Upcoming GPUs (2016)

[MiRai]

nVidia


GTX 1080

MSRP (Reference): $700
MSRP (Aftermarket): $600
VRAM: 8GB

Reviews:
Guru3D (http://www.guru3d.com/articles_pages/nvidia_geforce_gtx_1080_review,1.html)
HardwareCanucks (http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/72619-nvidia-geforce-gtx-1080-review.html)
Hard|OCP (http://www.hardocp.com/article/2016/05/17/nvidia_geforce_gtx_1080_founders_edition_review)
Hexus (http://hexus.net/tech/reviews/graphics/92846-nvidia-geforce-gtx-1080-founders-edition-16nm-pascal/)
TechPowerUp (https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/)
Tom's Hardware (http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1080-pascal,4572.html)

There is a useful mega-thread of reviews (both written and video) on the PC Master Race sub-reddit:
https://www.reddit.com/r/pcmasterrace/comments/4jqpzw/nvidia_benchmark_review_megathread/



Notes:


GP104 (Pascal)
Uses newer GDDR5X (G5X) VRAM
"Founder's Edition" will be available on May 27th, 2016, custom GPUs to arrive at a later date
"Founder's Edition" is the new name for "Reference Card"
Limited to 2-way SLI, but 3/4-way are possible through an "enthusiast key" (https://www.youtube.com/watch?v=2wBDt9tN5-c&t=7m10s) (no direct support from nVidia on 3/4-way)




GTX 1070

MSRP (Reference): $450
MSRP (Aftermarket): $380
VRAM: 8GB

Reviews:
NDA Lifts on May 17th

Leaks:
None

Notes:


GP104 (Pascal)
Uses older GDDR5 VRAM
"Founder's Edition" will be available on June 7th, 2016, custom GPUs to arrive at a later date
"Founder's Edition" is the new name for "Reference Card"
Limited to 2-way SLI, but 3/4-way are possible through an "enthusiast key" (https://www.youtube.com/watch?v=2wBDt9tN5-c&t=7m10s) (no direct support from nVidia on 3/4-way)






AMD

AMD will be revealing Polaris 10/11 at the end of May, but they are not expected to directly compete with GTX 1080/1070. AMD Vega architecture to be released later this year, which will include the new HBM2 memory.

More information will be added for AMD at a later date.

[MiRai]

________________________GPUs from 2015


nVidia


TitanX

MSRP: $1,000
VRAM: 12GB

Reviews:
Anandtech (http://www.anandtech.com/show/9059/the-nvidia-geforce-gtx-titan-x-review)
Guru3D (http://www.guru3d.com/articles-pages/nvidia-geforce-gtx-titan-x-review,1.html)
Hard OCP (http://www.hardocp.com/article/2015/04/14/nvidia_geforce_gtx_titan_x_video_card_review)
HardwareCanucks (http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/68992-nvidia-titan-x-performance-review.html)
KitGuru (SLI) (http://www.kitguru.net/components/graphic-cards/zardon/nvidia-geforce-gtx-titan-x-12-gb-sli/)
Legit Reviews (http://www.legitreviews.com/nvidia-geforce-gtx-titan-x-video-card-review_160070)
TechPowerUp (https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_Titan_X/)
Tom's Hardware (http://www.tomshardware.com/reviews/nvidia-geforce-gtx-titan-x-gm200-maxwell,4091.html)
TweakTown (SLI) (http://www.tweaktown.com/articles/7049/nvidia-geforce-gtx-titan-12gb-sli-two-much-better-one/index.html)
HardwareInfo (2/3/4-way SLI) (http://us.hardware.info/reviews/6033/nvidia-geforce-gtx-titan-x-sli--3-way-sli--4-way-sli-review-insane-performance!)

Notes:


Fully unlocked GM200-400 chip
3,072 CUDA Cores
Has the tendency to run warm which could result in throttling
Hybrid AIO WC from EVGA (http://www.evga.com/articles/00935/EVGA-GeForce-GTX-TITAN-X-HYBRID/)
ACX 2.0+ Cooler replacement from EVGA (http://www.evga.com/Products/Product.aspx?pn=100-FS-2990-B9)




980Ti

MSRP: $650-$750 (Different models vary in price)
VRAM: 6GB

Reviews:
Anandtech (http://www.anandtech.com/show/9306/the-nvidia-geforce-gtx-980-ti-review)
Bit-Tech (http://www.bit-tech.net/hardware/graphics/2015/05/31/nvidia-geforce-gtx-980-ti-review/1)
Forbes (http://www.forbes.com/sites/jasonevangelho/2015/05/31/nvidias-new-gtx-980-ti-delivers-titan-x-level-performance-for-350-less/)
Guru3D (http://www.guru3d.com/articles-pages/nvidia-geforce-gtx-980-ti-review,1.html)
HardwareCanucks (http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/69502-nvidia-gtx-980ti-performance-review.html)
KitGuru (http://www.kitguru.net/components/graphic-cards/zardon/nvidia-gtx980-ti-review/)
TechPowerUp (http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_980_Ti/)
Tom's Hardware (http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-ti,4164.html)
TweakTown (http://www.tweaktown.com/reviews/7170/nvidia-geforce-gtx-980-ti-reference-video-card-review/index.html)
Zol (Chinese) (http://vga.zol.com.cn/522/5229872.html)


Babeltech (Large Overview) (http://i1.wp.com/www.babeltechreviews.com/wp-content/uploads/2015/05/TheBigPicture.jpg) (No idea how accurate or trustworthy)

Notes:


Slightly cut-down GM200-310 chip
2,816 CUDA Cores
About 5-10% slower than a Titan X



-------------------------------------------------------------------------------------------------------------------------------


AMD


Fury X

MSRP: $650
VRAM: 4GB

Reviews:
Anandtech (http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review)
Guru3D (http://www.guru3d.com/articles-pages/amd-radeon-r9-fury-x-review,1.html)
Hexus (http://hexus.net/tech/reviews/graphics/84170-amd-radeon-r9-fury-x-4gb/)
Hot Hardware (http://hothardware.com/reviews/amd-radeon-r9-fury-x-review-fiji-and-hbm-put-to-the-test)
Legit Reviews (http://www.legitreviews.com/amd-radeon-r9-fury-x-video-card-review_167134)
Overclock3D (http://www.overclock3d.net/reviews/gpu_displays/amd_r9_fury_x_review/1)
Tech Report (http://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed)
Tom's Hardware (http://www.tomshardware.com/reviews/amd-radeon-r9-fury-x,4196.html)
TweakTown (http://www.tweaktown.com/reviews/7217/amd-radeon-r9-fury-high-bandwidth-memory-hbm-video-card-review/index.html)

Notes:


The first GPU to ever be equipped with HBM (High Bandwidth Memory)
AIO/CLC (All-In-One / Closed-Loop Cooling) requires you to find a place to mound the 120mm radiator
Runs quiet and cool
HBM is very fast memory which may help if you're pushing its 4GB limit as is shown in numerous 4K benchmarks
There will be an air-cooled version in the future



Review site choices based on either:


1) The first few that show up in Google
2) Sites that I've read in the past
3) Other mega-thread review threads that I find on other forums >_>

[MiRai]

The GTX 1080 reviews and benchmarks have begun to surface since the NDA was lifted earlier today. It looks to be 20-30% faster than 980Ti/Titan-X*, which, in my opinion, was expected.

*Unless you are willing to overclock, which then you can expect more performance, but may be limited by thermals.

I will be continuing to update the reviews slowly as they trickle out.

Also, as for pricing, it does seem that custom GTX 1080 boards (e.g. non-reference) will not be available until after Computex, so people will be paying the premium $700 price tag if they want an early version of the GPU.


I suspect that this approach is NVIDIA's way to milk early adopters - in the first weeks there will be only the Founders Edition available, until board partners come out with their custom designs, which will be well after Computex. I'm also worried that this could tempt board partners to pick $699 as baseline for their custom design products and that we'll see very few if any cards at $599. All this is quite unknown though at the moment, so let's wait a bit and see what happens.

Source (https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/32.html)

[Lyonheart]

My 980ti will be ok until I get a 4k monitor. By then there should be a 1080ti available.

[emitchell109]

Looks like I will have a reason to upgrade... Or at least a slim justification!
What are your plans upon release Mirai? Do you have a preferred board developer that you wait/look for? Or do you go with the founders editions?

[MiRai]

Looks like I will have a reason to upgrade... Or at least a slim justification!
What are your plans upon release Mirai? Do you have a preferred board developer that you wait/look for? Or do you go with the founders editions?
Personally, I go with EVGA because of their support and warranty options, but for this launch I would wait a few weeks to see what is actually happening with everything, but either way I won't be upgrading my hardware until the next iteration of GPUs, at the earliest.

People are very confused as to what nVidia is trying to pull, or if they really have enough stock of GDDR5X to have an ample supply of GTX 1080s, because this $100 "Founder's Tax" is really, really odd. I've been reading non-stop about this, and only a few seem to be actually defending this when everyone else is skeptical as to what is actually going to happen. nVidia also couldn't really explain what is so special about a Founder's Edition and why it requires the additional $100 price tag, and why board partners (e.g. EVGA, Asus, Zotac, MSI, PNY, etc) should even bother to price their boards at the MSRP of $599.

nVidia is saying...



It costs more to manufacture, but that hasn't been true about reference cards in the past since they've always been the cheapest.
The chips aren't binned (meaning they don't favor better overclocking)
The new vapor chamber cooler costs more and is better than prior versions of stock coolers, but it has already been shown to be inefficient which ends up throttling the GPUs (https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/29.html).
That they cannot speak for when partner GPUs will be available.
They used "premium materials" -__-


I'll be these nVidia guys couldn't wait to get off of the stage...




https://www.youtube.com/watch?v=bNCfn4y8dBw

[Hor]

I'm honestly thinking I will either go dual 1070s later, or grab a secondhand 980 TI and SLI with the one I already run since it's not like they're going to suddenly suck.

[MiRai]

I'm honestly thinking I will either go dual 1070s later, or grab a secondhand 980 TI and SLI with the one I already run since it's not like they're going to suddenly suck.
Yeah, if you're going to go with 1070s I would definitely wait for the price to come down, but people are trying to dump their 980Ti/TTX GPUs for pretty cheap, so if you want to grab something like an EVGA card (or some other manufacturer which allows for a transferable warranty), then I don't see 980 Ti SLI being a bad move, since you can likely 5-box at 4K with a few settings turned down to stay under 6GB and keep performance good.

[Hor]

Yep, and that's the winner. I picked up an MSI 980 TI for 350, so it'll match the one I already have. If the leaks are true, I'll definitely be moving to a pair of 1070s though, move the TIs to computer 2.

http://www.overclock3d.net/articles/gpu_displays/gtx_1070_3dmark_results_leaked/1

Thought you might be interested in seeing some of the info on the 1070, if it's legit it's definitely a good looking card.

[JohnGabriel]

The NVIDIA GeForce Experience now supports the GTX 1070. That means we can buy them soon? My 980 is giving errors and I cant wait much longer.

http://www.geforce.com/hardware/10series/geforce-gtx-1070

[Hor]

Yeah John, reviews of the Founders and board partner cards are starting to surface.

https://www.youtube.com/watch?v=Ce7-lv6pTv4

Here's a good one of the FE 1070, with an SLId MSI version coming shortly.

All looks very promising, I'm still waiting in the weeds to see how the new AMD cards look, if the 480 is as good as they claim, I suspect I might go team Red this time around.

[Pazgaz]

Just got 2 x 1080.
Waiting for the water blocks to arrive (should be today), and then gonna spend the weekend assembling and testing everything.
Just in time for legion :D

Sent from my A0001 using Tapatalk

[Pazgaz]

Before:
1721

After:
1722


And for some numbers:
I can run 5 instances of WoW at 4k resolution, everything set to max except AA. Getting 60 fps on master and 30 on slaves SUPER stable when running outside killing mobs and stuff.
I've yet to try Stormshield. Will update.

[Pazgaz]

Actually that was running on overclocked settings. When not overclocked it does go below 60 fps.

[Lyonheart]

I read that SLI scaling (http://www.guru3d.com/articles_pages/geforce_gtx_1080_2_way_sli_review,20.html) for the new cards is bad in most cases, due to the drivers not being optimized yet. So you might get better results in the future when updated drivers support SLI on these 10xx cards.

Im holding out for 1080TI or Titan (http://vrworld.com/2016/07/05/nvidia-gp100-titan-faster-geforce-1080/) versions of 10xx cards.

[MiRai]

I read that SLI scaling (http://www.guru3d.com/articles_pages/geforce_gtx_1080_2_way_sli_review,20.html) for the new cards is bad in most cases, due to the drivers not being optimized yet. So you might get better results in the future when updated drivers support SLI on these 10xx cards.
I was really disappointed in that article when I looked at it, and it seems that I'm disappointed in most review sites which are doing the 2-way SLI 1080 benchmarks, and that's because they're not putting them up against any other meaningful multi-GPU benchmarks. It just doesn't make any sense.

I mean... HTF can you review a GPU in SLI when you don't put it up against any other multi-GPU competition? That just doesn't make any sense to me. TechPowerUp did a very similar thing and put up 1080 SLI against 970 SLI (https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080_SLI/20.html), which is just as weird as not putting it up against any multi-GPU benchmarks at all, but as for a bonus, they also have an aftermarket Asus GPU which they're putting up against stock cards of the prior generation. Look how much faster this overclocked, aftermarket 1080 is compared to a stock 980 Ti!

Again, I'm not saying that the 1080 isn't a nice GPU, but this all just feels like such a weird release to me because it seems like information is being hidden, or skewed. Especially when you look at nVidia's most recent jab at AMD:

1723

It's convenient that the graph starts at 0.8, but someone from the Anandtech forum fixed the graph (http://forums.anandtech.com/showpost.php?p=38330016&postcount=18).



Im holding out for 1080TI or Titan (http://vrworld.com/2016/07/05/nvidia-gp100-titan-faster-geforce-1080/) versions of 10xx cards.
There's another thread on Anandtech (http://forums.anandtech.com/showthread.php?t=2479276) where people were talking about that article, and I agree with this particular post (http://forums.anandtech.com/showpost.php?p=38336508&postcount=11) at the moment. However, I'd also assume a price tag of at least $1200 if a Titan was going to show up anytime soon.

Titan "P" for "Privilege" :)

[Jar]

I'm waiting for Pauls Hardware reviews the 1060.....

[Pazgaz]

I think it's going to be 1200 for the 12gb ver and 1600 for the 16gb ver for titan.

[MiRai]

I think it's going to be 1200 for the 12gb ver and 1600 for the 16gb ver for titan.
Two reasons why I don't believe this to be true...

1) The market segment for the Titan GPUs is already incredibly small, and splitting them up into two, even smaller segments is probably going to hurt profits, more than help. I doubt nVidia is going to bother to spend extra money in order to manufacture two different types of ultra-top-end GPUs just to try and price gouge people at $100 per GB of VRAM.

2) The amount of VRAM on the PCB is determined by the GPU's memory interface:

GDDR5/X 256-bit = 2/4/ 8/16/32GB
GDDR5/X 384-bit = 3/6/12/24/48GB
HBM2 4096-bit = 4/8/12/16/20/24GB? (I'm guessing here, but it could also be 4/8/16/32GB like we see with 256-bit)

Now, all of the articles say that nVidia is going to "unveil" this new GPU, but that could mean anything. It could mean only spoken words with a picture on screen with no model or solid release date in sight, or a non-working mock-up model of the GPU, or a fully working GPU... we have no idea.

However, if nVidia is going to actually unveil a new Titan GPU at GamesCom and announce a release date in the near future, then my guess is that it's going to be G5X memory, and not HBM2. HBM2 isn't even available in the wild at this time, so for nVidia to offer it on a consumer-level GPU is absurd, because they release their GPUs like this: Tesla (Supercomputing/Deep Learning) > Quadro (Workstation) > GeForce (Gaming). I highly doubt that nVidia is going to cut into the super-scarce supply of HBM2 just so they can charge pennies on the dollar for it, when they could easily be making thousands of dollars per GPU in the Tesla or Quadro market.

So, with that said, if the GPU is going to be using the standard G5X VRAM, then it's either going to be 256-bit or 384-bit, and you can't have both 12GB and 16GB models on the same interface.

[mbox_bob]

2) The amount of VRAM on the PCB is determined by the GPU's memory interface:

GDDR5/X 256-bit = 2/4/ 8/16/32GB
GDDR5/X 384-bit = 3/6/12/24/48GB
HBM2 4096-bit = 4/8/12/16/20/24GB? (I'm guessing here, but it could also be 4/8/16/32GB like we see with 256-bit)
Err, the GB is determined by the density of the VRAM chips. The interface determines the number of chips.
GDDR5(X) chips have a 32bit interface, so you take you GPU interface, divide by 32 and that is the number of VRAM modules you will have on the PCB.
HBM1 is 1024 bits wide.

It then becomes a simple bit of math. gpu interface / VRAM chip interface * density / 8.

Funnily enough, the top end graphics cards tend to use the highest density chips available at the time, which is currently 8Gb (gigabit, not bytes), but it is not mandatory.

/e: it is possible to put more vram on the board, but this would then require the memory controller to switch between modules, so you would not gain any extra "bandwidth" out of doing so, and would probably then generate extra latency (in fact this is what the GTX 970 does, so it's 256bit interface is just marketing gumpf because it is not a true 256bit pipeline throughout the whole shebang).

/e2: The Titan-X (because someone will bring it up), does the memory controller switch to access the 24 VRAM modules (it has 4Gb modules), the internals of how it does it without impacting significantly on performance, who knows, and why the 970 does it so inefficiently in comparison, who knows. My guess is that the Titan-X does not switch to accessing a single module at a time, it is always addressing 12 modules simultaneously, whereas the 970 goes from addressing 7 modules to 1 module, then back to 7 etc (although only when it really needs that extra 512MB).

[MiRai]

bob, I know you like to pick on me, but here's where I'm coming from.




GPU
Memory Interface
Video RAM


GTX 460
192-bit / 256-bit
1GB / 1GB


GTX 470
320-bit
1.2GB or 2.5GB


GTX 480
384-bit
1.5GB


GTX 560
256-bit
1GB


GTX 560 Ti
256-bit / 320-bit
1GB / 1.2GB


GTX 570
320-bit
1.2GB


GTX 580
384-bit
1.5GB or 3GB


GTX 660
192-bit
2GB


GTX 660 Ti
192-bit
2GB


GTX 670
256-bit
2GB or 4GB


GTX 680
256-bit
2GB or 4GB


GTX 760
256-bit
2GB or 4GB


GTX 770
256-bit
2GB or 4GB


GTX 780
384-bit
3GB


GTX 780 Ti
384-bit
3GB


GTX Titan
384-bit
6GB


GTX Titan Black
384-bit
6GB


GTX 960
128-bit
2GB


GTX 970
256-bit
4GB


GTX 980
256-bit
4GB


GTX 980 Ti
384-bit
6GB


GTX Titan-X
384-bit
6GB


GTX 1060
192-bit
6GB


GTX 1070
256-bit
8GB


GTX 1080
256-bit
8GB




So, while I'm sure there is plenty of math involved, I think we can safely use a cheat sheet and say...

GDDR5/X 128-bit/256-bit = 1/2/4/8/16/32GB
GDDR5/X 192-bit/384-bit = 1.5/3/6/12/24GB

The only GPUs which deviate from this are the older 192-bit x60 GPUs, which only came with 1GB or 2GB each. I also leave out the 512-bit memory interface because AMD is the only one who has used it in recent times, but it follows the same guidelines where it's just twice of the 256-bit interface.


HBM1 is 1024 bits wide.

I was only going off of what AMD prints on their website (http://www.amd.com/en-us/products/graphics/desktop/r9), and what the sites which reviewed and benchmarked both the R9 Nano and R9 Fury X were stating, but I know very little about HBM beyond that.


Ground-breaking performance requires big bandwidth. AMD’s latest innovation delivers more than 3x the bandwidth per watt over GDDR5 along with a 4096-bit memory interface for incredible new advances in power and efficiency.

[mbox_bob]

:).... I was just saying.

I completely agree with your simplification; it was just that proposed 48GB card!!. You're going to start rumours.

/e re: HBM
In HBM1, each stack interface was 1024bits wide with 2Gb chips (256MB), 4 chips per stack (1GB per stack), the GPU memory controller interface was 4096bit, so, 4 stacks are needed.

HBM2 looks to be coming in 1024bits (still), with 8Gb chips in 4 or 8 chip stacks (rumours from Samsung), i.e. 4GB per stack (currently), potentially 8GB in a stack! With a 4096bit GPU interface, and it will pave the way for 16GB or potentially 32GB cards.

The math gets harder with HBM as you need to add in the number of chips per stack.

[Pazgaz]

http://www.overclock.net/t/1604953/various-pascal-titan-could-be-50-faster-than-the-geforce-gtx-1080-rumoured-to-launch-in-august

VR World is reporting that they've had hands-on time with the GP102-based GeForce GTX Titan card, with a few details on NVIDIA's new monster.

NVIDIA's next-gen Titan X successor will feature both 8+8 and 8+6-pin PCIe power connectors, with the PCIe power connectors being found at the front of the card, and not on top. The 8+8-pin power connectors will provide up to 375W TDP, while the 8+6-pin version will provide 300W. Performance-wise, it should be a monster with around 50% more horsepower over the already lightning quick GTX 1080.

The new Pascal-based Titan X successor will reportedly be 12 inches long, while the new Titan X successor will arrive in 12/16GB versions (as we've previously reported) rocking HBM2 memory. VR World's sources say that the new GPU is actually now bound by the CPU, with NVIDIA's engineers reportedly saying that even Intel's new Core i7-6950X isn't powerful enough to deliver the performance the new Titan cards need in enthusiast level scenarios.

We should expect NVIDIA to unveil its new Titan P at Gamescom, which takes place in Cologne, Germany between August 17-21, 2016.

[mbox_bob]

while the new Titan X successor will arrive in 12/16GB versions (as we've previously reported) rocking HBM2 memory.
Makes sense based on what they say about the 4096bit interface or 3072bit interface. Presumably the lessor interface will be because part of the memory controller is disabled (whether due to bad etching or just because).

[Pazgaz]

Also, if you look at the rx480, some of the cards have all 8 mem chips with 4 disabled while still having the same mem interface as far as I know.

[MiRai]

I completely agree with your simplification; it was just that proposed 48GB card!!. You're going to start rumours.
Well, we've already got GPUs with 24GB (http://www.newegg.com/Product/Product.aspx?Item=N82E16814133623), so 48GB probably isn't all that far off, but I don't think I'd try to game on them (or even want to afford them).


Also, if you look at the rx480, some of the cards have all 8 mem chips with 4 disabled while still having the same mem interface as far as I know.
I didn't say you can't have GPUs with two different amounts of VRAM on it, I said 12GB and 16GB were unlikely and I explained why. If you look at the table I created above you can see several different GPUs throughout the years which have had two different amounts of VRAM offered on it.

The RX 480 has a 256-bit interface, so it's capable of having any of the following amounts of VRAM on it:

GDDR5/X 128-bit/256-bit = 1/2/4/8/16/32GB

[MiRai]

The new Titan X has been revealed:

http://videocardz.com/62498/nvidia-announces-geforce-gtx-titan-x-pascal

12GB of G5X, and it's $1200 USD; available August 2nd (direct from www.nvidia.com).

[Lyonheart]

The new Titan X has been revealed:

http://videocardz.com/62498/nvidia-announces-geforce-gtx-titan-x-pascal

12GB of G5X, and it's $1200 USD; available August 2nd.

Wow! that seemed fast!?!? I wonder how long until 1080ti!

[MiRai]

Wow! that seemed fast!?!? I wonder how long until 1080ti!
If we go by the typical nVidia release schedule as of late, then I'd expect them to be released in October or November... just in time for the holidays. :)

[Pazgaz]

People are already saying this is a "small" Titan X (whatever that means) and that there will be another one later on (2017) with more CUDA cores and more RAM.
As for the prices, that's what you get with zero competition.

[MiRai]

People are already saying this is a "small" Titan X (whatever that means) and that there will be another one later on (2017) with more CUDA cores and more RAM.
As for the prices, that's what you get with zero competition.
Yes, the chip that this new Titan uses is cut down, and is not the full Pascal chip. In the realm of consumer graphics cards (as of the current and prior generation of chips), the full chip is labeled as G_X00—where _ is the first letter of the architecture that they're on (Kepler, Maxwell, Pascal, etc.), and X is a number (usually 1 or 2). However, there can be sub categories beyond that, so...


GTX 980 - GM204 (GeForce Maxwell 204)
GTX 980 Ti - GM200 (also Titan-X Gen1)

GTX 1080 - GP104-400 (GeForce Pascal 104)
GTX 1070 - GP104-200
GTX 1060 - GP106-400
GTX Titan-X - GP102-XXX (GTX 1080 Ti may be the exact same chip, or slightly cut down)

The big Pascal chip with HBM2 is currently P100. Whether or not we see GP100 is up to nVidia, but I've read some things about people now speculating that we won't see HBM2 until Volta, which is the chip after Pascal.

[Hor]

I'm honestly beginning to think we may not see a TI this time. UNLESS AMD pulls a serious sucker punch, like say something that's closer to a 1080 in performance but at a lower price.

Also, it appears that AMD is working on a fix to address Crossfire in windowed mode, which IF it happens could suddenly make those cards a lot more interesting.

[Pazgaz]

windowed mode CFX is why i moved from AMD to NV.
4 x 290x and only one working when boxing. really annoyed me :)