcomms dominance.
    Also, Nvidia really hasn't delivered well. Their chips are always less impressive than the marketing. Tegra 2/3 only had a single memory channel and mediocre GPU performance. Tegra 4 is better, but still doesn't have a unified shader GPU hick supports all the modern OpenGL/DirectX standards. I'm excited to see how this K1 chip does though....
  • jasonelmore - Monday, January 6, 2014 - link

    just look at the cores in the picture.

    I know these are not die shots, but it looks like Denver cores are roughly double the size of A15 Cores.
  • Krysto - Monday, January 6, 2014 - link

    They're 7-way vs 3-way superscalar so I'm not surprised. It's interesting that they're moving to dual-cores, though. So far Denver looks more like a competitor to Haswell/Broadwell than any other ARM chip. I'm actually starting to lose hope that it will be very efficient.
  • iwod - Monday, January 6, 2014 - link

    Since Nvidia are now licensing their Ge force Tech out, I wonder what's the implication of that.

    And no Maxwell?
  • Jeffrey Bosboom - Monday, January 6, 2014 - link

    Why would you want a larger instruction cache than data cache? Is this something specific to mobile?
  • Exophase - Monday, January 6, 2014 - link

    Enticing article. Not the news I was expecting today.

    One thing I don't agree with - I don't see why the Denver K1 has to be 28nm even if there's silicon already. TSMC 20nm silicon is easily possible at this point, it entered risk production forever ago and volume production is due to happen very shortly if not already underway. This could also explain why the Denver variant is coming later, although to be fair it's really coming earlier than any of us expected, unless this K1v2 is actually Parker..

    The only thing that makes me question 20nm is nVidia's more or less open criticism of following leading edge TSMC processes. And that on paper 20nm doesn't offer terribly great power consumption reduction.
  • errorr - Monday, January 6, 2014 - link

    Just to update, the 20nm FPGA's are already available as of November. Supposedly "volume" production should start in the next few weeks and ramp up through Q4.
  • Krysto - Monday, January 6, 2014 - link

    I think they want to make both K1 and K1v2 (Denver) on 28nm HPM, and then just jump straight to 16nm FinFET in 2015 (hopefully in first half of 2015). They are probably one of the very first customers of TSMC for 16nm FinFET, because they mentioned it since last year.

    I wish AMD would do the same and jump to FinFET as soon as it's available, and forget about extra costs. It would give them a much needed boost to shrink the gap between Intel in CPUs.
  • jospoortvliet - Monday, January 6, 2014 - link

    I feel the word 'details' is a bit of an exaggeration... ;-)
  • MrSpadge - Monday, January 6, 2014 - link

    At least it's more detailed than anything we officially knew before ;)
  • errorr - Monday, January 6, 2014 - link

    I'd be more worried about the cost of these. Even if they are 'faster" than A7 they are still going to be significantly hotter and suck more battery.

    Until tablet android becomes a viable ecosystem where there are actual apps optimized for those screens I don't see the point of Denver. Android tablets are about media consumption and other use cases are theoretical for now.
  • twizzlebizzle22 - Monday, January 6, 2014 - link

    Do we think Nvidia are going to meet the power and clock requirements in these implementations when it comes to the OEMs and mass produced chips with differing bins?

    Also can anyone explain or link me to some information on what a 3 or 7 way superscalar is?
  • dylan522p - Monday, January 6, 2014 - link

    How wide an architecture is.
  • winterspan - Monday, January 6, 2014 - link

    I'm no expert, but I believe it refers to how many instructions can be issued (i.e. worked on) simultaneously
  • Krysto - Monday, January 6, 2014 - link

    Someone posted this on Beyond3d:

    5 watts related to perf numbers in the table(365 Gflops), GPU should work at ~1Ghz to achieve those flops, it's definitely not for smartphones or even tablets, but rather for cars and chromebooks, but considering last summer demo with 1 watt GPU power consumption it should be scalable down to phones, that's why you can observe POP package here - http://www.nvidia.com/object/tegra-k1-processor.ht...

    Wow. It needs to reach 1 Ghz to get that performance? If that's true, then FU, Nvidia! I HATE it when both Intel and Nvidia do this. They mislead people with their benchmarks and "theoretical max performance" numbers or some crap like that, when in reality you're not going to be able to utilize that much performance because your devices will get TOO HOT within MINUTES!.

    I have an Nvidia GPU in my laptop, and it's useless. It advertises 3x the performance of the Sandy Bridge GPU, but when I play games, after about 30 minutes, it throttles to about half of that performance or less, making it barely any better than the Intel GPU. Then what the hell is the point of all that performance, if I can't USE it?

    I thought Nvidia learned a lesson with Tegra 4 so they don't make a "tablet chip" that inevitably becomes too performance focused, and draws too much power. It seems they are repeating the mistake with Tegra K1, though, and it won't actually be a "smartphone chip" that can actually be used for more than a few minutes under intense load.

    Nvidia has one more shot with Tegra K2 at 16nm FinFET. If they fail to make a proper "smartphone chip" that's efficient even at maximum load, then I'm giving up all hope on Nvidia.

    Damn it and I wanted them to do well, because I really think Qualcomm is about to turn bad and become uncompetitive due to its almost market monopoly at the high-end.
  • twizzlebizzle22 - Tuesday, January 7, 2014 - link

    Yeah these benchmarks are all impressive etc but when you see the likes of what qualcom are doing on the "last gen" implementations. As battery tech is only seeing marginal gains its interesting to see the power improvements which can be made throughout the soc. Tech like envelope tracking etc. Personally I don't game on mobile only in the desktop space so I'm interested to see applications and derivatives of the a57 which draw minimal power, I hope the next nexus focuses on a more power efficient chip in favour of chasing the court and clock count. As in the hangout with Peter Greenhalgh he mentioned not pushing the chips above 1.8 - 2.0ghz. I'm sure the 10% performance increase isn't going to be worth it with the extra voltage that needs to be pumped in, its not linear.

    I'm sticking with my Nexus 5 until a decent power efficient chip is made available whether that's with 16nm FinFET or even on the 20nm space, any mid to high range chip is powerful enough to run android pretty much jank free.
  • phoenix_rizzen - Thursday, January 16, 2014 - link

    Any SoC released since summer 2012 (so Qualcomm Snapdragon S4 Pro or newer) is capable of running vanilla Google Android or AOSP-based Android ROMs "jank free". And, each version of Android since 4.2 has required less resources to provide the same fluidity. Just look at what Motorola was able to do with the Moto X, running dual-core Krait 300 CPUs and older Adreno 320 GPU (aka Snapdragon S4 Pro), for example.

    However, it takes a very recent SoC (like Snapdragon S600 or newer) to run OEM-skinned Android even slightly janky, with TouchWiz and LG's versions some of the worst offenders. LG's version of Android 4.1.2 runs like molasses compared to Rootbox 4.2.2 or Carbon 4.3.1 on my Optimus G (quad-core Snapdragon S4 Pro).

    We don't need faster/more efficient hardware. We need faster/more efficient software that can use all the features of the hardware we already have. The Moto X is an excellent example of this. Motorola's optimised bionic and dalvik libs make the older hardware feel more powerful than it actually is. And many custom ROM devs have ported those to other Qualcomm-based phones with great results. Same with the Adreno drivers; there are heavily-optimised drivers floating around that make even Adreno 320 speed up by 10-20%; again ported to various custom ROMs.

Log in

Don't have an account? Sign up now