This was a server/professional release but the GP100 GPU that's the core of that reveal is likely to be the top tier of GPU coming to consumers too (because building a 600mm^2 GPU is hard and you want to sell as many as possible to make back R&D - also the CUDA core design process means building completely different designs for consumers would split the architecture weirdly and so far we generally see generational advances as a single wave). [More readable (technical) reporting of the facts]

That 600mm^2 is the same die area as the outgoing GTX 980 Ti uses only this a shrunk process (each transistor is much smaller so you get a lot more on every chip and they should hopefully produce much less heat to do the same work - the total power required seems to back up it may be 80% faster for the same power draw and high-end GPUs are generally limited by power use). The RAM is upgraded to the latest and greatest HBM2 technology, and (after all those nVidia Maxwell designs that basically ignored 64-bit floating-point performance to focus purely on FP32 (gaming) computational power) FP64 is back. An interesting aside to that is the new chip not only has a lot of FP64 units but the FP32 units can be used for 16-bit FP maths doing two computations per tick - this is interesting as some calculations in graphics can be done at 16-bit without it all going to Hell and this gives game engines a strong incentive to try it out (because you literally can double the calculations you do per second "for free").

And this isn't even the fastest version of that GP100 design (56 of the 60 blocks are enabled - this is done so a small defect in production doesn't force them to throw away the chip as they can disable the block with the defect and so increase yields to make it commercially viable; as time goes on the yields get higher and they can start shipping fully-enabled versions of the design). Here's texture map units, raster units, bandwidth, and floating point (maths that happens when rendering a scene) perf (approximately):

* Mainstream consumer Kepler models had disabled FP64 performance. ** Later, Maxwell removed most FP64 units from the design to get the most FP32 units into the die area (because the 20nm die-shrink failed to arrive so GPUs missed a jump which is why we're going from 28nm designs to a 16nm design).

AMD will respond with the Polaris design (also expected to come in broadly around this scale so it'll be interesting to see how they run for gaming, how they price them, how they build smaller mass-market designs for the cheaper cards, and if there's anything cool being added with these designs to take advantage of that very high memory bandwidth).

So if you were thinking of buying a fancy GPU in the next months, this is what you're going to be looking at getting hold of when the 16nm new designs arrive for consumers. Even if you're looking for a not-fancy new PC GPU, these big numbers may indicate the mainstream models will also be getting radically faster (although that HBM2 RAM is expensive so maybe they won't also see masses more memory bandwidth to go with the faster processors).

That memory bandwidth jump is pretty impressive. Combined with the improved multicore performance of DX12 we might actually see the elimination of stream stuttering with games that support DX12, assuming you have a powerful CPU as well. I feel like with DX11, DDR3 system RAM and current gen CPUs the Titan X is really being bottlenecked by a variety of other things. The performance increase in your average game over a standard 980 should be pretty massive but is often barely noticeable. I really hope Intel can pull some magic out of a hat for this year's CPU refresh, or at least bring 6-core models to consumer level processors. With DX12 those extra cores could make a real difference when paired with a beast of a GPU like this. I worry that no CPU on the planet right now would be able to stand up to this GPU.

Those are some impressive numbers but I wonder how long until games take advantage. 980s barely have any benchmark games within the mainstream.

@hayt: My gut says that VR is going to really push what cards can do in short order.

@joystick_hero: So I added the GTX970 to the chart for comparison. There are differences between generations (this isn't pure apples to apples as they redesign the blocks each time but it's not a totally unreasonable comparison to look at these numbers to get a rough idea of where each area could become bottlenecked with rendering [currently FP32 limited a lot, aka shader-limited] tasks).

I agree on the wider point. Either VR or demand for 4K content will push the requirements for high end GPUs quickly in the coming few years. As the crew is finding, dropping frames in VR is really not good (especially for some, Drew seems most sensitive).

Also we're seeing games that only just hit 1080p (with barely anti-aliasing) at the moment. This may lead PCs to start targeting a really high-detail (so derived from 4K rendering even without 4K output - more samples per pixel = more distant detail from polygons so reason to change the level of detail settings to render well out into the distance) using nothing much more than tweaked current engines with current assets.

Morrowind shipping engine:

Morrowind without the fog and some tweaked assets:

Seems cool although i dont understand half the numbers lol. I will probably need an upgrade for 4k sometime in the next year or 2 so ill be checking how good these end up and if the price is right.

So I'm building my first PC and was waiting for this stuff to start coming out before I start deciding on potential builds. Any news or rumors on prices and release dates? If I were to say build a computer around the GM200 (the 980Ti replacement?) what cost am I looking at in total?

*hugs his fx8350 and r9 290 before putting the barrel to its head and pulling the trigger* :'(

@nightriff: Definitely not yet. Nvidia will keep release dates and prices a secret as long as possible.

The consumer cards probably won't be out until next year at the earliest and you're probably looking at like $700-800.

Those are some big numbers. Or bigger at least. I wonder how soon we can see something that really juices this tech.

Ahhh forget VR, THIS is what has me drooling in the gaming scene. I can't wait for these things to come out and really see just what they're capable of!

I want it now! I want cake now!

The current rumor is end of may for release announcement during Computex.

If I made a proper salary, a lot of my money would likely go in upgrading my PC every year. Right now though, can't afford that. Got a 980ti not too long after it came out and it hurt pretty bad (I'm very happy with it, to be clear).

In fact, my next purchase would likely need to be in the CPU department, with that comes new motherboard and new RAM. *That* is a little scary. Anyway, so far my PC is doing quite fine. I'm ok with my 1920x1200 monitor and am not planning any VR either.

might be time to retire my GTX 480, we lasted awhile without any upgrades.

@mosdl said:

I think that's only for the GP104-based x70 & x80 cards. I wouldn't expect the GP100-based x80 Ti and x80 Titan cards to land until 2017.

I'm far from convinced by the business logic of that rumour. AMD have a Dual Fury card. That is a lot of card. And when have nVidia been happy to just ignore competition walking away with the crown of having the unquestionably fastest card out there? Half of their market dominance is built on "halo" products (which they charge for accordingly with prosumer models with FP64 perf [Titan Black vs GTX 780 Ti in the chart above being a great example] or just the most bleeding edge of single-GPU designs that stand above AMD). Polaris is coming. Polaris has to be a step above the Fury X before the end of the year (and rumours are most of that line will be out this Summer) even if their next single GPU doesn't beat the combined perf of a dual-Fury.

As far as business realities go, nVidia need to stop selling card for more than $400 or release a GP100-derived consumer model before the Winter. I don't think they have any interest in doing the former and seeing their revenues decline. Maybe we get the mid-tier core released first (with the next few months being the rumour) as a gaming slice of how they make this design on a smaller budget (replacing the 970/980) but they can't go 6+ months without the 980Ti being replaced with a GP100 design. If they do, there will be absolutely no reason to buy a GeForce card, it'll be Radeon all the way. Because just shrinking the Fury X to 16nm is enough to give them a great mid-range card and build a larger die monster on top of it (although Polaris is being talked about as more of a step forward than just doing a die shrink).

I mean, if you read SemiAccurate then you'll believe that nVidia are never shipping another product ever (another spectacular swing and a miss by those guys speculating on the missing GP100 last week :D) and have never actually shipped a functional product in all their years, but if we're going by the rumour-mill that's not fed on pure Green hate then things are looking better. They confirmed they are volume-producing these GP100 chips right now (it's just that buying $129,000 DGX-1 servers for research doesn't happen with overnight shipping and we're not going to be seeing P100 board under OEM server vendor distribution until early 2017 - this is standard lag for that part of the industry). I see them waiting for yields to go up but also coming up with a lot of dies that don't hit the 56/60 shader block required for going in their P100 servers. I'm expecting 2 units disabled consumer models to be where they monetise chips with defects covering two shader blocks.

I want to buy Polaris or Pascal, but with AMD releasing Vega at the end of the year, I'm curious about the benefits of waiting those extra 6 months. That would give VR a little more time to mature before I jump in, but then I'd be buying a headset with only 70% of its lifecycle remaining.

16nm will be here for a while. This is "the jump". Yes, we have incredibly stories about efficiency gains (Kepler to Maxwell, ushering in a new era of lean nVidia chips that ran cooler and generally were smaller than the equivalent AMD chip, a complete swap of the traditional comparison) inside a die process but those are the exceptions rather than the rule. The second generation on 16nm should be cool as everyone gets used to making large chips on 16nm (these are the first chips built to be anything like 600mm^2 on this process) but... we've got Navi a year later. It's not like we're not used to annual refreshes (it's just that this gen is a bit backed up from waiting for 16nm to actually work with large GPUs) so there is always reason to wait.

Hell, last time I saw their public charts then Vega was advertised as being the HBM2 launch. Well that's something that might not be waiting until after Polaris. It certainly seems like nVidia are happy with HBM2 yields to start pushing that with the GP100 now. By the end of the year, GDDR5X should actually be viable so there will be a tier between HBM and current stuff (as noted in my chart at the top, GDDR5 is somewhat stalled as far as nVidia wants to push it - AMD only got round that with a super-wide bus [throwing more silicon at memory controllers, more power to the memory subsystem, and just more physical memory chips at the problem]) but for this Summer's chips, I expect every high end card to mirror the Fury line and go HBM (nVidia are committing to HBM2, I remain unsure if AMD will ship their HBM1 controller on 16nm or also transition - again, the chart makes it look like HBM2 doesn't comes to AMD until early 2017 or end of the year at the earliest).

Ultimately it all depends on prices. We could see this new 16nm generation arrive and prices jump up (so they match outgoing prices per perf) which will mean buying a new card doesn't buy much more than the option for a new premium high end (like the GP100's Titan) and lower power use and so fan noise. But we should, going by other transitions, see the new cards coming in with each tier offering a reasonable boost in performance and lowering of power use for the price of the outgoing models. That is what we should all expect to happen and hope that the competition between nVidia and AMD (the Pascal design looks good on paper, we hopefully don't have to wait too long before we have numbers for how well game engines can exercise that raw power; Polaris rumours definitely make it sound like AMD are going to continue to be competitive in perf and get over their thermal issues that hampered their 28nm fight against Maxwell 2 GPUs) keeps them honest and fighting for dominance throughout the performance curve.

I want to upgrade my GTX970 when the next consumer model hits, but I'm not sure if there's a point when I'm still on a 3.2Ghz i72600K. These specs seem promising though.

@hayt: VR and 4K might do it, right now you can't really get 4K/60 on a single 980.

How much it matters will really depend on the price...or how powerful the version of it as a specific price is. $300 to $500 USD is probably the magic number, because cards in that range are the ones that get install bases, which is why the 970 is the most relevant card currently, despite other cards being twice as powerful.

Really looking forward to this next generation of GPUs but I always feel burned when they release the "ti" edition with more RAM and more cores unlocked so I think I'm just gonna wait for the inevitable 1080ti with 32gb of VRAM. I'm still waiting to see how good "Kaby Lake" is as Skylake so far has been a disappointment features-wise.

Buying a Vive and a 4k monitor around late August, hope these will be out by then. Looking forward to it all.

I have to upgrade from my GTX 770 this summer (no vacation for me!), but since I will also have to get a new motherboard, CPU and RAM I can't spend crazy amounts of money. Would getting a GTX 980 be stupid at this point? I am looking at a price of 5590 NOK (672 USD) for that one, while the 980Ti jumps to 7490 NOK (898 USD).

@jaqen_hghar: Wondering what's lacking about your system that means it can't handle a new GPU (as the Bombcast chat points to, CPUs like the i5-2500K isn't the best CPU you can get but if you're having issues with stuff like that in games, you need to check your overclock is stable [overclock gone wrong = CPU throttles down to stay cool and runs at something like half speed] and reinstall Windows rather than assume it's just an old CPU killing your system). It's not the perfect option to put off an upgrade but... an overclocked i5-2500K with some fast DDR3 (so not some 1333MHz "office" spec stuff - make sure you switched XMP on in your BIOS as otherwise the RAM runs slow, even if you brought nice stuff) is basically there for running a single overclocked Titan X as fast as a new i5. Because the old Intels overclocked well, they've got some headroom (and get a CM EVO 212 to cool it and you'll even be able to do that quietly despite slightly pushing the power demands towards 100 Watts) so if you've got one of those systems then it isn't actually an issue that there are much newer CPU/mobo/DDR4 options out there as you don't need to upgrade today.

Right now, I'd not buy a new GPU of any sort until we at least get some prices in a couple of months to make that a more informed decision (May to July we expect details, launch dates, etc for at least some of this new generation from both nVidia and AMD), baring emergencies (like your old one broke). Everything out right now designed for gaming (definitely up there in the price ranges you're looking at) will be getting replaced with a new model this year and hopefully all this Summer. Worst case, the new card will be much cooler and quieter and have a few extra bells and whistles to be unlocked by devs over time (so these Pascal chips having up to double perf when doing FP16 maths - you'll not see the results of that today but in a year, maybe some games start running far more easily if you got a new gen GPU). Likely case is the new card at the same price will be a significant step up in performance (or you can buy a GPu at a lower price tier for the same perf).

Wow.

So, here's the thing, you're already behind. If you buy a 980, you will be behind again in a few months after your purchase. I would wait. A 770 should still be a pretty capable card, assuming you are gaming at 1080p. I'm sure you'll have to turn a few settings down, but it'll definitely be worth your while to wait for the new cards to come out.

Before you go nuts buying an all new PC, try just upgrading to that GPU first. You'd be amazed at how long CPUs are good for these days. My old second PC runs an i5-750. I tried an experiment with that PC, replacing it's super old Radeon HD 6970 with my current GTX 970. Turns out that old i5-750 can still run Fallout 4, the Division, etc. maxed out at 1080p60, despite the fact that the CPU is 7 years old, and is below the minimum requirements on half the games I tested. So, before you go nuts, try just upgrading the GPU and see what happens.

@shivoa:Hmm, I could try to overclock my i5-2500k, as I haven't really done that. I just haven't dared to go into the overclocking game as I also don't want to go overboard on cooling. My CPU does reach 100% load on new games, but still. I also got 16GB DDR3 in there, so going to check out some overclocking first I guess.

And yeah, looks like waiting would be a better thing to do in any case. Thanks!

@drgreatjob: I hear you. I do only play at 1080p, so I'll try the overclocking CPU first, and then see what happens with the new cards. I'm one of those who don't really follow the "PC parts scene" outside of the time when I feel I need an upgrade, so I am always very much out of date when checking this stuff out.

Right now I'm running SLI and currently the Rift and Vive do not support SLI/Crossfire. I plan to upgrade my video card(s) with the next rollout of Nvidia and AMD GPUs. Should I plan to run only 1 GPU later on when VR matures with better games?

@jaqen_hghar: You should look at getting something other than the Intel cooler it came with. Something of a tower shape to push air towards the rear exhaust of your case.

The Cooler Master Hyper 212 EVO is a boring $30 buy but also really just great at dissipating about 100 Watts of heat with next to no real noise (compared to your GPU, it'll be almost silent). Your case does need to be big enough to fit it (it's tall) but that's the only major limitation of the cheap design. It's not the easiest thing to install (and unless you've got a void in the motherboard tray in your case, you'll need to take that out to install it) but it's also something you only need to do once and comes with instructions. With that installed your CPU will run cooler, quieter, and be free to overclock. Also, if you've had the same Intel cylinder cooler on there since 2012, thing will be caked in dust that'll not be helping it stay cool and quiet. Time for a $30 upgrade.

Even without increasing the voltage slightly, an i5-2500K will happily run around 4.4GHz (every CPu is slightly different, some overclock more than others but in general 2500Ks are good to great depending on the random chance of the production line as your one was made). That's a lot better than the 3.3GHz base (or even the 3.7GHz max turbo which it generally won't reach in games that push it hard because it doesn't like to clock every core up to that turbo at the same time) and you should find plenty of detailed instruction online in how to do it (find something written with your motherboard in mind if you're not familiar with the BIOS). The reason you paid for an Intel -K CPU is so it can be overclocked with nothing more than the click of a button.

While you're in the BIOS, check your RAM settings. XMP means your RAM will be running at the right (gaming) speeds to not hold back your CPU. But BIOSes don't turn it on by default (at least in my experience) so you may be running your RAM way slower than it can run at.

That should make sure you're system is more than capable of dealing with even the latest and greatest GPUs (at least single GPUs, SLI/several of them is when slightly more expensive CPUs or the newer ones start to make a difference). The new APIs (DirectX 12, Vulkan) are all about making the CPU even less involved in rendering and offloading even more work to the GPU without bogging down a mid-range CPU, so even if it was looking like maybe faster CPU were going to be in fashion for gaming, you've probably got another couple of years left before that's actually true (thanks to people switching APIs and so using less CPU resources to juggle the instructions for the GPU to render the scenes). Hell, even AI work is getting offloaded to GPUs in compute shaders because the performance of CPUs plateaued over the last few years and everyone saw GPUs as the great hope to avoid being stuck with what we had 5 years ago.

Glad to finally hear some news on this. I've been wanting cards from this line because the jump will last me for such a long time but I wonder if my i7 4770 is enough to take advantage of the new card

@colourful_hippie: Definitely yes - the 4770K is more than you need to feed a modern top-end overclocked GPU and that should continue to be true for these next gen of cards. You're barely behind a brand new 6xxx series Intel CPU with DDR4 when feeding a Titan X so I don't think these new gen of cards will show a massive issue.

Also, while tests for framerates will show the increasing load on a gaming CPU, that's not actually how you'll be using your CPU. If you've got a 60Hz limited monitor then you'll be upgrading your GPU and making the game show more shinies at... 60Hz. You don't need your CPU to be calculating 110 different world states every second and sending them to the GPU to render for 110fps. You only need it to calculate 60 of them and then send them to the GPU, which you will load down to only just finish them in time by cranking every setting up you can find (there are never enough settings to "max out" a game because you can almost always render at a higher resolution and so on and end up with art) at a modest CPU cost (like, seriously, what more does a CPU need to do when the resolution is increased if the GPU doesn't run out of VRAM and need to get new data shuffled in during rendering? CPU don't care).

So even with a somewhat unwarranted burden on the CPU from that test of an i5-2500K and an i7-3770K, they still seem completely fine driving a modern monster GPU. In the real world, I think they'll actually be doing a bit better (because they'll be feeding the GPU closer to 60Hz at max settings for all the possible CPU configs, assuming v-sync). Either way, I don't expect upgrading only your GPU will cause any issues or radically lower frame-rates compared to someone with the same GPU but a brand new system. Definitely not worth the cost of replacing the CPU, mobo, and RAM to build out a new system for gaming.

Something like an i7-4770K will definitely not be bottlenecking the system for games as long as you turn all the settings up, and if you're getting a new nice GPU then it'll be so you can turn all the settings up. I know that's why my GTX760 is getting replaced: not enough VRAM for all the high-detail textures needed (actually an issue in games ported from these new consoles), not enough perf to even get much beyond medium in the Division [it still looks lovely, that engine is impressive, but it could look nicer] to hit 1080p60, and it's time to buy something to let me enjoy VR-grade rendering and beyond.

@shivoa: Thanks for the info man! Didn't mean to hijack this thread though...
I actually have that cooler already, as I had some problem with heat on the CPU before this one. Feel kinda stupid having gone so long without taking advantage of it.

@shivoa: appreciate the info! Looking forward to seeing the consumer grade cards releasing