Pascal HBM compatibility with current MXM

Simple question. Any speculations or knowledge if Pascal with NVLink and HBM2 will be compatible with the MXM slot my current Clevo uses? Or do I need to upgrade this entire laptop which would then be cause to say ciao to mobile gaming and regain some sanity with desktop gaming?

 

It seems NVLink is tech to replace PCI/e, so it would make MXM obsolete... but it also is only meant for Tesla-class cards.

HBM2 according to meaker could work if it was rotated a bit, so we could expect that.

The real issue is weirdly-sized cards, like the low-power 980 will be shaped like.

 

Just pretend NVLink doesn't exist for now, because it's very much a server oriented feature. Unless nVidia can somehow convince PCI-SIG that NVLink should replace PCIe as the consumer standard (never gonna happen), you won't ever see NVLink come to consumer boards.

 

So is there something like NVLink for consumer use in the works? The concept seems to me match made in heaven for gaming. No more CPU bottleneck and faster communication between GPUs also? Wouldn't this make CPU power and SLI gaming heaven? Resolve so many issues that AMD fail and others trying to overcome with measures like Mantle and finally eliminating issues with SLI? Or is my understanding of NVLink completely off?

Well at least I can still hope that a Pascal HBM2 may fit in the current MXM.

 

Most likely, yes. I'm sure NVIDIA has something planned. Combined with DX12, games should perform much better.

There's a lot of posts about NV-Link in the other Pascal thread.

 

NVLink is not removing CPU bottlenecks for cards that would have affected any gamers in the first place.
Faster GPU communication doesn't mean a single thing for us in this instance, as it needs to be paired with XDMA-style designs so the memory data can get across enough. And as-is, PCI/e 3.0 does 32GB/s (SLI bridge is just under 1GB/s), and is sufficient for all SFR-type technologies that would be useful for the cards. Bumping the bandwidth to 64GB/s but leaving the card designs the same is pointless; it'll never transfer enough data if it couldn't do it on a 32GB interface already.
It still doesn't make SLI "better". It just means we can use SLI differently in newer games that are demanding on memory bandwidth and vRAM buffer size... and again, the cards need to be ready for it. Guess who's already ready for it and have been for about 2 years? AMD.
We don't need Mantle or DX12 or Vulkan to fix multiGPU problems. But it'll help, since devs will just be able to fix their own damn code and make things run better.

Your understanding of NVLink is entirely off =D

 

Actually AMD's CrossFire issues are mainly on the software side. The bridgeless XDMA CrossFire design when it works actually works better than SLI. The bottleneck isn't communication between the GPU and CPU, it's between the GPU themselves. And right now that stupid SLI bridge that everybody loves so much and is willing to pay big bucks for more bling is actually what's holding up the GPUs.

lol ninja'd

 

me

 

Interesting.

 

For now we may see nvlink replace the sli bridge.

 

They'll probably have their own version of XDMA CrossFire that I certainly believe, but I don't think NVLink in its original form will show up on consumer parts anytime soon, especially considering it's supposed to be a mezzanine card that snaps into a server board and replaces the PCIe slots.