Wireless PCIe To Enable Remote Graphics Cards

J. Dzhugashvili writes "If you read Slashdot, odds are you already know about WiGig and the 7Gbps wireless networking it promises. The people at Atheros and Wilocity are now working on an interesting application for the spec: wireless PCI Express. In a nutshell, wPCIe enables a PCI Express switch with local and remote components linked by a 60GHz connection. The first applications, which will start sampling next year, will let you connect your laptop to a base station with all kinds of storage controllers, networking controllers, and yes, an external graphics processor. wPCIe works transparently to the operating system, which only sees additional devices connected over PCI Express. And as icing on the cake, wPCie controllers will let you connect to standard Wi-Fi networks, too."

Comments

Re:I must admit...

[JesseMcDonald]

Some recent systems have IOMMUs which provide privilege separate between hardware devices much like normal MMUs govern software. However, unless this sort of IOMMU device is active, PCI and PCIe hardware is generally capable of transferring data to or from any other connected device, including any area of system RAM. Sometime this can even extend to external interfaces; for example, people have been known to take advantage of the DMA capabilities of the Firewire protocol to read the contents of RAM on an active system.

In general, non-hotpluggable hardware has been granted the same level of trust as the OS kernel, so no one worried very much about it. IOMMUs were more about protecting against faulty or corrupted software (device drivers) than malicious hardware. However, more and more hardware is hotpluggable these days. Also, some software interfaces are becoming too complex to really trust—consider, for example, the interface to a modern GPU, which must transfer data to and from RAM, and perhaps other GPUs, under the control of code provided by user-level application software (shaders, GPGPU). Without an IOMMU it is up to the driver software to prove that such code is perfectly safe, which is an inherently hard problem.