Framework Desktop Batch 7 Guild

Welcome Framework Desktop Batch 7 Guild members! ! ! ! ! ! !

7 Likes

I got the 128GB RAM model, and I’m planning on building an Unraid NAS with a Windows/Bazzite VM for (1) AI loads and (2) Steam games. Because I’m hoping to build this as a NAS, I’m REALLY hoping Framework will provide an alternate backpanel that exposes the ports for the PCIe card, per the discussion thread here:
Request: Case Opening for PCIe Card - Framework Desktop - Framework Community

I’m hoping to be able to use a Direct-Attached Storage DAS unit with SATA/SAS ports since this would provide overall better performance than a USB-based DAS. But to do this, I’d need to be able to access the ports of a PCIe card.

Anyhow, looking forward to seeing how others will be using and modifying their units!

4 Likes

64GB model here as I don’t really have a use case for 128GB. Realistically I’m not going to do much/anything with LLMs which require more than 48GB assigned to the 8060S. Also its a bit less of a commitment as we don’t know if AMD will continue to develop this product line.

For me its a small box which generates little heat, is super-efficient in terms of most workloads, plays games (at a PS5ish standard) and most important of all - doesn’t wail at you when the fans ramp up on cpu/gpu usage like everything else available in this power envelope.

I don’t envisage expanding anything as I’m aware of the limitations of the APU in terms of PCI lanes.

Edit - I’ve been intrigued about the APU for a while so I can also have a play around there.

6 Likes

Batch 7 ftw!

2 Likes

Thought i’d document my PC for the future upgrades.

Current idle current draw (via a Tasmota smart power plug)*

150W under my current win11 install, 120W under a Fedora live USB stick booted

5800X3D
6700XT 12GB
32GB 3200 ram (4x8GB sticks)
ASRock x570M mobo
2x NVME drvies (4TB and 2TB)
4TB HDD

Have a framework 128GB on order, it’s going to outperformance this by significant amounts at a fraction of the power…

3 Likes

395/128 ordered. (2TB ssd)
Mainly because I want to have an extension to my macbook pro, and I want to have a small linux box running at home.

I am building a few applications for managing and analyzing research data.
Basically I will have a database and be running ai agents for ingesting and storing data and then other ai agents that analyze the data.

3 Likes

Sounds interesting - I also ordered one (batch 10) with a similar setup to you, and looking to manage / analyse research data offline. I’d be keen to find out what sort of research it is, as I’ve been trying out so many different pipelines for analysis - consistency for me is quite difficult.

3 Likes

I ordered the 128GB mainboard and will put it in a mini-ITX case so I can get to the PCIe slot. I will dual boot Linux and Windows for an AI workstation and gaming box. Very excited to see all the great reviews and videos ovier the last couple weeks, can’t wait to join the fun!

What case you gonna use?

The research is usually different type of training studies of children with disabilities.

So the data is often a combination of person characteristics (like range of motion and type of disability) and performance in different tests. Strength, running/walking, etc.

I use an LLM as part of a python ingestion pipeline, the LLM matches the incoming data schema and creates a new table or appends data to the existing dataset.

Then there are a few steps that are regular data transformations.

Once I have the data in place I use google-adk to have a data scientist agent that performs a number of analysis on the data based on a theory. The result becomes a starting point for the real statistical analysis.

2 Likes

Ahhh, OK really interesting! A lot more complex than the data I analyse (qualitative 1:1 and contextual research interviews, alongside quantitative data). The key focus for me is keeping all the data stored and processed locally (keeping in line with permissions from the data subjects).

2 Likes

Ah, nice.
that sounds very interesting. qualitative analysis of interviews is not super-easy to do locally, so a very good use case for local LLM’s.
Storing the data locally is definitively a good priority.

1 Like

The Jonsbo t6 Black itx pc case Jonsbo introduces T6 ITX walnut handled case in metal and wood — available in black or silver | Tom's Hardware

Partly because it will put the PCIe slot at the top for easy access, as I want to do a no hassle multiboot with physically removable nvme drives using this: M.2 NVMe SSD to PCIe x4 Mobile Rack/Backplane with Removable Tray for PCI Express Expansion Slot, To

Hopefully it will all work together lol!

1 Like

that is a nice looking case!

1 Like

Need to start looking what NVME to get…or maybe 2

1 Like

I’m curious to see how long we have to still cancel our order. I’m 50/50 at this case, because I ended up getting a PC, so this would be purely for LLM stuff, and the token/s speed isn’t that great…. but it’ll be nice to keep stuff preloaded.

1 Like

I ended up getting a couple WD_Black NVMe on sale at Amazon during prime day. Got a good deal!

Yeah I just have heard lot of bad stuff about the WD so its not my current number 1 choice :stuck_out_tongue:

I guess I’ll find out in the next year or so whether they are any good, LOL :woozy_face::woozy_face::woozy_face::woozy_face::woozy_face::poop:

I can understand that. I feel a little 50/50 about cancelling based on a perception of ordinary performance, too. Especially so given the expandability constraints.

2 Likes