Linux

Sweet Spot on a Cheap Cloud GPU: MoE, Vision Ceilings, and VRAM Budgeting

Sweet Spot on a Cheap Cloud GPU: MoE, Vision Ceilings, and VRAM Budgeting

117 tokens per second from a “35-billion-parameter” model on a $0.27/hr GPU. The counterintuitive part: a smaller 27B model on the same GPU manages 33 tok/s — about 3.5× slower. The difference is not the parameter count on the spec sheet. It is how many of those parameters actually fire on every token.

This is the story of an afternoon spent learning to read a VRAM budget — and why “it fits” is the wrong question to ask.

From Local to Cloud GPU: Ollama on a Rented A100

From Local to Cloud GPU: Ollama on a Rented A100

134 tokens per second on a 35-billion-parameter model. That is what an NVIDIA A100 produces for $1.40 an hour — rented in the cloud, no hardware to buy, no drivers to fight. The plan was simple: spin up a pod, install Ollama, and see what consumer-unfriendly hardware actually buys you.

The GPU vanished the next morning. And the cheaper replacement taught me something the A100 couldn’t.

WiFi Access Point Bridge with Raspberry Pi 3 – ETH to WLAN

A Raspberry Pi WiFi Extender is a cheap and power efficient way of increasing the total range of your WiFi Network. In my case I needed to extend a WiFi coverage within the same LAN (bridge mode from Ethernet to WiFi).

I have found a working shortcut 😎 to create an access point following a GitHub project so I didn’t need to follow instructions from raspberrypi.org page – Using the Raspberry Pi as an access point to share an internet connection (bridge).

Upgrade Ubuntu 12.10 (Linaro 13.04) to Ubuntu 13.10 (Linaro 14.01) and later to Ubuntu 14.04 LTS

I wanted to keep my existing configuration and programs on my Cubieboard2 with Linaro Lubuntu, but at the same I wanted to upgrade the packages since the Lubuntu 12.10 is quite outdated. This post describes my steps to upgrade my Cubieboard2 to the newer version, which can be done in 3 steps:

  1. Update repository list to old-releases derivatives
  2. Upgrade Ubuntu 12.10 (Linaro 13.04) to Ubuntu 13.10 (Linaro 14.01)
  3. Ubuntu 13.10 (Linaro 14.01) to Ubuntu 14.04  LTS version (long time support)

Blogger blogspot to self-hosted WordPress

Google Blogger is very simple and easy to start with blogs unless you start to think of custom plugins, widgets, tools and more advanced themes. Many of us are following this path that starts in google blogspot and then migrate to WordPress. You can find several pages comparing these two platforms and also many howto pages describing the WordPress installation. In this this post I will list plugins that I found necessary to run my new WordPress blog, describe how easy it is to migrate posts from blogspot to WordPress and also describe changes in my old blogspot to redirect traffic to the new self-hosted WordPress post. 

Kippo honeypot on Cubieboard Ubuntu

Kippo is an SSH honeypot written in Python. Kippo is used to log brute force attacks and the entire shell interaction performed by an attacker so I have installed it on my cubieboard2 Lubuntu version. According to wiki a honeypot is a trap set to detect, deflect, or, in some manner, counteract attempts at unauthorized use of information systems.

So lets start to install and test my first honeypot.

Ubuntu – data stored in RAM filesystem – TMPFS

In my Cubieboard2 I use Lubuntu. Since I have connected barometer and arduino to my cubieboard, I read data from that devices and store it to cubieboard2 filesystem. Since NAND memory is not quite suitable for such frequent changes because of wear-out of the memory cells over time, I’ve created temp space in RAM memory. Cubieboard2 has 1GB DDR memory so there is some space for it.


I will use TMPFS and not older RAM disks – ramfs. Tmpfs is supported by the Linux kernel from version 2.4 and up. On reboot, everything in tmpfs will be lost. The memory used by tmpfs grows and shrinks to accommodate the files it contains and can be swapped out to swap space.
To get more comprehensive info regarding those filesystems read this post or this post.

DLNA media server on Cubieboard with Lubuntu

I have installed Lubuntu on my new Cubieboard 2. I have also attached HDD to SATA so I have more space to store my media. This posts describes the steps needed to install and configure media server to share my media to other devices (media players). I spent some time to select the right one and chose minidlna tool (also known as ReadyMedia, lightweight DLNA/UPnP-AV server targeted at embedded systems).

According to Ubuntu documentation the MiniDNLA daemon serves media files (music, pictures, and video) to clients on a network. Example clients include applications such as totem and xbmc, and devices such as portable media players, smartphones, and televisions.
MiniDLNA is a simple, lightweight alternative to MediaTomb, but has fewer features. It does not have a web interface for administration and must be configured by editing a text file.

Mount SATA drive in Cubieboard with Lubuntu

The main advantage of cubieboard is it’s SATA connector. I have old HDD from notebook, so I wanted to use it as a media server.

This post has 3 parts related to HDD connected to Cubieboard2 with lubuntu:

  • Mounting HDD
  • Backup NAND
  • Spin down HDD


There are many ways to see all disks and partitions but my favorite command is lsblk:

sudo lsblk -o NAME,FSTYPE,SIZE,MOUNTPOINT,LABEL
NAME        FSTYPE   SIZE MOUNTPOINT LABEL
sda                149.1G
nand                 3.7G
|-nanda     vfat      64M            Volumn
`-nandb     ext4     3.6G /
mmcblk0              241M
`-mmcblk0p1 vfat     241M            SSD

or also fdisk: