Recent content by Tim Murray

I'm assuming PLLs, as it's certainly not any sort of software DVFS cost. Experiment is really simple: get the system into an unloaded state, switch to userspace governor, lock a CPU benchmark (dhrystone or something absurdly simple) onto a single core, run a script on another core that...

sure, I should say that's what I measured on 5X/6P (but I seem to recall it's also true on Krait for CPUs).

changing clocks costs a lot of power, whether it's CPU or GPU. changing them every 10-20ms is catastrophic for power. source: my own experiments

Okay, so you're a reviewer, you want to know what an Android device is doing. You think, oh, I'll use random tool X that runs on the device and see what it says about the device's clockspeeds at various points in time. No, stop. Just stop. You're wasting your time, you'll be wasting my time...

if it's still going through dxgk, then Mantle seems like a big waste of time--the kernel thunking overhead and allocation lists will still kill perf. guessing they're working around it somehow, but carving out memory and suballocating was (last I checked, which was admittedly 15 months ago) not...

Why I left: lots of reasons, but the main one is that at this point I don't think HPC is where I want to spend the majority of my career. It was an unbelievable first job to have in terms of advancement and learning opportunities, but it was time to move on. Working on the Android RenderScript...

drive-by to say that although I left NV in July, I was the CUDA SW lead for Titan for a long time (over a year). B3D taught me well :) (this is the second #1 machine I was heavily involved with, I worked on Tianhe-1A as well) also, it should be noted that it's hard to get efficiency close to...

what CUDA failures are you referring to specifically?

How long until eezing reregisters and complains about moderators that ban him?

Friend works at Google, so I played with a N1 for a while a few weeks back. Hardware's good, but the UI remains (unsurprisingly) really messy. Some menus are buttons, some are sliders, some options are at the bottom of a page, some are at the top, some are buried under the menu button. It has a...

Er, it has no impact on PCIe bandwidth because it's not spinning across PCIe.

I don't know if blocking sync is implemented for OpenCL on NV hardware. The "fastest" way to wait on GPU event completion is to spinlock on a memory location, but that obviously pegs a CPU core. Blocking sync allows the thread to go to sleep and be awakened by the driver later. There can be...

The day when 32-bit OSes die altogether will be a very happy day for me.

Fade to Black was nothing compared to Flashback. At least, when I was 10 or so.

aw man, nobody thanks me for answering questions :(