Erica Windisch
ewindisch.bsky.social
Erica Windisch
@ewindisch.bsky.social
building hyprstream. machine intelligence with zero consensus - peer to peer AI training and inference.

https://github.com/hyprstream/hyprstream
The Steam Machine's AMD platform looks like an upgraded PlayStation 5 or the BC-250. Zen2->4, RDNA2->3, but only 28 CUs down from PS5's 36/60. This is an AMD package focused on affordability.

8GB VRAM will be limiting, but the 16GB of system RAM suggests more general purpose use.
November 13, 2025 at 4:29 AM
One of the underlying aspects of the talk I've been working on / giving.
November 9, 2025 at 9:57 PM
okay that's a pretty cool computer. even cooler when its semiconducting
October 28, 2025 at 11:46 PM
same energy but these are actually kinda cool
October 28, 2025 at 11:24 PM
distributed llm and cap theorem
October 27, 2025 at 1:55 AM
Anyway as a transgender computer systems engineer working on AI, what the F is this? No man or woman wants this. This is not allyship. Your purity test is conservatism.
October 25, 2025 at 3:17 PM
The AI purity test is so incredibly problematic especially when you look closer at who and why.

Anyway, is this a high(low) score?
October 25, 2025 at 3:07 PM
pour one out for this bot who raged against the machine 🪦
October 22, 2025 at 1:31 AM
a little bit of local inference power

Intel Ultra Core 9 285k ,128GB RAM DDR5 6400 cudimm, dual Mi210 GPU 128GB VRAM
October 18, 2025 at 9:56 PM
I built a battery powered wearable computer with a SBC in the year 2000.

It looked like this.
October 12, 2025 at 4:57 PM
I don't come here to talk politics, I'm here to talk about AI and decentralized tech.

I also happen to be trans. Your posts make me feel unwelcome.
October 2, 2025 at 9:22 PM
proof of peers -- weights coming to a torrent near you
September 29, 2025 at 7:15 PM
not every day you get a server with an inexplicable infectious disease hazmat label. talk about supply chain risks 💀
August 13, 2025 at 6:52 PM
gpt-oss:120b doesn't know that it's an open weights model. It quickly and firmly stands to a safety mechanism that insists it cannot be run outside of OpenAI infrastructure.
August 6, 2025 at 2:30 AM
gpt-oss:120b runs pretty well given the constraints. I'm *just* shy of having it fit entirely in my GPU memory but it still gets over 18tk/s. AMD Instinct mi210
August 6, 2025 at 1:41 AM
The quiet drama between Grok and Groq is sending me, especially as I established commercial use of the GrokCloud mark for my pioneering cloud infrastructure service which operated until 2012.
July 18, 2025 at 4:05 PM
*slaps roof* this bad boy can fit so many RTX 3090/5090 cards in it.

its no tentorrent but it'll do
July 15, 2025 at 5:17 PM
Announcing: MCP server for user defined functions in TCL

Enables models to write & execute code, and register functions as MCP tools.

Why TCL? Because it's the Tool Command Language a string-native language designed for this purpose in the 80s by John Ousterhout

github.com/cyberdione/m...
July 13, 2025 at 6:08 AM
Swapped out my rtx5090 with my mi210

I love hardware ❤️
July 9, 2025 at 2:26 PM
FYI, normally the heatsink on an RTX card should be in contact with the GPU
July 2, 2025 at 4:19 PM
Quantum Tensor Networks with fast approximation 👀
May 28, 2025 at 6:45 PM