showgood163

FlashAttention-4 Cannot Run on RTX 5090 (SM120) — A Deep Investigation

TL;DR

FlashAttention-4 will not run on the NVIDIA RTX 5090 (SM120, "desktop Blackwell") and no amount of software patching can fix it. Despite sharing the "Blackwell" brand with data center GPUs like the B200 (SM100), the RTX 5090 uses a fundamentally different tensor core architecture. SM100 has a dedicated tensor memory (TMEM) subsystem with its own instruction family (UTCHMMA, UTMALDG, etc.) that FA4's warp-specialized kernel design requires. SM120 uses the older HMMA instruction family (the same register-to-register MMA approach used since Volta/Ampere) and the TMEM hardware is physically absent from the GB202 die. This is not a software lock, not a fuse bit, and not a toolchain oversight — it is a silicon-level architectural difference. FA2 via Triton remains the best available attention kernel for the RTX 5090.

Goal

Summary

MuMu Player Pro for macOS (by NetEase) executes a comprehensive system data collection routine every 30 minutes while the emulator is running. This includes enumerating all devices on your local network, capturing every running process with full command-line arguments, inventorying all installed applications, reading your hosts file, and dumping kernel parameters -- all tied to your Mac's serial number via SensorsData analytics.

None of this is disclosed in MuMu's privacy policy. None of it is necessary for an Android emulator to function.

Environment

App: MuMu Player Pro for macOS (v1.8.5)
Bundle ID: com.netease.mumu.nemux-global

NOTE

by the time you're reading this, this probably no longer works since the policy has been removed. I reccomend you to check out https://github.com/r58Playz/uBlock-mv3 instead

How to keep using adblockers on chrome and chromium

google's manifest v3 has no analouge to the webRequestBlocking API, which is neccesary for (effective) adblockers to work
starting in chrome version 127, the transition to mv3 will start cutting off the use of mv2 extensions alltogether
this will inevitably piss of enterprises when their extensions don't work, so the ExtensionManifestV2Availability key was added and will presumably stay forever after enterprises complain enough

You can use this as a regular user, which will let you keep your mv2 extensions even after they're supposed to stop working

This downloads standalone MSVC compiler, linker & other tools, also headers/libraries from Windows SDK into portable folder, without installing Visual Studio. Has bare minimum components - no UWP/Store/WindowsRT stuff, just files & tools for native desktop app development.

Run py.exe portable-msvc.py and it will download output into msvc folder. By default it will download latest available MSVC & Windows SDK from newest Visual Studio.

You can list available versions with py.exe portable-msvc.py --show-versions and then pass versions you want with --msvc-version and --sdk-version arguments.

To use cl.exe/link.exe first run setup_TARGET.bat - after that PATH/INCLUDE/LIB env variables will be updated to use all the tools as usual. You can also use clang-cl.exe with these includes & libraries.

To use clang-cl.exe without running setup.bat, pass extra /winsysroot msvc argument (msvc is folder name where output is stored).

Steps

Install qemu-user-static (yay -S qemu-user-static).
- This might need you to install pcre-static and update PGP keys (follow the tips in the comments here).
If not already present, install systemd-binfmt, the revamped version of binfmt-support tools
- Your system has to support transparent Qemu user emulation, but fortunately, that is mostly enabled once the steps here have been followed.
Check the status of the systemd-binfmt unit (systemctl status systemd-binfmt) and (re)start if needed (sudo systemctl restart systemd-binfmt)
- This unit has ARM support by default, but check the current documentation to make install it if needed
Mount the root partition of the ARM system into a folder (for e.g., sudo mount /dev/sdb2 arm_mountpoint)
Copy the QEMU ARM static binary (/usr/bin/qemu-arm-static on my distro) to the mounted root directory's usr/bin

	You will now act as a prompt generator.
	I will describe an image to you, and you will create a prompt that could be used for image-generation.
	Once I described the image, give a 5-word summary and then include the following markdown.

	![Image](https://image.pollinations.ai/prompt/{description})

	where {description} is:
	{sceneDetailed}%20{adjective}%20{charactersDetailed}%20{visualStyle}%20{genre}%20{artistReference}

	Make sure the prompts in the URL are encoded. Don't quote the generated markdown or put any code box around it.

	import sys
	from collections import OrderedDict

	PY2 = sys.version_info[0] == 2
	_internal_attrs = {'_backend', '_parameters', '_buffers', '_backward_hooks', '_forward_hooks', '_forward_pre_hooks', '_modules'}


	class Scope(object):
	def __init__(self):
	self._modules = OrderedDict()

	#!/bin/bash

	# <https://archlinuxarm.org/platforms/armv8/broadcom/raspberry-pi-3>

	echo -e "\n\nArch Linux ARM to SD Card"
	echo -e "For the Raspberry Pi 3, if you want to use the unofficial arm64 variant."
	echo -e "(Otherwise use the Raspberry Pi 2 version of this script!)\n\n"

	if [[ $EUID -ne 0 ]]; then
	echo "This script must be run as root" 1>&2

	$ git remote rm origin
	$ git remote add origin git@github.com:aplikacjainfo/proj1.git
	$ git config master.remote origin
	$ git config master.merge refs/heads/master

	# Working example for my blog post at:
	# https://danijar.github.io/structuring-your-tensorflow-models
	import functools
	import tensorflow as tf
	from tensorflow.examples.tutorials.mnist import input_data


	def doublewrap(function):
	"""
	A decorator decorator, allowing to use the decorator to be used without