Comments Locked

90 Comments

Back to Article

  • Wreckage - Friday, May 6, 2016 - link

    I feel a great disturbance in the geForce, as if millions of voices suddenly cried out in joy, and quickly wanted to upgrade.
  • Horza - Friday, May 6, 2016 - link

    /cringe

    Lets hope the interesting info isn't completely drowned in marketing.
  • osxandwindows - Friday, May 6, 2016 - link

    For the gpu is small and full of performance.
  • davegraham - Friday, May 6, 2016 - link

    interesting...no HBM2 memory.
  • osxandwindows - Friday, May 6, 2016 - link

    HBM2 is coming in 2017
  • dragonsqrrl - Friday, May 6, 2016 - link

    This is gp104, nothing surprising
  • The_Admiral - Friday, May 6, 2016 - link

    Shut Up And Take my Money!!!!111
  • Stochastic - Friday, May 6, 2016 - link

    The critical question: will the 1080/1070 move the price/performance barrier forward? If so, by how much? With so much pent-up demand for FinFET GPUs, I fear that there will be quite a bit of price gouging in the first few months of sales. I hope I'm wrong.
  • RaistlinZ - Friday, May 6, 2016 - link

    1070 > Titan X at an MSRP of $379.99. Pretty impressive.
  • jjj - Friday, May 6, 2016 - link

    If they made that claim, it only stands in VR.
  • dragonsqrrl - Friday, May 6, 2016 - link

    Not sure about that. Based on the theoretical FP figures in the slides the 1070 is quite a bit slower than the 1080. There's more of a gap than usual, and it seems to suggest the 1070 performs somewhere between the 980 and 980Ti. Though at $380 that's still a good deal.
  • Meteor2 - Saturday, May 7, 2016 - link

    1070 has 72% of the FP performance of the 1080, so faster than Titan X -- and much cheaper.
  • dragonsqrrl - Saturday, May 7, 2016 - link

    We'll have to wait for reviews to know for certain, but one of the slides suggest the 1080 is about 25% faster than the TitanX, or in other words the TitanX should have about 80% the FP performance of the 1080. But like you said they also mention that the 1070, which has 72% the FP performance of the 1080, outperforms the TitanX...

    Maybe they were referring to different workloads, idk. But there seem to be a number of performance related inconsistencies in the slides that I haven't been able to figure out.
  • Yojimbo - Friday, May 6, 2016 - link

    I'm guessing he meant the 1070 is faster than the Titan X when using the single pass stereo. He kept repeating "2x the performance and 3x the efficiency of the Titan X" for the 1080 just before he made the claim about the 1070.
  • Jumangi - Saturday, May 7, 2016 - link

    That's a crappy comparison as the TitanX is a very poor value for its extreme cost. These cards need to be comared to the 980/970 in price and performance.
  • Cygni - Friday, May 6, 2016 - link

    Multi-billion dollar development budget and the best Nvidia's marketing department could do is "1080", a name that will probably confuse customers, and a slide with a functionally unlabeled Y-Axis? Cmon man.
  • Stochastic - Friday, May 6, 2016 - link

    We live in a world with the Xbox "One" and Battlefield "1". Nothing surprises me anymore. At least we have the PS4K to look forward to.
  • zaza - Friday, May 6, 2016 - link

    is Tom gonna get fired ?
  • willis936 - Friday, May 6, 2016 - link

    I too want to know this.
  • vladx - Friday, May 6, 2016 - link

    Tom is Jen-Hsun Huang's best guy, can't fire your best software guy.
  • mwallace - Friday, May 6, 2016 - link

    Sorry, but imho Tom deserves a public apology. He apparently forgot to cover his mike as he spoke to a colleague to queue the next presentation point, which led to the CEO's full-boss intimidation take down of the poor guy. The CEO's behavior was cringe-worthy, unprofessional and completely unnecessary. What an ass. I immediately lost interest (and part of my dinner) and killed the livestream. Nice job dummy.
  • Yojimbo - Friday, May 6, 2016 - link

    I agree that Huang overreacted, and unnecessarily publicly raked him over the coals a bit, but you certainly had a strong reaction to it.
  • Meteor2 - Saturday, May 7, 2016 - link

    That whole section was pretty funny. In the UK we call it 'banter' and it's a sign of mutual respect.
  • willis936 - Saturday, May 7, 2016 - link

    When a CEO worth a billion dollars rides your front side during a live streamed product launch of a line that cost enough to "go to mars" it stops being banter and starts being "Please don't fire me."
    I don't how chewing out a subordinate seems like banter after a long delayed product launch from a company that doesn't have confident stock value. Tensions were high.
  • jjj - Friday, May 6, 2016 - link

    OC versions at higher $$$ might means the vanilla ones are locked?
  • xdesire - Friday, May 6, 2016 - link

    Hmmm let's see, 600 bucks for 1080 and 380 for 1070. Looking at the numbers, 1070 seems a bit weak compared to its big brother. Now it's AMD's turn to show us Polaris
  • xdesire - Friday, May 6, 2016 - link

    1070 seems a bit TOO weak i mean
  • Strom- - Saturday, May 7, 2016 - link

    1070 too weak? It's faster than the Titan X.
  • lief1250 - Friday, May 6, 2016 - link

    1070 seems too slow for a cut down GP104.
    Usually the X80s are only around 20% faster than the X70s, this time they're 50% faster.
    1070ti maybe?
  • revanchrist - Friday, May 6, 2016 - link

    Maybe the performance numbers from Nvidia is based on 4K gaming. The much higher bandwidth of GDDR5X helps.
  • lief1250 - Friday, May 6, 2016 - link

    GFLOPS aren't affected by resolution afaik
  • antifocus - Friday, May 6, 2016 - link

    To be honest I dont know where do you get those numbers. If you are simply doing the comparison for GFLOPS, 980 is 32% faster than 970 according to wikipedia. 1080 is 38% faster than 1070.
  • lief1250 - Friday, May 6, 2016 - link

    I was using kepler as a reference point since it was the last die shrink, it had similar price constraints working with a new process.
    And yeah i guess my head calculation had a bit too much rounding off, you're right.
  • Jrood89 - Friday, May 6, 2016 - link

    So the question is, 2 1070 SLI or 1 1080
  • vladx - Friday, May 6, 2016 - link

    Go for single unless you want to play VR games since 90FPS is not enough for that.
  • Jrood89 - Friday, May 6, 2016 - link

    Just 4k... My 780 is not enough and have been waiting for Pascal.
  • vladx - Friday, May 6, 2016 - link

    That gonna be a huge upgrade, GTX 1080 should be about 5x faster. Enjoy xD
  • exileUT - Saturday, May 7, 2016 - link

    Using Bench on this website, I compared a GTX780 to a GTX980 Ti + 25% (GTX1080 est.) The result is 2.17x faster. (ex. 30fps on GTX780 = 65fps on GTX1080)

    The flagship nvidia cards have been 50% faster each gen on average between the last 4 generations. GTX1080 Ti should be 50% faster than a GTX980 Ti. It doesn't matter what is possible on the new node, they have targets during design that don't change without competition.
  • vladx - Saturday, May 7, 2016 - link

    Jen Hsung said 1080 is 2x performance Titan X with 3x efficiency.
  • exileUT - Saturday, May 7, 2016 - link

    And before that he showed the slide where it was 20% faster than Titan X. The TDPs are 250w and 180w. It might deliver 2x performance in specific VR scenarios.
  • vladx - Saturday, May 7, 2016 - link

    Still delivers like I said, VR is the future anyways. AMD will go below 15% until they release Vega to compete since Polaris 10 can't.
  • exileUT - Saturday, May 7, 2016 - link

    I just checked nvidias website. They show Tomb Raider at 1.75x faster than a GTX980. Witcher 3 1.66x faster. VR is 2.6x faster using a synthetic benchmark. It's fun to speculate, but none of us will know until the benchmarks are out. You are probably right about AMD, sadly.
  • jasonelmore - Friday, May 6, 2016 - link

    How long is the 1080? The picture on the keynote shows a 14" long GPU
  • dragonsqrrl - Saturday, May 7, 2016 - link

    The stock PCB is 10.5", standard
  • tr1age - Friday, May 6, 2016 - link

    The lack of cuda cores compared to the 980ti will it hit render performance?
  • T1beriu - Sunday, May 8, 2016 - link

    Probably not, but the lower number of cores will be compensated by the higher frequency.
  • SirGCal - Friday, May 6, 2016 - link

    And now the wait for Ti and Titan's continues... Such a tease...
  • dragonsqrrl - Friday, May 6, 2016 - link

    You might be waiting for a while...
  • theMillen - Friday, May 6, 2016 - link

    my guess would be late q1 2017
  • jasonelmore - Saturday, May 7, 2016 - link

    GP100 is already in production.. Imperfect dies are already being saved back for the 1080Ti, i doubt we'll have to wait very long. 6 months at the most.
  • exileUT - Saturday, May 7, 2016 - link

    The last two Ti models released on average 7 months after the x80 model.
  • Dug - Friday, May 6, 2016 - link

    Why wait when it's as fast as 2 980's in SLI and is priced at $600?
    Once the Ti comes out, are you going to wait for the 1180?
  • RussianSensation - Saturday, May 7, 2016 - link

    It's just logic. If 1080 is 25% faster than a 980Ti and that took 1 year, GP100 with 3584 CCs would be up to 40% faster once it's also overclocked and it'll take about a year to get here. If NV released a 3840 CC full fat chip 780Ti successor for $699, then 1080 for $599 looks really bad for anyone that has a decent GPU that's powerful enough to grind out over the next 12-15 months.

    No doubt, 2560 CC 1080 with 1.733Ghz Boost and 2.11Ghz overclocking is impressive but that makes $700-750 GP100/102 EVEN more impressive in light of that!
  • Yojimbo - Friday, May 6, 2016 - link

    Jen-Hsun Huang should think of more adjectives besides 'amazing'.

    The simultaneous multi-projection seems pretty awesome for VR. If it has a similar output quality it seems to undercut AMD's focus on multi-GPU rendering for VR. I imagine they can still output to the various viewports at various resolutions (multi-res rendering). But it seems to me that the SMP method of correcting for lens distortion might allow for less accurate correction than the standard method (potentially) does. With SMP there are just 4 different projections per eye, each projection itself monolithic. But with the standard method it seems one should be able to correct as accurately as one wishes/has the information for. I have no idea what is done in practice, though. That's if I'm understanding the two methods correctly.
  • Yojimbo - Friday, May 6, 2016 - link

    Oh the other thing I noticed is that, despite saying there can be up to 16 view ports at a time, Huang only proposed to use 8 of them for VR. What's the application for 16 then? A Light field HMD?
  • D. Lister - Saturday, May 7, 2016 - link

    VR probably incorporates two view ports simultaneously, so 2x8=16.
  • Yojimbo - Saturday, May 7, 2016 - link

    In his demonstration he showed 2x4 = 8.
  • Mr Perfect - Friday, May 6, 2016 - link

    Jen-Hsun will probably stop using"Amazing" when he no longer considers leather jackets business wear.
  • D. Lister - Saturday, May 7, 2016 - link

    Seeing him on stage, I alway get this feeling that the next time he would ditch the whole biker thing and just go full awesome with green spandex, and a cape with nvidia logo on it. :|
  • xaueious - Friday, May 6, 2016 - link

    Nvidia's home page is showing 1.7GHz as base clock, but in their presentation their sample was clocked at 2.1GHz on the stock cooler. If this is any potential indication, even though the 1070 is severely cut down, it might be able to make up for a lot of headroom with overclockability.
  • revanchrist - Friday, May 6, 2016 - link

    Maybe they'll handpick the best overclockers to sell as founder's edition?
  • vladx - Friday, May 6, 2016 - link

    Yep, makes sense.
  • D. Lister - Friday, May 6, 2016 - link

    "Not mentioned: that all of the consoles are also based on AMD's GCN architecture"

    Good then, that we have you to gleefully point that out. :p

    "...they're going to want to focus on getting GTX 700 series users to upgrade"

    So then, I'm guessing, the whole "power of 10“ twaddle was about performance/watt compared to what, Kepler?

    Anyway, let's hope review samples are sent sooner rather than later. The benchies could be an interesting read. I myself am really looking forward to the 1080‘s Ti variant to replace my 780Ti.
  • Yojimbo - Friday, May 6, 2016 - link

    I think 'power of 10' meant the '10' in 1080.
  • D. Lister - Saturday, May 7, 2016 - link

    Oh, I thought there was some math in there. :(
  • Mr Perfect - Friday, May 6, 2016 - link

    Any chance you guys will do up an actual article(pipeline size is fine) on all of this? I tend to give live blogs a complete miss because they read like a Twitter feed, but there's clearly good data in here. A nice write up would be appreciated.
  • exileUT - Friday, May 6, 2016 - link

    GTX1080 is 25% faster than GTX980 Ti. $599 is 13% more than $530. I guess that's a good deal? 250 watts vs 180 watts. I hope AMD competes this generation. I still remember the 9800 Pro from 2003. You can be great again guys...
  • AnnonymousCoward - Saturday, May 7, 2016 - link

    I wonder if NVIDIA is intentionally restraining GTX10 performance, so that they can keep coming out with faster cards for the next several years. 16nm should give 3x the density of 28nm, and there's a new architecture here too. Why only +25%?!
  • jasonelmore - Saturday, May 7, 2016 - link

    because they must milk 16nm for a very very long time. HBM got ditched, the only thing holding these pascal cores back is memory bandwidth.
  • dragonsqrrl - Saturday, May 7, 2016 - link

    How was HBM ditched? HBM2 is going to be paired with AMD's upcoming 'big die' GPU (Vega 10), and is already shipping in the Tesla P100 (GP100).

    Based on the specs and estimated performance figures for GP104, I doubt it'll be bottlenecked much if at all by its available bandwidth with GDDR5X.
  • exileUT - Saturday, May 7, 2016 - link

    They get an R&D budget and price, performance, efficiency targets. If the product sells well, they hit their targets. What's possible doesn't matter. It takes a world war or space race to see what that is.
  • dragonsqrrl - Saturday, May 7, 2016 - link

    TSMC's 16nm node offers roughly 2x the density of their 28nm, not 3x.

    "Why only +25%?!"
    Because you're comparing GM200 to GP104, not GP100. What's the basis of your expectation that the performance increase be significantly greater than that?
  • AnnonymousCoward - Saturday, May 7, 2016 - link

    The 2x process shrink, the supposed 17B transistors (double), and the new architecture. Shouldn't doubling the transistors give at least the performance of 2 of the old dies?
  • dragonsqrrl - Saturday, May 7, 2016 - link

    Where are you getting 17B transistors from? I haven't seen any official confirmation of transistor count yet, but based on it's die size early estimates placed GP104 at roughly 8B transistors, similar to GM200.
  • paddytokey - Saturday, May 7, 2016 - link

    I'm pretty sure that's exactly what is happening. Like with the 680, the 780 and the 980 ;)
  • dragonsqrrl - Saturday, May 7, 2016 - link

    The MSRP for the 980Ti is still $650. It's a pretty good deal considering the performance increase,the new features, the lower TDP and significantly improved perf/w, and the fact that prices almost always drop below MSRP in the months following launch.
  • exileUT - Saturday, May 7, 2016 - link

    I guess, newegg has one for $530 after MIR. I'm waiting for the 1080 Ti since it lines up better with Intel Kaby Lake Q1 2017. It should be around 3.5x performance increase over my GTX680/2500k. I upgrade every 5 years or so when the stars align.
  • RussianSensation - Saturday, May 7, 2016 - link

    Kaby Lake is launching in Q3-4 2016. If you are getting 1080Ti/Vega HBM2, your best bet is Skylake-E $389 -- successor to i7-6800K Broadwell-E.
  • lilkwarrior - Friday, May 6, 2016 - link

    Being a 980TI SLI owner, it seems, as usual, to wait for 1080TI
  • Ashinjuka - Saturday, May 7, 2016 - link

    #PowerOfTom
  • jasonelmore - Saturday, May 7, 2016 - link

    So what about Async Computer?
  • tyger11 - Saturday, May 7, 2016 - link

    I guess I'm missing it in the posts, but what's a "Founder's Edition" - just an overclocked version of the regular one?
  • nyoungman - Saturday, May 7, 2016 - link

    Watching the video of the event, it isn't entirely clear to me what the Founder's Edition is. It sounds like NVIDIAs reference design and cooler as opposed to cards from partners. Maybe they select the ones that are more overclockable for Founder's Editions -- not sure. I also read somewhere that they could (only) be ordered online.
  • nyoungman - Saturday, May 7, 2016 - link

    There is a recording of the event over on Twitch. It starts at about 27 minutes in:
    https://www.twitch.tv/nvidia/v/64989878

    NVIDIA also posted their video introducing the 1080 which includes the specs:
    https://www.youtube.com/watch?v=JUL6Gc2g4kg1
  • yhselp - Saturday, May 7, 2016 - link

    Nvidia are getting close to charging triple what they used to traditinally. The GTX 1080 packs a medium-sized GPU and costs up to $599; whereas a few short years ago the 560 Ti - the premium mid-sized GPU - cost $249. How long are Nvidia going to delay big Pascal, and how much is it going to cost?! Jesus...

    Where are you AMD. We need competition badly.
  • yhselp - Saturday, May 7, 2016 - link

    edit: $699*
  • D. Lister - Saturday, May 7, 2016 - link

    The 1080 is in the same tier as the 580, which had 3 billion transistors and came with 1.5GB VRAM and was released in 2010 for $499.
  • Meteor2 - Saturday, May 7, 2016 - link

    I found the statement 'can play all available games at maximum detail at more than 60 FPS' pretty compelling.
  • Mattly - Monday, May 9, 2016 - link

    I don't know if it's an American thing, but I found him repeating himself constantly really irritating. It's like he's talking to a small child or someone really slow.
  • Hrel - Monday, May 9, 2016 - link

    I wonder if the GTX 1070 is actually a 5.5GF card with 6.5GB of RAM and marketing just didn't get the memo again... or something.

Log in

Don't have an account? Sign up now