What will nvlink bring to the consumer platform? And nvlink vs PCIe 4.0

[idieorudie]

Nvlink is supposed to be useful in servers and workstations but what will it bring to the consumer market?

Also nvlink vs PCIe 4.0?

What will be the difference?

[Hatsune Miku 「 」]

nothing. It supposed to speed up gpu and cpu data transfer for super computing. Not like such an expensive technology is being added for consumers who don't need it.

Why do people always assume that once something is for enterprise and servers that it'll have an application in consumer pcs.

[AnthonyPaull]

Nvlink is supposed to be useful in servers and workstations but what will it bring to the consumer market?

Also nvlink vs PCIe 4.0?

What will be the difference?

 

Another shitty proprietary platform.

 

Amd and Nvidia are having a Dick waving contest, and AMD are kinda just casually standing there doing it with one hand like 'Yeah, we'll release some shit that's like yours but cheaper" and Nvidia are like "FUCK YOUUUUU GONNA KILL AMD N SHIT FUCKING GSYNC GIVE US YOUR MONEYYYYY!!!!!"

 

Basically, Nvidia re becoming the Apple of Graphics cards. Steer clear of their shit.

[idieorudie]

What about PCIe 4.0?

Will it be beneficial?

[Sauron]

What about PCIe 4.0?

Will it be beneficial?

 

Right now even pcie 3.0 is unnecessary above 4x (although nvidia obviously has to lock slifor no reason...).

[idieorudie]

Existing enthusiast GPUs can hardly saturate PCIe 2.0 let alone 3.0

So Hbm is useless?

[idieorudie]

Right now even pcie 3.0 is unnecessary above 4x (although nvidia obviously has to lock slifor no reason...).

Umm so Hbm will not increase bandwidth?

[idieorudie]

The benefit of HBM is to grant the GPU chip faster access to the graphics memory (VRAM) modules. It won't affect the data rate between the card and other components.

Oh so Hbm is used for fast data transfer between gpu and vram! Thanks!

I thought it affects the date transfer rate of gpu on the pcie bus

[Sauron]

Umm so Hbm will not increase bandwidth?

 

It will increase the bandwidth between the vram and the gpu, not between the vram and the rest of the system (or not as much). Maybe it will start needing pcie 3.0 8x to be optimal, but we're still a long way from needing anything more. Bare in mind pcie 3.0 16x would have 4 times the bandwidth that is useful right now.

[idieorudie]

It will increase the bandwidth between the vram and the gpu, not between the vram and the rest of the system (or not as much). Maybe it will start needing pcie 3.0 8x to be optimal, but we're still a long way from needing anything more. Bare in mind pcie 3.0 16x would have 4 times the bandwidth that is useful right now.

Ohh ok thanks!

[Pauly.B]

Eventually a version of nvlink may merge over to the consumer market (post volta gpu's perhaps?)

[WAStanley734]

The reason for this is multi gpu interconnectivity. The general person with 1 gpu will never use it but those who sli their systems need the added bandwidth, especially if dx12-1? continue to get better at multithreading. If you were to do the calcution to fully use the current high end gpu in 4x sli you would exceed the 16 gb/s on 3x. And since they are going all the way to 80gt/s then it'll be able to be a standard for a few years. Amd won't be able to touch this interface for multi setups, if it works like they claim. We also don't know the transfer rates for the hmb and hmb2 gpu boards. Could significantly be increasing. If you drastically reduced the time it takes to access the memory, ie ram disk, the throughput is increased drastically. So is it for consumers? Yes. If you want it. We are able to purchase server boards, cpus, and ram, which are significantly more stable. This is a step in the direction of say 4x titan z's used to render a game(8 gpus). Now we are getting to the point where does the cpu have enough lanes to handle this!? Which is where as a gpu manufacturer they want to be, "oh your cpus can't keep up with our multi gpu setup!?" "Oops." Lol.