NVIDIA GeForce GTX 1180 PCB: more power, new SLI finger

[Rune]

1 hour ago, TVwazhere said:

Tonight on LTT Smackdown Vs Raw: Leadeater Vs Swatson.

 

But really, I'll be happy when these are out regardless. My 1080 is starting to show its age and can't even properly push 1440p max settings at 144hz anymore. Maybe I am in need of a full system upgrade. Hm...

[TVwazhere]

Just now, Rune said:

I wanted to do something like this, but I'm at work  

[S w a t s o n]

1 hour ago, Rune said:

 

But really, I'll be happy when these are out regardless. My 1080 is starting to show its age and can't even properly push 1440p max settings at 144hz anymore. Maybe I am in need of a full system upgrade. Hm...

I take offense to being the one getting slammed but he's the mod so I expect it :^ )

[bowrilla]

9 hours ago, Tedny said:

looks like GDDR6 will 2GB for one chip

I'm afraid that it's still just 8Gb / 1GB. Neither SkHynix nor Micron list anything bigger than 8Gb GDDR6 modules and it's not like it's something terribly new at this point. While possible I wouldn't count on NVIDIA just having a deal with both suppliers for them to not go public at this point with a 10/12/14/16 Gb version of their GDDR6 memory chips.

 

If that is actually a 1180/2080 PCB than that card will end up with 8GB of VRAM. Which is a hefty bummer!

[Nowak]

Fake news. This is a workstation PCB.

[bowrilla]

19 minutes ago, NowakVulpix said:

Fake news. This is a workstation PCB.

 

13 minutes ago, VegetableStu said:

current NVLink connectors aren't oriented like that. you could argue for both ._.

And Buildzoid mentioned that NVLink has less pins I think.

[bowrilla]

26 minutes ago, Tedny said:

It will be 4k GPU, 8GB will kill it like a selling product. + this PCB looks more like 1160,but that 6-pin and 8-pin  weird for just Middle range card. So, Maybe ram manufacture just don't tell us what can GDDDR6 can be.... yea... you know, those guys.... 

It looks like a beefy 8 phase VRM design, it got a new generation SLI header, 6+8 pin power design … nope, I'm afraid this is a high end PCB. It obviously aims at the 1080 rather than the 1080ti. GDDR6 might be faster but still. I'm not sure it will be such a great competion against the 1080ti. I'm not expecting more than 10% better performance. The PCB doesn't even have space considering all the traces for another 4 VRAM chips. Now considering SkHynix or Micron actually have 16Gb chips on hand – that'd mean this PCB would come with a whopping 16GB of VRAM. Compare this to the current flag ship and it's very very very much unlikely. GDDR6 is still new, they're not even maxing out the bandwith yet. There might be a 12/14/16 GB version coming but that'll be the ti version.

[Brehohn]

I love bare PCB's! They are really cool to look at before they have all of the capacitors and such on them.

That has to be NVLink or some form of it. My thought is, maybe they are actually trying to make SLI better and more consistent? SLI has a max bandwidth of 2GB/s on the High-Bandwidth Bridge. NVLink has a maximum bandwidth rate of around 35-40GB/s. With that being said, putting NVLink onto Geforce cards makes sense because people are now trying to push multiple 4K monitors at high refresh rates.

The cost though, they might make it a dumbed down version of the Quadro NVLink to where it will support what we need just fine, but not be overkill and expensive.

Hopefully what I'm thinking makes sense.

[leadeater]

1 hour ago, Brehohn said:

NVLink has a maximum bandwidth rate of around 35-40GB/s

NVLink is actually a multiple lane connector so can go much higher than that. NVLink 2.0 is 25GB/s per Sub Link and has 6 so will do 150GB/s per direction, though Tesla cards have two NVLink connectors so I don't know how that effects things if this card is based off that with a single connector. Maybe it'll have to be 150GB/s split in half for in and out.

 

Not that it matters much here but the number of devices dictates the number of sublinks you can combine to each device.

Quote

According to depictions in Nvidia's blog based publications from 2014 NVLink allows bundling of individual links for increased point to point performance so that for example a design with two P100s and all links established between the two units would allow the full NVLink bandwidth of 80 GB/s between them.

https://en.wikipedia.org/wiki/NVLink

NVLink 1.0 is 20GB/s 4 Sub Link which is where the 80GB/s comes from.

 

Really cool connection technology just not sure how applicable it is to a gaming GPU/SLI.

[schwellmo92]

7 hours ago, Swatson said:

I take offense to being the one getting slammed but he's the mod so I expect it :^ )

Leadeater is right, historically SLI scaling has been pretty but it has gotten extremely poor in the last couple of years, here’s an article showing that https://babeltechreviews.com/the-50-game-gtx-1070-ti-sli-review/3/

[jasonvp]

18 minutes ago, schwellmo92 said:

Leadeater is right, historically SLI scaling has been pretty but it has gotten extremely poor in the last couple of years, here’s an article showing that https://babeltechreviews.com/the-50-game-gtx-1070-ti-sli-review/3/

As I'm fond of reminding people when this "SLI is dead" thing comes up: it depends entirely on which games you're playing.  The games I play all scale fairly well across an SLI'd pair of GPUs.  Dice's Frostbite engine, for instance, supports SLI very well.  Ubi's AnvilNext (eg: Rainbow Six Siege) does, too.  It hasn't always, but it does now.  Ubi's Dunia (FC5) does, superbly.  One of the best I've seen, actually.

 

Some big-named AAA devs are still supporting it.  Some don't.

 

 

[Brehohn]

3 hours ago, leadeater said:

NVLink is actually a multiple lane connector so can go much higher than that. NVLink 2.0 is 25GB/s per Sub Link and has 6 so will do 150GB/s per direction, though Tesla cards have two NVLink connectors so I don't know how that effects things if this card is based off that with a single connector. Maybe it'll have to be 150GB/s split in half for in and out.

 

Not that it matters much here but the number of devices dictates the number of sublinks you can combine to each device.

https://en.wikipedia.org/wiki/NVLink

NVLink 1.0 is 20GB/s 4 Sub Link which is where the 80GB/s comes from.

 

Really cool connection technology just not sure how applicable it is to a gaming GPU/SLI.

Good to know. However, wouldn't this still be useful by giving SLI more bandwidth to have a higher pixel clock? With how affordable 4K gaming is becoming, and now with 120hz displays, we are going to need something much faster than High-Bandwidth bridges, correct?

[vanished]

11 hours ago, NowakVulpix said:

Fake news. This is a workstation PCB.

I really doubt that.  I could be wrong but here's why I say it:

• There's not enough memory.  Look at something like the P6000, or any of the quadros compared to their GeForce counterpart.  The have way more VRAM than the consumer cards at the same tier of processing power, and yet if we believe the assertion that GDDR6 will be 1 GB per chip, that means this is only an 8 GB card, or 16 if they put more on the back.  That's barely enough to be the 1180/2080, never mind a quadro equivalent
• Why is the NVlink connector (or whatever it is) mirrored?  If this was a workstation card, shouldn't it have the normal connector we've already seen on that kind of card?  Why change it?
• It's too early.  We know nvidia's release cycle is to come out with the *80, and then later the lower things like *70, and the higher stuff like *80 Ti, Titans, etc. and the Quadros would be a part of that second wave.

[TheLaserCucumber]

21 hours ago, YongKang said:

The new SLI finger should be a key interest here

That is not a SLI finger, that is a NVLink connector.

[leadeater]

5 hours ago, Brehohn said:

Good to know. However, wouldn't this still be useful by giving SLI more bandwidth to have a higher pixel clock? With how affordable 4K gaming is becoming, and now with 120hz displays, we are going to need something much faster than High-Bandwidth bridges, correct?

Sort of, not a lot of information is passed between GPUs in SLI so the required bandwidth is low(ish), however that is current generation SLI. GPUs render each frame interdependently of each other and pass the resulting frame back to the master GPU to monitor output, those final render frames aren't actually a lot of data. Without looking in to it the newer High Bandwidth SLI Bridge was probably required for 4K 120Hz-160Hz (something like that) since it's still a fair amount of bandwidth at that frame size and refresh rate, no where close to memory bandwidth though.

 

A bi-directional 150GB/s interconnect on the other hand is getting close to GPU memory bandwidths, still only half way though.

[Valentyn]

Time to upgrade from my Titan V? It’s struggling in WoW at 6880x2880 since they added DX12z

 

Gimme that 50% increase NGreedia.

[Syntaxvgm]

This is either a tesla or quadro.. or we're getting some form of NVLINK like thing to replace SLI, which I doubt. But if it were the case I'd think this would mean AFR or split rendering would be left to DX12/vulcan whatever and this would be a different type of thing altogether, maybe with shared vram

Doubt it though. The only thing I see that you could make an argument for this being a consumer card is the 2 fan headers. I don't think we've had dual fan workstation cards yet, all have been single blower? Edit: I see others have pointed out the reverse connector. 

 

Also this looks like it has full size dp and a mini-dp connector. What's with that? Maybe that's usbc?