TR Forums

https://wccftech.com/nvidia-geforce-gtx-1180-specs-performance-price-release-date-prelimenary/
GeForce GTX 1180 GTX 1080
Architecture Turing Pascal
Lithography 12nm FinFET 16nm FinFET
GPU GT104 GP104
Die Size ~400mm² 314mm²
CUDA Cores 3584 2560
TMUs 224 160
ROPs 64 64
Core Clock ~1600MHz 1607MHz
Boost Clock ~1800MHz 1733MHz
FP32 Performance ~13 TFLOPS 8.7 TFLOPS
Memory Interface 256-bit 256-bit
Memory 8-16 GB GDDR6 8GB GDDR5X
Memory Speed 16Gbps 10Gbps
Memory Bandwidth 512GB/s 320GB/s
TDP 170-200W 180W
Launch Q3 (July) 2018 July 20 2016
Launch MSRP ~$699 $599-$699 (Founder's)

(how do I display this in a nice table? Posting it takes out all the extra spaces I added)

Looks like a July launch
~50% performance improvement over the GTX1080.
Same TDP as the GTX1080

DPete27 wrote:

(how do I display this in a nice table? Posting it takes out all the extra spaces I added)

The Code tags are your friend.

+50% boost can be a major leap. But how would that impact the sales of the previous generation then, given they can't cut the prices by 50% less. The main reasons they won't release a tech that makes a giant leap ahead but rather will drip feed the new tech in slow and steady advancing varients to keep the older tech and the prices relevant for some time.

TheNinja wrote:

But how would that impact the sales of the previous generation then, given they can't cut the prices by 50% less.

Same way they've done every other generation transition -- gradually discontinue the previous generation. When they released Pascal they stopped making Maxwell parts: why would this generation be different?

K-L-Waster wrote:

When they released Pascal they stopped making Maxwell parts,

Isn't GeForce MX130 still Maxwell?
It seems like they will frequently just rename an old part and continue to sell it. e.g.: the Fermi GPU in GeForce GT420/GT430/GT440 became GeForce GT415M/GT420M/GT425M/GT435M became GeForce GT530 became GeForce GT520M/GT525M/GT540M/GT550M/GT555M became GeForce GT620/GT630 became GeForce GT620M/GT625M/GT630M/GT640MLE became GeForce GT730 became GeForce GT710M/GT720M became GeForce GT810M/GT820M over a five year stretch with only minor improvements.

To confuse the situation as much as possible, NVidia even sells multiple GPUs all under the same model number.

The same way they do every other release. The faster chips will be priced at a point that makes performance per $ slightly more attractive. If it is 50% more powerful, expect to pay 30%+ more. Gone are the days of them holding a price point and just giving us extra power for the same price.

That's why entry level chips like the 1060 are pushing what was once firmly enthusiast pricing, and the top of the range is up near $1,000. My 960 likely cost me about the same as a new 1050 right now, and I bet they aren't that far apart in terms of performance.

NarwhaleAu wrote:

My 960 likely cost me about the same as a new 1050 right now

"right now" prices are not a fair market assessment, since crypto mining has driven prices through the roof. For example, in January 2017 you could buy an RX470 for $130 (regretting not jumping on that deal), that card beats your $160 GTX960 by 50%.

K-L-Waster wrote:

Same way they've done every other generation transition -- gradually discontinue the previous generation. When they released Pascal they stopped making Maxwell parts: why would this generation be different?

..it might depend on their release of their mid-range cards. IF they delay it like they have this release, (insead of a typical 4 month follow-up), it could be a better profit ratio for Nvidia. Maybe ditch the 1080ti, and keep the rest of the range with a $50/75/100 cut in retail prices. Perhaps making the new card a poor value for miners, while generating greater interest in gamers to spend-more on the new higher performance card?

Of course they probably won't (and instead release the midrange cards 4th quarter), but it is a strange market right now.

I'll believe it when I see it.

My GTX 1080 is in dire need of an upgrade.

Nvidia can afford to wait as they are the performance king.

I really hope they don't push the TDP too far off the 1080. The 1080 is such a great card with a wonderful mix of power consumption and performance. Honestly I think it's one of the best products Nvidia has ever released.

Nvidia. HDMI 2.1 (with VRR) or you will have me at least considering AMD (or Intel *snickers to himself*).

This Tom's article is stating a July 30 launch date for Turing. Also, it was interesting to see the 3rd party card validation process laid out.

End User wrote:

I'll believe it when I see it.

My GTX 1080 is in dire need of an upgrade.

Nvidia can afford to wait as they are the performance king.

May I suggest THIS UPGRADE.

I'll be curious to see if this next generation will be enough to push AAA games at 4K @ 90FPS. My 1080Ti can handle 4K@60+FPS, but we'll need a great deal of grunt to tame 4K. And then the question becomes "Why run 4K@90FPS if 1440P@200FPS is possible?"

chuckula wrote:

End User wrote:

I'll believe it when I see it.

My GTX 1080 is in dire need of an upgrade.

Nvidia can afford to wait as they are the performance king.

May I suggest THIS UPGRADE.

Certainly, you may suggest anything... But I prefer my apples direct from the tree or supermarket. As far as gaming goes, when Apple stops screwing it up with proprietary goofiness I might take a second look. "Might" being the key operative word...

Judging from how well Pascal Founder's Editions did for Nvidia, they will almost certainly do it again. I'm guessing the actual time to upgrade will be when the 1180ti comes out, or at least AIB custom cards.

1180 might beat 1080ti, but it'll almost certainly launch at $700 minimum.

Will there be a 1180 Ti, though? It was safe to assume before because there was "big" Maxwell, Kepler, Pascal, etc... But is there a "big" Turing? If the rumors about the 1180 prices are true, I suspect there might not be a bigger version coming.

chuckula wrote:

End User wrote:

I'll believe it when I see it.

My GTX 1080 is in dire need of an upgrade.

Nvidia can afford to wait as they are the performance king.

May I suggest THIS UPGRADE.

Those are some powerful ARMs you've got there.

drfish wrote:

Will there be a 1180 Ti, though?

Of course, but they always start with the xx80 non-Ti version, then make a Titan card to sell for a fat profit for a few months, then unleash the xx80Ti which is equivalent to the Titan at ~60% of the cost.

I'd expect a similar delayed release of the Turing family as Pascal saw.

It just feels like it's going to be different this time around. If Turing is a dedicated gaming chip, there's nothing bigger to cut it down from like before. Volta never turned into a $1,500 gaming chip. If Ampere replaces Volta, where else does that leave gaming cards to come from? It's all speculation, but we could see Turing "Ti" as a card with extra RAM and slightly higher clocks instead of 50% more silicon like before (more like a 1070 vs. a 1080).

DPete27 wrote:

I'd expect a similar delayed release of the Turing family as Pascal saw.

I'll wait for the von Neumann family.

We totally need a Dijkstra card now also. If it tells you how bad and wrong your code is when running GPU compute apps, then so much the better.

drfish wrote:

Will there be a 1180 Ti, though? It was safe to assume before because there was "big" Maxwell, Kepler, Pascal, etc... But is there a "big" Turing? If the rumors about the 1180 prices are true, I suspect there might not be a bigger version coming.

IMO there almost certainly will be. What better way to make use of that shiny new 7nm process than to make a big stonking GPU after all. They'll build it for Teslas/Quadros and eventually trickle down to a Titan then a x80ti after each higher-end market has had its fill.

I posted this over at beyond3d, figured there are a few people here that may be interested:

Re: the leaks about a "new boost clock algorithm" for Turing:

It seems to me as though Nvidia's boost clock algorithm is already quite mature, and as such there is minimal clock headroom left for overclocking beyond what the boost algorithm can already achieve. If Nvidia wants to "pull an AMD" and push Turing toward the outer edge of the thermal/power envelope they will only gain perhaps 5% over what is already achievable. That leaves process improvements to make up the rest of the expected performance gains over Pascal, a notion I find dubious given the nature of the "12nm" (i.e. 16nm+) process involved.

All this makes me wonder, perhaps even hope (naively) if perhaps the leaks as they have been interpreted thus far don't quite tell the whole story. If Turing is to gain any sizable amount of performance over Pascal on the average (gaming workloads), given the same number of ALUs and only minor architectural changes, there will need to be some major clock speeds improvements to achieve this. Given what I've already mentioned about the likelihood of this prospect, to me this only leaves one possible solution:

The return of clock domains.

Discuss.

@techguy: The spec comparison in my OP seems to back up your suspicions. Nvidia is potentially leveraging a larger die (for more coarz) and faster GDDR6 RAM to achieve their performance improvements this time around while standing on the great architectural foundation they established with Pascal. It would interest me if reviewers could take steps to negate the >cores and >VRAM bandwidth on Turing to attempt to eke out what performance gains come directly from actual architectural improvements in Turing.

DPete27 wrote:

@techguy: The spec comparison in my OP seems to back up your suspicions. Nvidia is potentially leveraging a larger die (for more coarz) and faster GDDR6 RAM to achieve their performance improvements.

We should probably define the basis of comparison. I believe that GTX 1180 will be faster than GTX 1080 Ti, based on the latest rumors/leaks. Given the bus-width differential between these chips (256-bit vs. 384-bit), the additional clock speed of GDDR6 will only serve to bring the bandwidth up to approximately the same level as 1080 Ti/Titan X (Pascal). With that in mind, GDDR6 itself is unlikely to be a large contributor to any performance differential.

As for the die size difference, "GT104" will almost certainly be larger than GP104, though with fewer ROPs and narrower memory interface it will probably end up smaller than GP102.

Good point:

GeForce GTX 1180 GTX 1080Ti
Architecture Turing Pascal
Lithography 12nm FinFET 16nm FinFET
GPU GT104 GP102
Die Size ~400mm² 471mm²
CUDA Cores 3584 3584
TMUs 224 224
ROPs 64 88
Core Clock ~1600MHz 1480MHz
Boost Clock ~1800MHz 1582MHz
FP32 Performance ~13 TFLOPS 11.3 TFLOPS
Memory Interface 256-bit 352-bit
Memory 8-16 GB GDDR6 11GB GDDR5X
Memory Speed 16Gbps 11Gbps
Memory Bandwidth 512GB/s 484GB/s
TDP 170-200W 250W

Thanks for posting that table.

Quick corrections:
GTX 1080 Ti = GP102 and die size = 471 mm^2

Edit: I see you caught that already