fxkamd

Thank you to tiny corp for pointing out some problems running BERT training with Tinygrad on AMD GPUs in this Tweet. We had a few engineers at AMD take a look at the problem and they were quickly able to reproduce it.

What they found was an issue related to CWSR (compute wave save restore), which is a mechanism that allows our driver and firmware to preempt and reschedule long-running compute waves on our GPUs. The GFXv11 GPU line requires a workaround to set COMPUTE_PGM_RSRC1.PRIV=1 when dispatching a compute kernel. Normally this is handled by the AQL DISPATCH packet. However, since the Tinygrad implementation leverages a custom runtime, it requires this workaround in its PM4-based dispatch. This patch is specific to GFXv11 GPUs. Other GPUs do not require it and should not use this workaround. The following KFDTest patch can be used as a reference: https://github.com/ROCm/ROCT-Thunk-Interface/commit/507637ed5b82197eecbf483cdc1234939766549a

While inv

fxkamd

This is Felix Kuehling, long time KFD driver architect. I started looking into the TinyGrad source code yesterday, focusing on ops_kfd.py, ops_hsa.py and driver/hsa.py, to understand how TinyGrad talks to our HW and help with the ongoing debugging effort from the top down. This analysis is based on this commit: https://github.com/tinygrad/tinygrad/tree/3de855ea50d72238deac14fc05cda2a611497778

I'm intrigued by the use of Python for low-level programming. I think I can learn something from your use of ctypes and clang2py for fast prototyping and test development. I want to share some observations based on my initial review.

ops_kfd looks pretty new, and I see many problems with it based on my long experience working on KFD. I think it's interesting, but probably not relevant for the most pressing problems at hand, so I'll cover that last.

ops_hsa uses ROCr APIs to manage GPU memory, create a user mode AQL queue for GPU kernel dispatch, async SDMA copies, and signal-based synchronization with barrier packets

Mnkai

TDP and Turbo Parameter Modification with MSR on Non-Overclockable CPUs

Disclaimer

• Modifying MSR may void your CPU's (or system board's) warranty. Proceed with caution. I am not responsible for any damage caused by this article.
• MSR addresses vary significantly between CPUs. Check your CPU's MSR address using Intel's documentation.
• This has only been tested on the Intel i7-8550U (Kaby Lake R).
• This article is a translation of this article. If you can read Korean, I recommend reading that article instead.

Introduction

braian87b

In experience to get a proper working multiple wan configuration using mwan3 starting from scratch you should:

Important: this works well on OpenWRT 15.05.1, on newer versions there was some breaking changes, for example, the wan ifaces have ipv6 capability and now are named with letters ("wan, wanb... , wanc" instead of "wan, wan2... wan3" so wanb6 means 2nd wan ipv6.): https://github.com/openwrt/packages/blob/master/net/mwan3/files/etc/config/mwan3

The official documentation seems to be very detailed and up to date, I recommend reading those first: https://openwrt.org/docs/guide-user/network/wan/multiwan/mwan3 but I recommend to give a look at my config file below, since my approach for policyes is very nice.

First of all: Activate conntrack, docs says that is important and neccesary to get MWAN3 work properly, and it is needed to reboot:

searls

# This is just a scratchpad after I hacked what I needed in an irb session
require 'octokit'

Octokit.configure do |c|
  c.login = 'searls'
  c.password = 'c0d3b4ssssss!'
end

client = Octokit::Client.new
repos = client.repos #Note, for an org's repos, see `client.orgs.first.rels[:repos].get.data`

mariusGundersen

Programmer Collective Nouns

• An Enumerable of Rubyists @raganwald
• An Indentation of Pythonistas @raganwald
• A fold of Haskellers! @ReinH
• A Din of Twitterers @raganwald
• A callback of JavaScripters @irvingreid
• An NCC-1701 of Java Programmers @raganwald
• A relation of SQLers @raganwald

spikegrobstein

upstream plex-upstream {
	# change plex-server.example.com:32400 to the hostname:port of your plex server.
	# this can be "localhost:32400", for instance, if Plex is running on the same server as nginx.
	server plex-server.example.com:32400;
}

server {
	listen 80;

	# server names for this server.

jlong

var parser = document.createElement('a');
parser.href = "http://example.com:3000/pathname/?search=test#hash";

parser.protocol; // => "http:"
parser.hostname; // => "example.com"
parser.port;     // => "3000"
parser.pathname; // => "/pathname/"
parser.search;   // => "?search=test"
parser.hash;     // => "#hash"
parser.host;     // => "example.com:3000"

rofl0r

/* Support for executing system calls in the context of the game process. */

static const int injection_size = 4;

static const char nop_code_bytes[injection_size] = {
    /* This is the byte pattern used to pad function
       addresses to multiples of 16 bytes. It consists
       of RET and a sequence of NOPs. The NOPs are not
       supposed to be used, so they can be overwritten. */
    0xC3, 0x90, 0x90, 0x90