I’m trying to be patient, but watching these reviews of new systems is hard. The Beelink GTR9 looks impressive, but I’m really drawn to the Frameworks ITX design.
Beelink looks good, but has issues with it’s NIC disconnecting under load.
Looks like Corsair AI Workstation 300 has a Mini-ITX motherboard, but there is surprisingly very little info about it, even though it shows as in stock now. It does feature 300W FlexATX PSU, but no pictures of the inside of the case, so who knows. Based on one of the Reddit threads, it doesn’t have a PCIe expansion slot, but not clear if it’s a custom fit board or a standard MiniITX one.
Thanks! Looks like we shouldn’t expect any progress within the next 2 weeks for Batch 13?
Is all of Batch 12 128GB shipped now?
Not all, your customer page will have the most up to date info. Feel free to check back in two weeks, but I can’t guarantee an update. I just know Q4 right now 
3 Likes
Looks like the Lenovo one that was launched in China only
Looks like it uses the same Sixunited motherboard as GMKTek Evo X2 and other similar mini-pcs, but I wonder if they use a better cooling system, as it has a larger case similar to Framework.
Hard to know as, in the absence of reviews, I’d say it’s vaporware
There are a few people on Reddit who have it, but they haven’t answered any questions about thermals or internals, other than confirming the absence of PCIe slot.
In any case, I’m now seriously considering just getting the GMKTek unit - it doesn’t seem like PCIe slot on framework is usable, and since APU and RAM are soldered in anyway, there not much difference between getting Framework and any other Strix Halo mini-pc. The only thing that I’m not sure about is the level of fan noise on GMKTek unit and thermals overall.
One big difference between the FW and others is the cooler design
If I had to pick one from the pool of others it’d be HP
Yeah, when it comes to a 2k PC, I will rather wait 2 months more than risk buying from a different vendor
Edit: There’s always the 4k option from NVIDIA
Just to spare others the click, this isn’t even x86, it’s just another AI only option.
You’re right, I should have said it was only for those looking for a PC for AI and that the link is to an article on NVIDIA’s expensive and late DGX Spark AI mini-PC. So, if you are only interested in x86 don’t exhaust yourself by clicking on this link 
DGX Spark has similar performance to Strix Halo for twice the cost. It’s also a custom ARM chip and runs NVidia’s Ubuntu fork - not sure if you can install any other ARM Linux there. It’s more like a dev kit for larger DGX ecosystem than a general purpose AI computing device. I’d buy one for $2K, but not for $4K.
BTW, I’ve ordered GMKTek – there is a promo on Amazon and I got 128GB version for $1850 with 2TB SSD - $240 less than my Framework Desktop WITHOUT an SSD.
I’m still keeping my Framework preorder just in case, but if I like the device, I’m going to cancel it. I like Framework and I wouldn’t mind paying extra, but it doesn’t seem to be problem free either, and the useless PCIe slot is a real disappointment. We’ll see.
2 Likes
I’m starting to waffle on keeping my pre-order and not just for the delays, though that certainly isn’t helping.
As for the DGX Spark. It’s a different beast. If you’re buying one for pure AI usage, then it’s superior to the 395+. The AMD is 50 TOPS while the DGX Spark is like 1000 TOPS.
They are not really apples to apples. The DGX Spark is not made for consumer use while the 395+ is AI capable, but primarily geared towards consumer use.
50 TOPS for AMD is just the NPU. DGX Spark has better graphics compute, a bit below 5070 level which is pretty impressive for the device that small, but it has only slightly higher memory bandwidth (273 GB/s vs 250 GB/s) and less powerful CPU too.
AI usage is too generic. If you want to use it with LLMs, there is not that much advantage compared to Strix Halo, definitely not worth the price. For image generation it makes more sense to just get a dGPU. So, that basically leaves fine tuning larger models and more “traditional” ML flows that use CUDA.
It would be more popular if it didn’t cost that much. I think for $2500 it would be a great competitor for Strix Halo models, I’d definitely buy one for that price.
1 Like
Nvidia’s DGX Spark’s intended use is for Enterprise AI Inference Backends for corporate privacy over hosted API inference. It has significantly higher AI performance ~20x than the AMD AI 395+ for inference purposes.
With that said, we can just agree to disagree.
Well, got my GMKTek unit. In the process of creating a recovery drive for pre-installed Windows and getting Fedora 43 beta installed. The fans kicked in a few times, but they are not that annoying, similar to an average Windows laptop. Definitely not desktop-level noise (so far).
2 Likes