Huge GTX 1180 leak from Tech Power Up, Volta based (possibly), 12nm Finfet node

[Master Disaster]

Tech Power Up have the GTX 1180 listed with lots of info...

 

Quote

The GeForce GTX 1180 is a graphics card by NVIDIA that is not released yet. Built on the 12 nm process, and based on the GV104 graphics processor, the card supports DirectX 12.0. It features 3584 shading units, 224 texture mapping units and 64 ROPs. NVIDIA has placed 16,384 MB GDDR6 memory on the card, which are connected using a 256-bit memory interface. The GPU is operating at a frequency of 1405 MHz, which can be boosted up to 1582 MHz, memory is running at 1500 MHz.

We recommend the NVIDIA GeForce GTX 1180 for gaming with highest details at resolutions up to, and including, 3840x2160 (4K).
Being a dual-slot card, the NVIDIA GeForce GTX 1180 draws power from 1x 6-pin + 1x 8-pin power connectors, with power draw rated at 200 W maximum. Display outputs include: 1x HDMI, 3x DisplayPort. GeForce GTX 1180 is connected to the rest of the system using a PCIe 3.0 x16 interface. The card measures 267 mm in length, and features a dual-slot cooling solution.

Quote

The GPU entry lists the specifications for the graphics card – most of which are identical to the leaks we have seen before. The NVIDIA GTX 1180 graphics card will be manufactured on TSMC’s 12nm FinFET process which will introduce significant power efficiency upgrades on a core to core basis.  The exact core count will be 3584 CUDA cores divided in 28 SMs. 64 ROPs and 224 TMUs make up the rest of the specifications. According to the same entry, the memory in question will be the GDDR6 variant with up to 16 GB worth of DRAM.

So about the whole Volta thing, there's a chance this might be an ID error and TPUs DB is detecting Turing as Volta, remember Turing is a process shrink of Volta so if TPUs DB hasn't been updated it's entirely possible that it's being incorrectly identified

Quote

Here is where things get slightly murkier, however, the TPU GPU DB entry lists the card as a volta variant, which is odd considering the previous leaks have told us that the architecture in question will be called Turing. This might simply be a MachineID error where the GPU-Z tool is reading the architecture incorrectly. Turing is essentially a process shrink and optimization of the Volta architecture and will bring significant performance and power efficiencies to the next-generation graphics card.

The GTX1180 is expected to launch around CES for anywhere between $699 & $799

Quote

The expected launch date for the next generation of NVIDIA Turing graphics cards is probably around the Computex time frame. Earlier reports have indicated that they could hit the shelves as soon as July which means a tease or soft launch at Computex next month might not be entirely out of the question. As far as pricing goes, keep in mind that the crypto bubble has mostly burst (as far as graphics card pricing goes anyways) so we are expecting a slightly toned down MSRP of somewhere around $699-799 but that sticker will mean nothing if retailers and e-tailers decide to charge a heavy premium for them depending on demand.

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

https://www.techpowerup.com/gpudb/3224/geforce-gtx-1180

 

Interesting stuff, $699 launch price but we all know they're going to be much more expensive than that.

 

Will you buy one?

[williamcll]

I wonder if it will push the price of current volta chips down.

[Tech Enthusiast]

21 minutes ago, Master Disaster said:

Will you buy one?

Apart from waiting for benchmarks from Steve to show up, i would much rather get a 1170 or 1180ti. 

The xx80 never appealed to me tbh. The xx70 is a lot more value for the money, while the xx80ti is a good chunk more horses for slightly more money.

 

A few years back i would have said "i will wait for a LTT benchmark first", but they are too much about not stepping on anyone's toes to be trusted anymore. Maybe trust is the wrong word, but they sugar coat results so much that it is sometimes hard to tell how the product really did.

 

Edit: Also there is a very slight chance that AMD can offer something to compete with. Latest rumors suggest that i am just a dreamer tho.

[Notional]

Interesting. But are we sure this is a GTX card? 16GB of vram for a consumer gaming card doesn't sound right. Especially in these inflated DRAM pricing times. Then again, it could be a Titan GTX card.

[porina]

So on core counts it is on parity with 1080Ti so maybe only a small difference in performance depending on where clocks end up. More ram (more cost?) but lower bandwidth... unless there is a new trick which adds value, I don't think it'll make an impact for those already at 1080Ti level.

[porina]

4 minutes ago, Notional said:

Interesting. But are we sure this is a GTX card? 16GB of vram for a consumer gaming card doesn't sound right. Especially in these inflated DRAM pricing times. Then again, it could be a Titan GTX card.

Maybe due to the memory controller it has to be a "power of two" to be effective. That would make it 8 or 16GB options. Maybe they feel 8GB is too low for the early flagship 11 series card? Especially given 1080ti and Titan Xp are at 11 and 12GB already. I suppose on that note, a split launch of a Titan at 16GB and 1180 at 8GB could make sense, with the 1180 somewhere between 1080Ti and 1080 pricing.

[WereCat]

8 minutes ago, porina said:

So on core counts it is on parity with 1080Ti so maybe only a small difference in performance depending on where clocks end up. More ram (more cost?) but lower bandwidth... unless there is a new trick which adds value, I don't think it'll make an impact for those already at 1080Ti level.

Considering that Titan V performs just about 20%-30% better than the 1080ti in games, the GTX 1180 will probably be sitting at or just above the Titan Xp performance with a lower TDP. Just my speculation, I agree that this will probably won't affect us people with 1080ti... well, at least until Ray tracing becomes a thing.

[Notional]

3 minutes ago, porina said:

Maybe due to the memory controller it has to be a "power of two" to be effective. That would make it 8 or 16GB options. Maybe they feel 8GB is too low for the early flagship 11 series card? Especially given 1080ti and Titan Xp are at 11 and 12GB already. I suppose on that note, a split launch of a Titan at 16GB and 1180 at 8GB could make sense, with the 1180 somewhere between 1080Ti and 1080 pricing.

That could definitely be the case. However, NVidia tends to save on memory on their GPU's. Think 3GB 780 vs. 4GB 290, or the 3GB 1060 vs. 4GB 580 (that no one should buy). Heck even the 3½GB 970. NVidia has improved on this matter though, like the 1080ti with 11GB. We will have to wait and see. Personally, I prefer GPU's to have a lot of Vram, so we can have high quality textures.

[TheRandomness]

Lmao

 

Read the part that says ‘This is just a placeholder’

 

@Master Disaster Just saying, but a human person manages the database.

[Tech Enthusiast]

1 minute ago, TheRandomness said:

Lmao

 

Read the part that says ‘This is just a placeholder’

Don't you dare stopping the hype train.

We have been waiting for new GPUs for ages. Let us have some test tears already!

[leadeater]

5 minutes ago, Notional said:

That could definitely be the case. However, NVidia tends to save on memory on their GPU's. Think 3GB 780 vs. 4GB 290, or the 3GB 1060 vs. 4GB 580 (that no one should buy). Heck even the 3½GB 970. NVidia has improved on this matter though, like the 1080ti with 11GB. We will have to wait and see. Personally, I prefer GPU's to have a lot of Vram, so we can have high quality textures.

1180Ti 8GB (15% less SMs) & 1180Ti 16GB cos Nvidia 

[Notional]

Just now, leadeater said:

1180Ti 8GB (15% less SMs) & 1180Ti 16GB cos Nvidia 

Well the first one would just be the 1180 non Titanium

[porina]

5 minutes ago, WereCat said:

Considering that Titan V performs just about 20%-30% better than the 1080ti in games, GTX the 1180 will probably be sitting at or just above the Titan Xp performance with a lower TDP. Just my speculation, I agree that this will probably won't affect us people with 1080ti... well, at least until Ray tracing becomes a thing.

Titan V has 5120 cores vs 3584 mentioned here. Without considering clock, that's a 43% brute force advantage. I guess we don't have much to compare, has anyone tried to work out a "gaming IPC" between Volta and Pascal?

 

Just now, Notional said:

That could definitely be the case. However, NVidia tends to save on memory on their GPU's. Think 3GB 780 vs. 4GB 290, or the 3GB 1060 vs. 4GB 580 (that no one should buy).

I think where they go 3GB/6GB is due to the way they handle memory, and have something comparable to 3-channel (or multiple thereof). 

 

Just now, Notional said:

Heck even the 3½GB 970.

As they said in the lawsuit, they offered 4GB, you got 4GB. Never mind 0.5GB was slower to access.  

 

Just now, Notional said:

NVidia has improved on this matter though, like the 1080ti with 11GB. We will have to wait and see. Personally, I prefer GPU's to have a lot of Vram, so we can have high quality textures.

Depends on game support... unless you're into heavily modding games how much do you really need?

[leadeater]

4 minutes ago, Notional said:

Well the first one would just be the 1180 non Titanium

Well there was a much bigger SM count difference this generation but I wouldn't put it past Nvidia to pull a 1060 again but on higher end cards this time around, 1070Ti was only a half slap in the face compared to that.

[Notional]

1 minute ago, porina said:

I think where they go 3GB/6GB is due to the way they handle memory, and have something comparable to 3-channel (or multiple thereof). 

 

As they said in the lawsuit, they offered 4GB, you got 4GB. Never mind 0.5GB was slower to access.  

 

Depends on game support... unless you're into heavily modding games how much do you really need?

Yeah, the memory controller dictates the amount of vram for sure.

Indeed, but they lost that lawsuit, so it doesn't really matter what they said.

 

That's what gamers said back in the 780 days, heck even the 680 days, when AMD's offerings had 1 GB of vram extra. That extra GB came in handy a lot. I will admit that over 8GB doesn't seem to make a lot of sense right now. But then again, we don't know what industry push the next gen consoles will bring. The last time, PS4 and XBone got almost 12 times the VRAM from the previous gen. That's why we saw a game like Watch Dogs, which was a true next gen game, have high res textures, and a lot more unique textures on the screen at any given time. Suddenly you needed 4GB of vram, and the 780 users got shafted as a result. VRAM is just a silly thing to run out of.

[Notional]

3 minutes ago, leadeater said:

Well there was a much bigger SM count difference this generation but I wouldn't put it past Nvidia to pull a 1060 again but on higher end cards this time around, 1070Ti was only a half slap in the face compared to that.

Indeed, but that is also partly because the 1080 and 1080ti are two vastly different chips, and not just the same chip cut down. 

[Droidbot]

1 minute ago, porina said:

Titan V has 5120 cores vs 3584 mentioned here. Without considering clock, that's a 43% brute force advantage. I guess we don't have much to compare, has anyone tried to work out a "gaming IPC" between Volta and Pascal?

 

I think where they go 3GB/6GB is due to the way they handle memory, and have something comparable to 3-channel (or multiple thereof). 

 

As they said in the lawsuit, they offered 4GB, you got 4GB. Never mind 0.5GB was slower to access.  

 

Depends on game support... unless you're into heavily modding games how much do you really need?

"IPC" is lower on Volta than Pascal. 

 

Volta is squarely a enterprise architecture. Anybody who says otherwise can be pointed to /dev/null. 

 

It's meant for the growing neural networking space, (hence the "Tensor cores") and that's where 70% of Nvidia's profit on the TitanV and Tesla V100 will come from. 

 

There will be a second architecture for consumers, and Nvidia has been heading that way for a while now. We haven't gotten top tier silicon on day one for years, and with the rise of GPGPU in the enterprise space we will never get top tier sillicon day one. 

[Humbug]

Looks great from Nvidia.

 

But I hope TSMC’s 12nm FinFET can handle the demand, of miners + gamers.

[Bcat00]

So a slight improvement, not surprising considering there's no competition at the high end.

 

since 4k is now within reach with 1 Gpu, im guessing they are going for more add in modules than raw horsepower.

[TheRandomness]

12 minutes ago, leadeater said:

1180Ti 8GB (15% less SMs) & 1180Ti 16GB cos Nvidia 

Hey am I allowed to mark my post as an important post so it can be displayed at the top?  Like, seriously, it says right at the bottom of the post.

[leadeater]

10 minutes ago, Notional said:

Indeed, but that is also partly because the 1080 and 1080ti are two vastly different chips, and not just the same chip cut down. 

Yep, but I was meaning literally another 1060 but on the 1180Ti 

[leadeater]

6 minutes ago, TheRandomness said:

Hey am I allowed to mark my post as an important post so it can be displayed at the top?  Like, seriously, it says right at the bottom of the post.

Dunno, I'm more leaning towards lock myself since this really isn't anything at all.

[Notional]

2 minutes ago, leadeater said:

Yep, but I was meaning literally another 1060 but on the 1180Ti 

Ah, well that would not surprise me. But that used to be the x80ti and Titan variants

[leadeater]

6 minutes ago, Notional said:

Ah, well that would not surprise me. But that used to be the x80ti and Titan variants

lol actually very good point, mind blanked on that. They might need to now with the Titan V how it is.

[xWood4000]

If I had the money I would definitely buy the 1180 instead of the 1080 because I want a next-gen card as soon as possible. I'm realistically waiting for the 1160 or 1170.