Odd, I’m told my Laptop should be here tomorrow.
Did you actually get a shipping email? Has anyone? I haven’t gotten one yet nor has my credit card been charged.
It’s getting a little close to the end of November; is Framework going to make their shipping target for this batch?
I haven’t been charged yet either.
Still waiting also.
Just received an email for my batch 2 pre-order.
Your pre-order is now expected to ship in the week commencing December 8th.
Yes. It’s been sitting in ANCHORAGE, AK since Nov 22 2025 07:46 pm. Great. I wanted to make sure it didn’t go to the Bermuda Triangle that was Memphis TN that my Framework Desktop was stuck at for 5 days. Now it’s Anchorage Ak. I guess, I got what I wanted. It’s not stuck in Memphis.
Just got charged this morning, saying it’ll ship “within the next few business days.” Makes me think it’ll ship this week, not the “week commencing December 8th”?
It’s in my hands peeps! I haven’t opened the shipping box yet, but I have it!
Ok, so now that I’ve installed Windows (Linux is coming next, I have an 2230 drive that’s blank right now, waiting for a Linux install.) I can say a few things about the system.
- DIY hardware install was a non-issue, and non-story. Highest praise I can give there.
- I wish the driver bundle was separated out. I don’t need to install 16 driver packages. I only need a few of them. Unless Framework is going to keep this up to date every single day to make sure that the drivers are all day 1 updates from the various vendors, I’d prefer a master zip file that I can pick and choose what I need to install.
- This keyboard on a windows laptop is pretty close to on par with a ThinkPad laptop. I think it might even be the exact same as far as feel goes. They knocked the keyboard out of the park with this feel. The backlight is also a very well done. I’d say that’s on par with the Macbook backlighting in most cases.
- The trackpad is VERY good. Fairly close to a Macbook trackpad, and that is again one of the highest praises that I can give it. It doesn’t quite seem to understand two tap is right click, all of the time, but it seems to be on it 99% of the time.
- I put 64GB into my system and the nVidia module. I’ll report back for LLM usage for programming. I’ve been using the Kwen Claude Thinking model with Zed on my RTX 4000 (20GB) workstation. I’ll see what the 5070 (8GB) backed with 64GB of system RAM can do.
3 Likes
After doing a Rust project, in an already existing code base, I ran qwen3-30b-a3b-thinking-2507-claude-4.5-sonnet-high-reasoning-distill with a 120k context window. That uses around 17GB of RAM, so way over the 8GB of the GPU’s RAM window. To it’s credit, the system remained responsive during the whole thing. I’m using this as a mobile workstation and I’m getting mobile workstation performance. Playing music in the background, didn’t skip a beat while doing the LLM workload. I gave it 3 files from my project that I wanted it to refactor one function for. It handled the task correctly, but slowly. It takes a VERY long time to read the 32K context that I gave it–around 5 minutes. (I’m not an LLM writing this, I just like em-dashes, I was using them before they were cool, ok?!) Once had processed the context it went into a thinking phase that happened at around 2tok/sec. It’s not ideal, but it IS working.
On a AMD 9800X3D, RTX 4000 on my desktop system, that has 20GB of GPU RAM, it easily does 120tok/sec with the same level of input. This is ALL about the memory wall. System RAM is pretty fast, but the travel time from the GPU to the system memory will always be orders of multitude slower than the direct from the GPU’s memory. There are so many factors that are involved here, but let’s just say this. Having memory that’s physically closer to where the computation is happening, having a bigger bus (more lanes of the highway), and having a higher speed limit will all help the speed that the computation can happen. Mitigating the memory wall has been a computer science problem for more than the length of 2 average computer engineer careers at this point. This system is good, but it’s not magic. This is part of the reason why RAM prices have sky rocketed recently. Need more DRAM to keep up with the next AI startup.
I wouldn’t buy this if you’re doing strictly LLM workloads. There are better places to spend your money. I’d recommend the Framework Desktop with 128GB of RAM to be honest. That’s Threadripper levels of memory bandwidth in something that you can carry around. I don’t recall the benchmarks I do on that system, but it’s around the RTX 4000 level of performance.
4 Likes
Whether any userspace programs will ever make use of the NPU driver support that’s been in the kernel for a while now is anyone’s question, but if it ever does actually happen I’m sure that’ll help. The hybrid models that make use of CPU (which, for Zen 5, is a beast for AI/SIMD workloads), NPU, and GPU seem very cool to me.
The use of the NPU for parts of workloads seems undervalued IMO. Sure, it might not even be as fast as the iGPU/CPU working in tandem, but in a laptop use case where power/thermal budget is the limiting factor, I can only see it as helping. Of course, that is if it ever gets integrated into user-space software.
On a different note, I got my shipping confirmation today. The tracking website "aftership” (which one of Framework’s tracking links point to) shows it should be here by Monday. Directly tracking it with the FedEx link doesn’t show a date, however. Let’s hope that it doesn’t get stuck in Anchorage like some other people’s orders have.
Since today is the last weekday of November, I’m hoping that my order ships today since Batch 2 is supposed to ship in November…
I never got an email saying my order was delayed until December 8th so I’m crossing my fingers.
No email and still waiting…
Still in Anchorage!
Well today’s the last day of November… card still isn’t charged yet… never received the dreaded December 8th email…
While they’re going to be late with the original estimate, technically they’re not late based on the “3-18 days” in the batch preparation email. That, however, doesn’t make it any less irritating.
My card has not been charged yet either and still waiting…
For what it’s worth, I’ve found it pretty quick and easy to extract drivers from the driver bundle, then only install what I want. That’s not as quick and user-friendly as some may want, but once someone’s done it once or twice, it’s a reasonable operation.
2 Likes
My order is scheduled to be delivered tomorrow. For those who haven’t been charged yet, I ordered a DIY Edition w/ AI 9 HX 370, no memory or storage, & no GPU; I wouldn’t be surprised if the orders that haven’t been charged or shipped yet are because of difficulty sourcing storage and/or memory.
3 Likes
I do recall there is a way to do it, but I don’t have the instructions on that readily available. It would be nice if Framework linked to documentation on how to do that from the downloads page if you want to cherry pick your drivers.
1 Like