Is USB RAM or a USB processor a feasible concept? [closed]

It's possible to put useful special-purpose processing power out at the far side of the USB bus, but it's not very practical for a general-purpose processor. So it's not like you're going to be able to upgrade your slow Pentium M by popping a Core i7 USB dongle into the side of your machine.

A real-world example of useful special-purpose processing power over USB are those USB dongles that contain dedicated H.264 encoder hardware to accelerate re-encoding video files. The Elgato turbo.264 HD is one such device.

FireWire could conceivably be a better choice for trying to add external general-purpose processing power to a system. FireWire doesn't have USB's restrictive Host vs. Device dimporphism; FireWire devices are all peers on a network. FireWire devices can do DMA transfers to each other, so you could conceivably craft something of a NUMA system via FireWire links. Intel's upcoming Light Peak optical interconnect is another possibility.

It's possible, but would be so slow to be useless.

Windows Vista introduced a feature called ReadyBoost which uses a USB stick sort of as a RAM extension. Since USB is so slow compared to RAM, it would use the USB stick as a faster alternative to the hard disk's pagefile.

A general-purpose processor connected over USB would be so slow as to be pretty useless except in perhaps a very specialized situation.

• For comparison, the peak bandwidth of USB2 is 480 Mbit/s, whereas the peak bandwidth for DDR2-5300 is 5333 Mbyte/s, which is 88 times faster.

• Flash memory's write and read speeds doesn't come close to USB2's maximum throughput.

• The latency of DDR2-5300 is around 6ns, while the latency of USB flash drives is on the order of 1-10ms, which is hundreds of thousands times slower.

USB 2.0 has a maximum bandwidth of 480Mbps. DDR3 1600 has a bandwidth of 12800MBps. Notice the b vs B.

This is certainly not plausible. CPUs would be even more bandwidth limited.