[WCCFTECH] NVIDIA GeForce GTX 1180 leaked info and performance

[Guru]

8GB is rarely needed because game developers are conserving data, because most GPU's are still 4GB of vram, so they need to create games that work best with most gpu's, that means 3GB and 4GB GPU's. If ALL gpu's were 8GB tomorrow, you'd see that all new games would take advantage of that and most will consistently use more than 6GB of vram.

We can easily have games use up to 16GB of vram, issue is very few cards have that amount of vram. Again you could have a GTX 2080ti with 10k cores, 16GB vram and there will be games that still make it bow down. All we need to see is a half decent v-ray implementation and it will tank the performance even on a 2080ti with 10k cores and 16GB vram. Very high levels of tessellation, particle physics, huge amount of soft shadows and dynamic lightning will also tank performance on any card.

 

[Fallen Kell]

Doesn't Fallout 4 also run much smoother on 8GB cards, in comparison to 2GB and 4GB cards?

Yes, and there is a similar modding community that has been releasing higher resolution texture packs, higher polygon models/skeletons, and improved lighting/shadows for both these games. And it simply comes down to does your card have enough memory and power to run it.

I mean, seriously, go look at some of the screenshots of some of those mods out there. We are simply at the mercy of how fast the graphics cards are for this game engine since the publisher has encouraged and helped create such a good modding community and not restricted them in any real way. People have been essentially competing with each other for making the best looking additions to the game. Just take a look at lush grasses for Skyrim and tell me that graphics cards could have done that when the game released...

 

[Headfoot]

Fallout gets bottlenecked on CPU single thread speed very quickly, even in vanilla, much less with mods that add lots of objects. Huge draw call counts.

And the "Games with mods can use lots of VRAM" argument is fine I guess, but yeah we all realize if you load in mods whose purpose is to use more memory, it's gonna use more memory. Its the exception that proves the rule.

 

[William Gaatjes]

Fallout gets bottlenecked on CPU single thread speed very quickly, even in vanilla, much less with mods that add lots of objects. Huge draw call counts.

And the "Games with mods can use lots of VRAM" argument is fine I guess, but yeah we all realize if you load in mods whose purpose is to use more memory, it's gonna use more memory. Its the exception that proves the rule.

That makes me wonder what can also be done, is that a game publisher limits the vram use but the gpu designer might come up with performance optimizations in the driver. For example based on lookup tables that are calculated in advance and layout in memory and not calculated real time. When these values are needed, a simple lookup might in some ocassions be faster than calculating.
Or some performance enhanced texture filtering and blending that needs lot of memory. All that is static can be precomputed and stored in available video ram andf available when needed.
Then again, one cannot exchange everything for large lookup tables in memory because it will eat in available memory bandwidth.



edit. was system ram.

 

[Ajay]

It's 10 nm.

This close to 7nm, interesting choice. With the delay I thought for sure that NV was holding out for 7FF.

 

[Ken g6]

This close to 7nm, interesting choice. With the delay I thought for sure that NV was holding out for 7FF.

It's probably because AMD isn't giving them good competition. This way they can wait another two years and produce another die shrink with whatever process makes sense then.

 

[jpiniero]

This close to 7nm, interesting choice. With the delay I thought for sure that NV was holding out for 7FF.

Not really. Remember it was like at least 8 months from when the first 16 nm products showed up before nVidia released the 1080. They want yields to be in great shape first. They seem to be waiting for 7FF for Tesla though.

 

[Samwell]

This close to 7nm, interesting choice. With the delay I thought for sure that NV was holding out for 7FF.

You can't launch 7nm consumer gpus before mid 2019. Means you had to wait for 1 more year. And it's not 10nm, it'll be 12nm.

 

[jpiniero]

You can't launch 7nm consumer gpus before mid 2019. Means you had to wait for 1 more year. And it's not 10nm, it'll be 12nm.

The thing is that 12 nm doesn't really offer much of a shrinkage. Kind of have to go to 10 nm to make it worth the effort. If they aren't going to do 10 nm, they would just rebrand GP102... and that seems kind of silly given how the prices are because of miners.

 

[MadOver]

Anyways, wont be a mid range GPU to start, so AMD should be fine..

 

[Ajay]

You can't launch 7nm consumer gpus before mid 2019. Means you had to wait for 1 more year. And it's not 10nm, it'll be 12nm.

As already mentioned, that’s nuts - 12 FF is just a tweaked 16 FF?!

 

[Qwertilot]

As already mentioned, that’s nuts - 12 FF is just a tweaked 16 FF?!

Did you see how much better gv100 was in performance /watt than gp100? A whole mass of compute specific stuff limiting its gaming performance of course.

If they launch 12nm gpu's they'll give the expected increases over Pascal.

 

[Samwell]

The thing is that 12 nm doesn't really offer much of a shrinkage. Kind of have to go to 10 nm to make it worth the effort. If they aren't going to do 10 nm, they would just rebrand GP102... and that seems kind of silly given how the prices are because of miners.

No, 10nm is too expensive for the effort. Designing a 12nm chip is cheap. A rebrand GP102 will give you max 10%, that's nothing. New architecture on 12nm should be around 40% faster with Diesizes of a bit over~400 and 600mm². As 7nm is very near, this should be a fast Generation, like GM200, which only lived for 15 months. There is already Volta in 12nm and other chips in 16nm, therefore many chipparts are already designed for the process or the changes are small. Volta Rops, TMUs,new cache architecture, great, you just more or less "copy and paste" them in a different chip. In 10nm, you need to develop every single part of the chip new and the problem is that 10nm has tripple patterning, which makes development much harder.

As already mentioned, that’s nuts - 12 FF is just a tweaked 16 FF?!

Maxwell also had no new process, 10nm is just not the way to go because it's not good enough for the development cost. 40% should be ok for the moment till 7nm gets here in end of Q4 19, Q1 2020.

 

[Tweak155]

https://wccftech.com/nvidia-geforce-gtx-1180-12nm-database/

 

[jpiniero]

No, 10nm is too expensive for the effort. Designing a 12nm chip is cheap. A rebrand GP102 will give you max 10%, that's nothing. New architecture on 12nm should be around 40% faster with Diesizes of a bit over~400 and 600mm².

You'd have to think if they were just going to use Volta they would have released it by now. And that would be awfully big for x104 and x102, even though are going to charge a ton for it.

I actually think they will skip the first gen 7 nm for gaming chips and use the second gen with partial EUV.

 

[Samwell]

You'd have to think if they were just going to use Volta they would have released it by now. And that would be awfully big for x104 and x102, even though are going to charge a ton for it.

I actually think they will skip the first gen 7 nm for gaming chips and use the second gen with partial EUV.

It won't be pure Volta, because Volta is lacking too many graphics features. More like a Volta 2, in a way comparable with Maxwell 1 and 2. Maxwell 1 had the shader structure changed, but most graphics stuff was from kepler and it had Keplers DX11.0 Feature Set. Maxwell 2 didn't change shaders, but added DX12.1. Volta has the same Feature Set as Pascal and is still lacking some DX12 features, which Vega and Skylake got. I'm sure the next graphics gen will get the full DX12.1 Feature Set. But for that they need a few architecture changes additionally to the stripping off pro features from the chips. Maxwell X04 and X02 were the same size, 16nm is now old enough and not so expensive anymore. The problem is Ram, i think we will see 16GB with the 1180, but it'll be damn expensive.

As for 7nm, depends how fast they can get 2nd gen 7nm. They won't give amd more than 6 months advantage in the new process.

 

[Mopetar]

This close to 7nm, interesting choice. With the delay I thought for sure that NV was holding out for 7FF.

Why? NVidia will get a lot of people to buy twice if they come out with a new card in the next few months and release a 7 nm card ~18 months later.

They know that they've got a big chunk of the enthusiast market, so I don't a reason for them not to put out a new card as soon as they can instead of waiting for a new process that may take quite a while to mature. Add in a lack of competition from AMD (combination of performance and availability), and there's no real risk to NVidia having AMD leapfrog them.

 

[cytg111]

So, a 30% improvement over 1080ti? Not bad, not enough to upgrade but not bad!

 

[Qwertilot]

That would be upper mid range don't forget. Unless they've radically changed strategy (why?!), there should eventually be a 'biggest' chip (1080ti) which will be ~30% further on.

 

[cytg111]

That would be upper mid range don't forget. Unless they've radically changed strategy (why?!), there should eventually be a 'biggest' chip (1080ti) which will be ~30% further on.

Yup, that might be my ticket.

 

[TheF34RChannel]

That would be upper mid range don't forget. Unless they've radically changed strategy (why?!), there should eventually be a 'biggest' chip (1080ti) which will be ~30% further on.

I'll be getting both sequentially. Need some more power asap but can't wait for the Ti.

 

[happy medium]

It look like we will see a 50% performance jump over the 10xx series,
https://wccftech.com/873853-2/

 

[alcoholbob]

I have no idea whether this is remotely accurate or not, but the performance chart shows a 50% increase over the last gen card its replacing. So if true, would not call that a "crappy gain".

Cards replace price brackets to a consumer, not some artificial internal tier bracket. Fact is you can't say a 699 card is "replacing" a 549 card to consumers, to them it's a 699 card replacing a 699 card. Incremental progress.

 

[TheF34RChannel]

It look like we will see a 50% performance jump over the 10xx series,
https://wccftech.com/873853-2/

Just their typical tabloid conjecture nothing to see there.

 

[happy medium]

Just their typical tabloid conjecture nothing to see there.

But yet they got most of it right with the gtx1080/1070.....
https://wccftech.com/nvidia-introduce-gtx-1080-gtx-1070-battlefield-5/

they did a good job with the gtx1080ti also.
https://wccftech.com/nvidia-pascal-refresh-geforce-20-speculation/

So yes there is plenty to see here.

I'm sure a gtx1180 for $599 that is even 35% faster than a $700 gtx1080ti will fly off the shelves.