What happens when NVLink launches?

[WhiteSkyMage]

Hey guys,

 

I couldn't find any good info about all that. If they are changing the PCI-E to NVLink, we would obviously need a new MB right?

 

So in that case,I take it, it would be just one sided compatibility of MBs for Intel CPUs and Nvidia GPUs, unless those boards have a PCI-E as well...I don't see how this is going to work...With that kind of bandwidth, no game will actually challenge a GPU unless it is running 8K res...and there are not many monitors...or at least not many I know of on the market... And we are talking, NVLink combined with a GPU with HBM memory and some big chip.

 

Well I could take this article as a grain of salt: http://www.kitguru.net/components/graphic-cards/anton-shilov/nvidia-changes-roadmap-volta-is-now-due-in-2018/

but 2016 is the year of VR...and we actually might need this performance afterall...

 

 

what do you guys think?

[dizmo]

I'm quite excited for NVLink. Sure, we don't need it now, but unless someone develops something better people will never be pushed to make something that maxes it out.

I'm in the "sit and wait" boat when it comes to Pascal. I've seen the numbers, the rumors, but until we see some actual real world tests, it's all for naught.

Look at the Fury. Was supposed to be an amazingly epic card. It's not.

[Swndlr]

I'm quite excited for NVLink. Sure, we don't need it now, but unless someone develops something better people will never be pushed to make something that maxes it out.

I'm in the "sit and wait" boat when it comes to Pascal. I've seen the numbers, the rumors, but until we see some actual real world tests, it's all for naught.

Look at the Fury. Was supposed to be an amazingly epic card. It's not.

Funny thing is I actually planned on upgrading to a Fury, too bad it wasnt what I hoped it would be

[WhiteSkyMage]

Funny thing is I actually planned on upgrading to a Fury, too bad it wasnt what I hoped it would be

Yeah, I was too, even though I was expecting also bad surprices like what we got in the end...

[Stefan1024]

Hey guys,

 

I couldn't find any good info about all that. If they are changing the PCI-E to NVLink, we would obviously need a new MB right?

 

So in that case,I take it, it would be just one sided compatibility of MBs for Intel CPUs and Nvidia GPUs, unless those boards have a PCI-E as well...I don't see how this is going to work...With that kind of bandwidth, no game will actually challenge a GPU unless it is running 8K res...and there are not many monitors...or at least not many I know of on the market... And we are talking, NVLink combined with a GPU with HBM memory and some big chip.

 

Well I could take this article as a grain of salt: http://www.kitguru.net/components/graphic-cards/anton-shilov/nvidia-changes-roadmap-volta-is-now-due-in-2018/

but 2016 is the year of VR...and we actually might need this performance afterall...

 

 

what do you guys think?

 

This feature is important for GPGPU computing like in a super-computer. There the GPUs are often starving for data.

But a "nomal" gamer can't make a big use out of it....

[Glenwing]

NVLink isn't for desktops. It will be used in supercomputers. It will most likely only be featured in Tesla cards, the GeForce lineup will just use PCI Express.

[WhiteSkyMage]

NVLink isn't for desktops. It will be used in supercomputers. It will most likely only be featured in Tesla cards, the GeForce lineup will just use PCI Express.

I hope that's the case...

[manikyath]

NVLink isn't for desktops. It will be used in supercomputers. It will most likely only be featured in Tesla cards, the GeForce lineup will just use PCI Express.

i feel like this will be the most sensible answer.

 

i doubt anyone is gonna buy a new mobo specially for their nvidia cards.

[dizmo]

Funny thing is I actually planned on upgrading to a Fury, too bad it wasnt what I hoped it would be

I was hoping it'd be a powerhouse.

But, like everything that's FTM...it's not nearly what it's cracked up to be.

[Enderman]

I hope that's the case...

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

[WhiteSkyMage]

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

I am actually a Nvidia fanboy, but who would change their mobo just for a new GPU? And I do not see this happening since it would look the same as with Manttle. Nvidia will have to subsidize a lot of companies and you will have to choose real wise then. Just stick to 1 company or use the old PCIe for switching between the 2 when u need. If I have to change it I will however. I did have to switch to AMD for those bitcoins I remember...

Oh just forgot to say, AMD would be facing big challenge if Nvidia and Intel were to collaborate for that Nvilink, and I would hate to see even less competition, even if AMD stays in the end. And who knows what would be the cost of a board with NVLink....Nvidia will just slap some price and reviewers would go "it's new so it's expensive, wait for later to get it cheaper" and there will be no later just like with G-Sync is still $200, imagine your fancy NVLink mobo will be $400. Ok - i5 + NVLink mobo + high end GPU = kidney.

[DXMember]

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

I'm sorry, but what?

could you imagine paying like 300$ royalty fees to nVidia for a motherboard featuring nVLink interconnect and another 200$ for every intel CPU regardless of wether you use it or not?

Just look at the moneymilking G-Sync is!!!

People getting 3-Way 980ti SLI with surround G-Sync esentially wasting like 600$ for a feature that is designed for pesant's that can't push enough frames to cap the refresh rate of the monitor.

I'm sorry but why are these people getting a 800$ adaptive refresh rate monitor when they won't even drop below the 60 fps to take advantage of it?

How come that people that would actually benefit from feature like this - the likes of 750ti or 950 users that can only pay like 150$ for a GPU have to dish out about 4 times more for a monitor with a propriatary tech that's not even likely to be relevant for however long is the time of service for a monitor...

[KemoKa]

I was hoping it'd be a powerhouse.

But, like everything that's FTM...it's not nearly what it's cracked up to be.

The fury series was never built with the intention of being a great performer in the current APIs. They were built specifically to rock the floor in DirectX 12 and Vulkan, where they can take advantage of their Asynchronous Shaders architecture. The only reason that people are disappointed in the Fury series of cards is because, as per AMD, they're not receiving instant gratification. But the benchmarks are showing real promise.

Not trying to be a fanboy, just saying it like it is. Nvidia cards are better at DX11, but their parallel performance in games sadly leaves much to be desired.

The Fury X is delivering on its promise of 8.6 Teraflops of performance. The computational horsepower is there, it's just not being utilized properly for gaming because the APIs we still use are several years old.

[WhiteSkyMage]

The fury series was never built with the intention of being a great performer in the current APIs. They were built specifically to rock the floor in DirectX 12 and Vulkan, where they can take advantage of their Asynchronous Shaders architecture. The only reason that people are disappointed in the Fury series of cards is because, as per AMD, they're not receiving instant gratification. But the benchmarks are showing real promise.

Not trying to be a fanboy, just saying it like it is. Nvidia cards are better at DX11, but their parallel performance in games sadly leaves much to be desired.

The Fury X is delivering on its promise of 8.6 Teraflops of performance. The computational horsepower is there, it's just not being utilized properly for gaming because the APIs we still use are several years old.

AMD are thinking future that's true, even though this time it was a little too quick so this gen is considered a leap to the new but a failure at present. It will take time for devs to go AAA titles on DX12. For now, we will still see games that are on DX11. I'd say next year would be the year of DX12.

[KemoKa]

AMD are thinking future that's true, even though this time it was a little too quick so this gen is considered a leap to the new but a failure at present. It will take time for devs to go AAA titles on DX12. For now, we will still see games that are on DX11. I'd say next year would be the year of DX12.

People are always downplaying the Ashes benchmark because it's just one benchmark from an alpha version of a game with no driver support, but I think it really shows the potential for performance increases in DX12 with Async shaders. It's more impressive to me that that's what they got even with no driver support in an unstable alpha version of a game. People will say and think what they like - I think it shows real promise.

[Glenwing]

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

Bandwidth improvements won't improve the performance of applications that don't use a lot of bandwidth (i.e. games). Why would anyone want to pay extra for something that they don't need, which prevents them from using any other graphics cards in that slot, including previous generation NVIDIA cards?

[MageTank]

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

You overreacted to his post. The context in which he was speaking (He is the OP of this thread) is that he was worried NVLink would require new motherboards. He hopes that Glenwing is correct, as it would do away with the notion that he would have to buy a new board once NVLink becomes a thing. It was not that hard to understand.

 

The fact that you aggressively questioned him being a fanboy, and even mentioned (we) when complaining about him being relieved that NVLink may not be coming to normal consumers, suggests that you might be an Nvidia fanboy. After all, NVLink is proprietary to Nvidia, and the only way you would benefit from it would be if you were already set on purchasing a product that possessed that technology once it came out.

 

Now, to clarify, i am only jumping to a broad conclusion based on little information that i know about you. Much like you just did to OP. I could very well be wrong about you (it doesn't matter either way) and you could just be an average consumer that hopes to see technology advance for the sake of getting more performance for your dollar. Point is, if you don't want me jumping to this conclusion about you, do not do so to another person.

[dizmo]

The fury series was never built with the intention of being a great performer in the current APIs. They were built specifically to rock the floor in DirectX 12 and Vulkan, where they can take advantage of their Asynchronous Shaders architecture. The only reason that people are disappointed in the Fury series of cards is because, as per AMD, they're not receiving instant gratification. But the benchmarks are showing real promise.

Not trying to be a fanboy, just saying it like it is. Nvidia cards are better at DX11, but their parallel performance in games sadly leaves much to be desired.

The Fury X is delivering on its promise of 8.6 Teraflops of performance. The computational horsepower is there, it's just not being utilized properly for gaming because the APIs we still use are several years old.

I think that's one of AMD's problems. They should have released this card next year when the API's are ready, and implemented in titles.

I mean it's all great and good to say you have this awesome tech that will be coming out soon...but people look more at what's going to benefit them now, not in the future.

Plus at the rate video cards progress by the time enough games use DX12, they'll be due for their next line anyway. Just seems like a poorly thought out release.

I do kind of see why they did it. If they just rehashed again people would be outraged, and they wanted to be FTM with HBM.

[Enderman]

I am actually a Nvidia fanboy, but who would change their mobo just for a new GPU? And I do not see this happening since it would look the same as with Manttle. Nvidia will have to subsidize a lot of companies and you will have to choose real wise then. Just stick to 1 company or use the old PCIe for switching between the 2 when u need. If I have to change it I will however. I did have to switch to AMD for those bitcoins I remember...

Oh just forgot to say, AMD would be facing big challenge if Nvidia and Intel were to collaborate for that Nvilink, and I would hate to see even less competition, even if AMD stays in the end. And who knows what would be the cost of a board with NVLink....Nvidia will just slap some price and reviewers would go "it's new so it's expensive, wait for later to get it cheaper" and there will be no later just like with G-Sync is still $200, imagine your fancy NVLink mobo will be $400. Ok - i5 + NVLink mobo + high end GPU = kidney.

Ok so we should all stick to PCIe for the next 200 years of  PC computing because that way people can use backwards compatible cards, right?

Let's just completely stop innovating and improving technology for the next few centuries.

 

I'm sorry, but what?

could you imagine paying like 300$ royalty fees to nVidia for a motherboard featuring nVLink interconnect and another 200$ for every intel CPU regardless of wether you use it or not?

Just look at the moneymilking G-Sync is!!!

People getting 3-Way 980ti SLI with surround G-Sync esentially wasting like 600$ for a feature that is designed for pesant's that can't push enough frames to cap the refresh rate of the monitor.

I'm sorry but why are these people getting a 800$ adaptive refresh rate monitor when they won't even drop below the 60 fps to take advantage of it?

How come that people that would actually benefit from feature like this - the likes of 750ti or 950 users that can only pay like 150$ for a GPU have to dish out about 4 times more for a monitor with a propriatary tech that's not even likely to be relevant for however long is the time of service for a monitor...

Who told you that you need to pay $300 royalty fees and $200 more for a CPU?

Intel CPUs already cost over $200

Also you obviously don't know how gsync works, because the point of gsync is to NOT hit the maximum refresh rate of your monitor

If you're running games at 4k ultra, even 3 GPUs wont keep you over 60fps or 144fps 100% of the time

Frame drops and stuttering is what gsync reduces

 

Bandwidth improvements won't improve the performance of applications that don't use a lot of bandwidth (i.e. games). Why would anyone want to pay extra for something that they don't need, which prevents them from using any other graphics cards in that slot, including previous generation NVIDIA cards?

lower latency, higher CPU-GPU communication, possibly even using higher speed ram for vram

its all about having options and alternatives in the future

 

You overreacted to his post. The context in which he was speaking (He is the OP of this thread) is that he was worried NVLink would require new motherboards. He hopes that Glenwing is correct, as it would do away with the notion that he would have to buy a new board once NVLink becomes a thing. It was not that hard to understand.

 

The fact that you aggressively questioned him being a fanboy, and even mentioned (we) when complaining about him being relieved that NVLink may not be coming to normal consumers, suggests that you might be an Nvidia fanboy. After all, NVLink is proprietary to Nvidia, and the only way you would benefit from it would be if you were already set on purchasing a product that possessed that technology once it came out.

 

Now, to clarify, i am only jumping to a broad conclusion based on little information that i know about you. Much like you just did to OP. I could very well be wrong about you (it doesn't matter either way) and you could just be an average consumer that hopes to see technology advance for the sake of getting more performance for your dollar. Point is, if you don't want me jumping to this conclusion about you, do not do so to another person.

I don't think you realize how many times motherboards have changed expansion slots...

remember PCI?

yeah, that old slot that is no longer on motherboards? yup, its becoming obsolete.

Of course if you want backwards compatibility there are still motherboards with BOTH TYPES OF SLOTS

like seriously, motherboards are already CPU specific

if you lived back when AMD and intel used the same socket you would have probably made the same argument

look how far we've come now

it seems like you're just afraid of change, like most people here

[PlayStation 2]

so you're hoping that we dont get massive performance improvements?

why??? are you an AMD fanboy or something?

Let's be honest here: would you really buy a new motherboard JUST to use a fucking GPU?

It's not just inconvenience, it's abandoning an already established standard.

[Enderman]

Let's be honest here: would you really buy a new motherboard JUST to use a fucking GPU?

It's not just inconvenience, it's abandoning an already established standard.

well considering the thousands of people that already buy a new motherboard "JUST" to use a new CPU, then sure if it gives a significant performance improvement

 

or maybe you would prefer that intel CPUs also use FM3+ and be limited to older and worse technology?

[PlayStation 2]

well considering the thousands of people that already buy a new motherboard "JUST" to use a new CPU, then sure if it gives a significant performance improvement

 

or maybe you would prefer that intel CPUs also use FM3+ and be limited to older and worse technology?

yeah but you'd be a retarded prick if you upgrade every year to the new socket

personally i'd upgrade like every 4 years but i don't got a desktop so eh

it's not the socket, it's the features the CPU has

[Enderman]

yeah but you'd be a retarded prick if you upgrade every year to the new socket

personally i'd upgrade like every 4 years but i don't got a desktop so eh

it's not the socket, it's the features the CPU has

exactly

you dont need to immediately upgrade even if nvlink was put on motherboards...

 

maybe in a few years you want to upgrade your CPU, so when buying a new motherboard you get one with nvlink

or if youre building a PC and cant afford a nvlink GPU, you get a regular GPU and a motherboard that has both PCIe and nvlink so that you can upgrade to a new nvlink GPU in the future

 

its all about having options, just like motherboards with PCI and PCIe

nothing forces you to use one or the other, but PCI slowly goes obsolete

[PlayStation 2]

exactly

you dont need to immediately upgrade even if nvlink was put on motherboards...

 

maybe in a few years you want to upgrade your CPU, so when buying a new motherboard you get one with nvlink

or if youre building a PC and cant afford a nvlink GPU, you get a regular GPU and a motherboard that has both PCIe and nvlink so that you can upgrade to a new nvlink GPU in the future

 

its all about having options, just like motherboards with PCI and PCIe

nothing forces you to use one or the other, but PCI slowly goes obsolete

PCI is going obsolete, PCI Express is here to stay for at least another half-a decade.

SSDs can't go that fast nor can GPUs take advantage of even an 8x bus.

[Enderman]

PCI is going obsolete, PCI Express is here to stay for at least another half-a decade.

SSDs can't go that fast nor can GPUs take advantage of even an 8x bus.

yeah i know

 

i never said nvlink was going to take over in 2 years, thats ridiculous

 

but its likely that we will be using that (or something even better) in 10 or 20 years when PCIe does go obsolete