Originally Posted by Squeek
For compute power it's great but i'm not sure how great it will be for games in comparison to a dedicated API like mantle.
But I'm pretty uneducated on the subject
For this to reach consumer X86 space, we would need to see AMD and Intel adopt it.
AMD is Nvidia's competitor
Intel ran Nvidia out of their platforms by rejecting their chipsets entirely.
The only hope this has is in HPC. It is more than likely specialized only for PowerPC and IBM since they are working with Nvidia now.
Nvidia is unable to bring this to consumers and it will remain an HPC only type of thing. At best they will be able to use it to make two GPUs talk over existing PCIe, but it won't be communicating with system memory or anything without going over PCIe. Which means it won't ever be able to bring consumers benefits that APUs can bring, and instead will only accelerate pure GPU workloads unless Nvidia can squeeze their way into a platform, someone else's CPU, or make their own ARM CPU.Edit for clarity: Nvidia NvidiaLink requires the CPU or Chipset to cooperate with sharing system memory. Nvidia doesn't have the ability to create their own CPU outside of ARM. Intel and AMD can use x86 to bring this to gamers and such. Nvidia can do Shield type devices or desktop ARM systems for playing Android games at best. In a consumer system where Nvidia can't get access to chipsets and CPUs to put their NVLink tech in, they can at best make multiple GPUs talk to each other really fast and share memory with the GPU only.
For CPU sharing, Nvidia will have to work with someone else who will cooperate, which only so far seems to be IBM. But Nvidia is completely screwed because they have to basically depend on someone else to make an HSA style platform and they have no control over the CPU. They also don't use the most common platform that consumers of this kind of technology would like. SO you will only really see NVLink be used to full potential where they can access the CPU and chipset design and throw their ideas into it.
Consumers won't get the HSA stuff unless they want to abandon x86.
But none of those options are really appealing for most people, unless you have specific CUDA applications that you need to see rapidly accelerated.
I will be floored if this leaves a niche of HPC. Nvidia has been kicking and moaning a lot lately and they're doing everything they can to put lipstick on a pig, from 3fps increase with skewed bar graphs that don't start at zero to comparing framerates in games that don't even display the same things on the screen depending on if you're using Mantle or DX.
I'm just trying to save a lot of you from disappointment, because Nvidia is by far in the worst position when it comes to gaming and this homogeneous computing thing. Intel GPU sucks but they can at least use x86 and APU style acceleration. Intel and AMD are moving on from just making big/fastest or power efficient GPUs and Nvidia is doing everything they can to keep up, but they can't do anything because they can't make x86 CPUs and AMD and Intel don't want anything to do with their chipsets. Most chipsets don't even do that much anymore because a lot of what they do has been moved on die to the CPU, because "MUH SOC!"Edited by sdlvx - 3/25/14 at 9:03pm