NVLink, is this bad?

[nothereanymore]

This is proprietary hardware, will people who want NIVIDA cards have to buy a new motherboard that supports this? If so, I might just leave these greedy fucks...

[Victorious Secret]

Fairly certain its between the GPUs and doesn't require hardware changes, but correct me on that one if its otherwise. 

 

Also, greedy fucks? Yea, fuck Nvidia for doing RD to better secure their future as a company and their future bank accounts, like a company should. Cause no other company does this kind of stuff. No. Screw commerce, am I right? 

[Frankie]

Fairly certain its between the GPUs and doesn't require hardware changes, but correct me on that one if its otherwise. 

 

Also, greedy fucks? Yea, fuck Nvidia for doing RD to better secure their future as a company and their future bank accounts, like a company should. Cause no other company does this kind of stuff. No. Screw commerce, am I right? 

 

I am reading on it right now and apparently it is and I quote:

 

NVIDIA® NVLink™ is a high-bandwidth, energy-efficient interconnect that enables ultra-fast communication between the CPU and GPU, and between GPUs. The technology allows data sharing at rates 5 to 12 times faster than the traditional PCIe Gen3 interconnect

 

So if it hit consumer market I would guess that it could replace PCIe. But that is innovation and the point about greed that @Noirgheos did is ridiculous. How the fuck would that make Nvidia greedy ?

[nothereanymore]

Fairly certain its between the GPUs and doesn't require hardware changes, but correct me on that one if its otherwise. 

 

Also, greedy fucks? Yea, fuck Nvidia for doing RD to better secure their future as a company and their future bank accounts, like a company should. Cause no other company does this kind of stuff. No. Screw commerce, am I right? 

 

I am reading on it right now and apparently it is and I quote:

 

 

So if it hit consumer market I would guess that it could replace PCIe. But that is innovation and the point about greed that @Noirgheos did is ridiculous. How the fuck that make Nvidia greedy ?

If they make it so you have to have it to use an NVIDIA card, and I don't see why it would replace PCIe. It'd be weird if AMD was using NVLink for their cards.

[Frankie]

If they make it so you have to have it to use an NVIDIA card, and I don't see why it would replace PCIe. It'd be weird if AMD was using NVLink for their cards.

 

I found this, it indeed does replace PCIe. It is logically engineered with Nvidia cards. So I don't know how would they make it work on AMD cards that have completely different architecture.

 

 

Edit: source http://www.nvidia.com/object/nvlink.html

[Guest Kloaked]

If they make it so you have to have it to use an NVIDIA card, and I don't see why it would replace PCIe. It'd be weird if AMD was using NVLink for their cards.

 

Just because I make myself a badass sammich doesn't mean I have to share it with everyone.

 

 

Probably a bad analogy. But you get my point.

[nothereanymore]

I found this, it indeed does replace PCIe. It is logically engineered with Nvidia cards. So I don't know how would they make it work on AMD cards that have completely different architecture.

 

 

Edit: source http://www.nvidia.com/object/nvlink.html

Like this, it looks great, but will it be necessary to use an NVIDIA card, and will MOBOs support this AS well as PCIe so we cant still use AMD cards on the same motherboard... we don't know yet.

[Frankie]

Like this, it looks great, but will it be necessary to use an NVIDIA card, and will MOBOs support this AS well as PCIe so we cant still use AMD cards on the same motherboard... we don't know yet.

 

We gonna probably see two versions of the same MOBO. one with PCIe and one with NVLink. That makes the most sense to me right now. Having both on one MOBO would be probably really expensive and impractical and probably bigger than E-ATX MOBOs

 

800th post hype.

[nothereanymore]

We gonna probably see two versions of the same MOBO. one with PCIe and one with NVLink. That makes the most sense to me right now. Having both on one MOBO would be probably really expensive and impractical and probably bigger than E-ATX MOBOs

 

800th post hype.

The two versions sounds nice, as long as they're the same price. And once again, a split happens. It's to be expected with Intel and AMD, but now GPUs as well? I don't like where this is going...

[Klerk]

Well Nvidia themselves used their own resources to develop so it will be really stupid for them to share the technology to their competition. And as to whether having this feature on a motherboard means we won't be able to use AMD cards will be a stupid move on both motherboard manufacturers and Nvidia and I highly doubt they will do this.

[Frankie]

The two versions sounds nice, as long as they're the same price. And once again, a split happens. It's to be expected with Intel and AMD, but now GPUs as well? I don't like where this is going...

 

Same price ? Probably not. But still, this is for super-computers and it gonna take a long time before it hits consumer market if even.

[Briggsy]

just like Gsync, the consortium/group responsible for motherboard standards are already working on a future open standard of pcie that isn't ready yet. Furthermore, the communication between gpu's will be the focal point of nvlink, because system ram on mainstream boards will still be far behind in bandwidth, rendering NVlink an enthusiast feature for future flagship Nvidia GPUs used in SLI configs. Nvidia's MO seems be an effort to mirror industry standards, but release their own proprietary version first, and make the industry look likes its copying Nvidia. 

[Dabombinable]

just like Gsync, the consortium/group responsible for motherboard standards are already working on a future open standard of pcie that isn't ready yet. Furthermore, the communication between gpu's will be the focal point of nvlink, because system ram on mainstream boards will still be far behind in bandwidth, rendering NVlink an enthusiast feature for future flagship Nvidia GPUs used in SLI configs. Nvidia's MO seems be an effort to mirror industry standards, but release their own proprietary version first, and make the industry look likes its copying Nvidia. 

I don't even understand why Nvidia needs to create a new standard, when all new motherboards are PCIe 3.0, and no GPU in existence uses all of the bandwidth of even PCIe 2.0 x16.

[Briggsy]

I don't even understand why Nvidia needs to create a new standard, when all new motherboards are PCIe 3.0, and no GPU in existence uses all of the bandwidth of even PCIe 2.0 x16.

 

Its just another "we did it first" thing.

[Frankie]

I don't even understand why Nvidia needs to create a new standard, when all new motherboards are PCIe 3.0, and no GPU in existence uses all of the bandwidth of even PCIe 2.0 x16.

 

Super-computing ?

[Dabombinable]

Super-computing ?

Possibly, but even then PCIe 3.0x16 wouldn't be saturated.

[Frankie]

Possibly, but even then PCIe 3.0x16 wouldn't be saturated.

 

You never know what they have behind the curtains.

[Dabombinable]

You never know what they have behind the curtains.

Skeletons perhaps?

[Hunter7263]

Possibly, but even then PCIe 3.0x16 wouldn't be saturated.

For gaming we dont come anywhere close but there are applications such as protein folding that have been bottlenecked by the pcie lanes since version 3 was introduced and are eagerly awaiting pcie4. Most people have no idea that that is the case. Think about it currently for games we get info from the cpu passed to the gpu the gpu does some stuff might check with the cpu again and then outputs it. In supercomputing the cpu can send a piece to a cuda core have it start and then send another piece to another core and so on. When those peices get finished they get sent back. The GPU isnt used in the same way you are used to.