** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Orangey]

That's the bandwidth between VRAM and GPU core.

 

[shadow_boxer]

Really torn between pulling the trigger on a 980Ti or waiting for pascal. Still on 780 but it's starting to show and fancy getting a couple of new titles.

 

[Smrtka]

Really torn between pulling the trigger on a 980Ti or waiting for pascal. Still on 780 but it's starting to show and fancy getting a couple of new titles.

I am on a same boat. However I have read some articles that the Pascal card because of their new 3d memory or whatever will not use PCI-e, meaning I will need a new system

 

[hominid]

I am on a same boat. However I have read some articles that the Pascal card because of their new 3d memory or whatever will not use PCI-e, meaning I will need a new system

I would be very surprised if they did that!

 

[ukxenon]

I am on a same boat. However I have read some articles that the Pascal card because of their new 3d memory or whatever will not use PCI-e, meaning I will need a new system

just another excuse to upgrade 2016 will be a good year {i hope}

 

[Perfect_Chaos]

I am on a same boat. However I have read some articles that the Pascal card because of their new 3d memory or whatever will not use PCI-e, meaning I will need a new system

You must be talking about NVLink, that's not meant for us.. PCIE will still be used.

 

[Orangey]

Nvidia will use PCIe 4 when it comes out late 2016 / early 2017.

NVLink is going to be used in supercomputers only and the only one I have heard of using it is the Oak Ridge coming online 2018. So a little dishonest to push it as a feature of Pascal...

And regarding the "3D memory" it is their sly way of saying HBM, which AMD already did.

 

[Kaapstad]

Will 4GB cards be done away with seeing as lots of VRAM is the requirement even at 1080p?. 8GB be the minimum?, I doubt I would even buy a 6GB card now?. I know the top end pascal GPU will surface first? but what comes after that and how much memory will those cards have I wonder?. Why has VRAM become the focus of todays pc video games? some games textures don't look that great imho.

I think it will be small Pascal first which means performance about 15% better than the current GM200 cards.

Small Pascal will come with up to 8gb VRAM which means that Maxwell Titan X will still be 2160p king in SLI.

Big Pascal when it lands will be a beast with performance about 180% of current GM200 cards and the Titan versions will pack 16gb of VRAM.

 

[Dicehunter]

You know if a 1080 pascal comes you'll buy. The Ti won't be until late summer or autumn at the earliest.

I just sold my Titan X last night. (On Intel Graphics atm ) I'll be going for 1080 pascal. That should be a big enough jump from Titan X.

Nope I'm not doing another 780 Ti to 980 situation.

I'll keep hold of the 980 Ti until Nvidia release full fat big daddy Pascal, I'm in no rush to get rid of the 980 Ti

 

[nox_uk]

kinda hoping someone will leak some performance numbers to get rough idea how much faster these might be. I usually get hyped up by new releases/upgrades and then realise that they didn't make that much difference in the big scheme of things :/


nox

 

[drunkenmaster]

I am on a same boat. However I have read some articles that the Pascal card because of their new 3d memory or whatever will not use PCI-e, meaning I will need a new system

It might be used in the same way, the way some quad sli boards add an extra pci-e switch to 'add' lanes between the chipset and pci-e slots. These don't increase gpu to cpu speed, just gpu to other gpu speed and is nearly useless for gaming. They don't send much data across the pci-e bus to sync in sli/xfire so doubling bandwidth does nearly nothing. Really it just means that it's easier to give each gpu equal bandwidth to the cpu rather than one slot getting 8x and another getting 4x or something.

The main reason for NVlink is to increase GPU to GPU talking in completely custom compute systems with more than 4 gpus all being used for compute where gpu to gpu talking can be way higher than simple synchronisation done in sli/xfire.

Nvlink will absolutely have zero benefit in gaming. It's built on top of pci-e anyway and IBM already had an effective NVLink, just without Nvidia involved which was again built on top of pci-e, this is gen 2(in effect) and renamed with Nvidia involvement. The gen 1 version had zero impact on anything but IBM systems.

Either way for PC systems pci-e 4 comes out in a year or two which comes pretty close to matching NVLink making the benefits even smaller. IIRC Nvlink is pretty much 80GB/s and pci-e 4 bumps connection speed to 64GB/s. Again the real benefits come when a system is held back dramatically at 32GB/s which gaming is absolutely not, compute can be.

Absolutely, without question, Pascal will launch as pci-e 3 cards(maybe supporting pci-e 4 but not likely). They will not release desktop cards that only work on NVLink, literally no chance at all of that.

As someone else mentioned, 3d memory has no bearing on this anyway. Nvidia says 3d mem because they don't want to say HBM created by AMD.

 

[drunkenmaster]

http://wccftech.com/nvidia-pascal-volta-gpus-sc15/

NVIDIA didn’t reveal they name of the Semiconductor Foundry but it was confirmed that TSMC would be supplying the new GPUs. Now this might not be a significant bit as it has been known for months and we know that NVIDIA’s Pascal GP100 chip has already been taped out on TSMC’s 16nm FinFET process. This means that we can see a launch of these chips as early as 1H of 2016.

How on earth can this website be taken seriously by anyone. Nvidia didn't reveal the foundry name but confirmed it was being made at TSMC.... what a insane sentence.

 

[MooMoo444]

Maybe NVIDIA themselves did not reveal, but it was confirmed by other sources.

 

[Orangey]

What they mean is "Nvidia were named as one of the launch partners of 16nm".

In the wccf world this is what passes for confirmation.

 

[TNA]

How on earth can this website be taken seriously by anyone. Nvidia didn't reveal the foundry name but confirmed it was being made at TSMC.... what a insane sentence.

Haha! I thought the exact same when I read that. I was like O.o

Then I read it again and was like o.O

lol

 

[Smrtka]

It might be used in the same way, the way some quad sli boards add an extra pci-e switch to 'add' lanes between the chipset and pci-e slots. These don't increase gpu to cpu speed, just gpu to other gpu speed and is nearly useless for gaming. They don't send much data across the pci-e bus to sync in sli/xfire so doubling bandwidth does nearly nothing. Really it just means that it's easier to give each gpu equal bandwidth to the cpu rather than one slot getting 8x and another getting 4x or something.

The main reason for NVlink is to increase GPU to GPU talking in completely custom compute systems with more than 4 gpus all being used for compute where gpu to gpu talking can be way higher than simple synchronisation done in sli/xfire.

Nvlink will absolutely have zero benefit in gaming. It's built on top of pci-e anyway and IBM already had an effective NVLink, just without Nvidia involved which was again built on top of pci-e, this is gen 2(in effect) and renamed with Nvidia involvement. The gen 1 version had zero impact on anything but IBM systems.

Either way for PC systems pci-e 4 comes out in a year or two which comes pretty close to matching NVLink making the benefits even smaller. IIRC Nvlink is pretty much 80GB/s and pci-e 4 bumps connection speed to 64GB/s. Again the real benefits come when a system is held back dramatically at 32GB/s which gaming is absolutely not, compute can be.

Absolutely, without question, Pascal will launch as pci-e 3 cards(maybe supporting pci-e 4 but not likely). They will not release desktop cards that only work on NVLink, literally no chance at all of that.

As someone else mentioned, 3d memory has no bearing on this anyway. Nvidia says 3d mem because they don't want to say HBM created by AMD.

Thanks for clarification on the PCIe bit, no the questions is worth waiting for this or get 980TI?

 

[Seanspeed]

How on earth can this website be taken seriously by anyone. Nvidia didn't reveal the foundry name but confirmed it was being made at TSMC.... what a insane sentence.

I'm guessing they mean that while Nvidia didn't say anything themselves, the site found another source to confirm where it was being sourced from.

Thanks for clarification on the PCIe bit, no the questions is worth waiting for this or get 980TI?

I cant think of any good reason to not wait at this point unless you desperately need a top line graphics card for whatever reason.

Buy a 980Ti today and in 6 months there's going to be something that costs like 50% less, performs similarly and probably has new features/advancements on top of that. If you can live with that and not feel any regret, go ahead.

 

[TNA]

Small Pascal will come with up to 8gb VRAM which means that Maxwell Titan X will still be 2160p king in SLI.

Yeah, when it works that is

Thanks for clarification on the PCIe bit, no the questions is worth waiting for this or get 980TI?

Definitely wait. Not long now and you get new faster tech for cheaper

 

[Orangey]

I'm guessing they mean that while Nvidia didn't say anything themselves, the site found another source to confirm where it was being sourced from.

They don't have sources.

 

[drunkenmaster]

Thanks for clarification on the PCIe bit, no the questions is worth waiting for this or get 980TI?

My take is, I'd buy a Fury over a 980ti if I was going to buy this generation rather than wait for a 16nm architecture designed more specifically around HBM, higher bandwidth and DX12 features. I also didn't bother buying a Fury

I still might if some great deal comes up on one somewhere in the next couple months but I really don't see it being too worthwhile. 16nm/designed specifically for DX12 architectures will be so much better over the next 2-3 years than this gen working on what is mostly DX12 games from next year onwards.