3 ms·
Blows it away? 16 way vector integer processor - giving ~4 Gigaops of integer compute performance (see https://github.com/hermanhermitage/videocoreiv/wiki/Vide
by hermanhermitage 14y ago
Blows it away?
16 way vector integer processor - giving ~4 Gigaops of integer compute performance (see https://github.com/hermanhermitage/videocoreiv/wiki/VideoCore-IV-Kernels-under-Linux https://github.com/hermanhermitage/videocoreiv/wiki/VideoCor... for how to access it from user land).
24 Gigaflops via the QPU (from the 3d shader pipe) - which we have people beavering away on to open up soon.
Yes the ARM is weak on this one, but don't underestimate the power of the functional blocks and the open source and reverse engineering community.
Edit: see https://github.com/hermanhermitage/videocoreiv/wiki/VideoCore-IV-Programmers-Manual https://github.com/hermanhermitage/videocoreiv/wiki/VideoCor... for more info, or check the root wiki page. Volunteers wanted - all shapes - all sizes - all capabilities (join the mailing list, or come visit us on IRC).
- nitrogen 14y agoIn your efforts to RE the GPU, have you uncovered any hope of getting faster I/O? I'd love to use all that GPU power, but the applications I have in mind require I/O to match (at the very least, solid 150 mbits/s USB2.0 and 25mbits/s Ethernet).
- hermanhermitage 14y agoThe only options I can think of for fast IO would be exploiting the MIPI CSI and DSI interfaces. I think you can get 2 Gbps input on CSI and same on DSI for output. You'd need some fairly specialist interface chips. What would be nice in a future SoC would be say 1 Gbps Ethernet or USB3 right into the core of the beast. The current USB situation is rather disappointing - I would imagine the team would be avoiding the current IP in a hurry on future devices.