Framework’s new desktop is mini ITX
Framework’s new desktop is mini ITX
arstechnica.com
Framework’s first desktop is a strange—but unique—mini ITX gaming PC
41 评论
-
-
Soldered CPU on a desktop: DOA.
SOLDERED RAM ON A DESKTOP: lol. Lmao even. What the fuck. What ever happened to "you should be able to fix your stuff"? Their laptops have user-replaceable ram btw. This is a joke.
-
I pre-ordered the 128GB SKU in the second wave. Soldered RAM doesn't matter to me if I am already maxing out what the platform is capable of. If I can dynamically configure the memory allocation between the CPU and GPU, this will be an extremely potent little AI workstation. I'll be able to cancel the pre-order of things aren't what I expect, and it isn't much of a loss for me ($100 refundable deposit).
I do agree that this branching away from Framework's roots, but I am still very happy that they are doing interesting things. I've always thought that what Minisforum has been doing with their SFF workstations has awesome, so I'm glad to see other companies wading into the same space.
-
They have confirmed you will be able to set a 96GB GPU memory split (windows, 110GB on linux)
-
-
You forgot no serviceable GPU
-
I don't think it's a bad product but it's just so weird coming from Framework. Given how their entire shtick is easy repairs, easy upgrades, less waste by being able to recycle some of these parts you swap out into other devices, etc. It's just weird to have a soldered CPU and RAM on a desktop FROM FRAMEWORK.
-
I think it's more than weird. Soldered ram on a desktop would surprise me even from HP.
-
-
This sounds more like an LLM box than a desktop. High memory bandwidth, GPU cores, kinda like the Mac minis running Exos. With 128 GB it would beat the pants off a much more expensive nvidia.
-
It looking like an AI machine is more AMD's doing than Framework's. Framework just grabbed one of, if not the highest performing mobile chip, which happens to be shipped with a big ass gpu
-
Well, I mean they didn't need to get full bandwidth on the memory, they could have just gone with socketed RAM. And 128GB max is a hell of a lot for a desktop, but not a machine that you want to load multiple large models into and have accessing as fast as possible.
I'm considering buying the bare board for a local processor after someone else takes the risks and benchmarks it.
-
-
-
I thought their whole mindset was repairability and upgradeability. This is just an all in one board with a case. With a starting price of $1,000 even comparable Mac mini's are $300 less if you want performance or just get a console if you want to play games.
Would have loved if their desktop would have just rehoused the 13 or 16 motherboards with upgraded cooling potential and even maybe thunderbolt to GPU adapters
-
This computer is not a gaming machine (as they falsely advertised), because you can definitely get both more powerful and more upgradeable for less.
It is exclusively an AI computer. Its whole advantage is to provide massive high bandwidth memory. It serves only that purpose and it serves it amazingly well. For anyone else, it is not a good value proposition. But in the AI space, this machine is a fantastic value. There is nothing else out there with 128GB that even comes close in price. An RTX 5090 has 32GB and costs 2000 USD alone, without any other component. The nvidia digits probably won't have 128GB and certainly not at that price.
This competes against Apple's computers with their HBM where people run LLM lovally and it does it using many more standardized components and with a much more reasonable price.
This machine serves a niche exclusively and I don't blame anyone for dismissing it, but it's because it serves a very specific use case for which there is little to no alternatives.
Edit: yes the nvidia digits will have 128GB of shared memory, with 1 petaflops of int4 compute, source nvidia: https://nvidianews.nvidia.com/news/nvidia-puts-grace-blackwell-on-every-desk-and-at-every-ai-developers-fingertips starting at 3000 USD.
-
or just get a console if you want to play games
No. Consoles are just glorified DRM machines.
-
the one exception being a steam deck
-
-
Near as I can tell? Their stakeholders aren't happy with sales up to date. And considering one of their bigger ones is already running video "reviews" on how amazing this is and how you should buy it...
-
-
Durability with soldered cpu and ram? Wtffff
-
I'm looking forward to somebody building a cluster with these and running a full size DeepSeek R1.
For about 26.000€ you can get a cluster with enough RAM to run it. For comparison, a single Nvidia GPU with 80GB VRAM costs about 21.000€, bringing the total cost to about 400.000€ for just the GPUs to run it.
-
I was looking forward to a small replacement to my gaming pc. But the cost is wayyy out of my expectations.
I could get a very nice matx build for that budget.
-
Could you make something similar at the same size though? Small form factor PC’s have always been expensive.
-
-
Think of the minecraft servers this bad boy could run
-
-
The only way I can wrap my head around this is as a really expensive console, where the standard for modularity and upgradability is non-existent...
Just kindof a bummer. And I adore small computers.
It's a super cute device with lots of fun little details, I just don't see why go with this over a normal SFF PC for the price. It's easier? Cuter?
-
Exactly, there's more modularity in my ITX X570 Mb, although locked to the AMD CPUs, I can pop in these new 5800X3D and other chips that weren't available when I bought my motherboard, supports crazy high OC DDR4 speeds and even manages to deliver a full 16x PCIE 4.0 Slot plus 2 M.2 connectors. In ITX size.
-
-
I wish they had gone with an even smaller Mini-STX format like the AsRock DeskMinis but it's still cool.
41 评论
i'm already tired of people losing their minds over the soldered ram, as if it was in framework's control or that socketed ram could even compare to this... modular ram at this spec is coming but it isn't ready for this generation of chips. if you don't have a need for the insane vram just don't buy it, or get the case and build your own standard ITX system. no one else is providing this chipset to desktop users or running at its full tdp and framework made it available, i cannot find the problem.
For real. People are assuming LPDDR5x is the same as DDR5. You cannot get the speeds of LPDDR5x without the memory being as close to the compute as physically possible.
I don't think people are assuming that. People where assuming that Framework only made tech where repairablity was going to be the main focus because that's how Framework presented themselves. So if Framework wants to branch out to tech that's not strictly repairability focused it'll take time for consumers to adjust their expectations of Framework.
Its like when Ferrari puts out a $200 T-shirt, people will be like "stick to cars Ferrari lol". But, turns out Ferrari actualy makes almost half their money selling keychain and $200 T-shirts.
The problem for some Framework fans, myself included is "opportunity cost". What essentially happened was that AMD piggybacked onto Frameworks event to release a product that Framework spent half the event talking about.
Framework basically just released a PC case but, those already exists lol. And that's where the opportunity cost comes in, if Framework is spending time on a product a lot of their fans don't care about then Framework isn't spending time on products those fans do care about or products that better follow the companies mission statment which makes us upset.
People seem to be split on the desktop though. Just like a lot of my "normie" friends don't care about buying a repairable/upgradable laptop, I don't care about the benefits this Desktop offers. Personally I've just come to terms that the desktop just isn't for me and that's OK. I would have preferred Framework spent this opportunity to make a repairable tablet or something similar but, not everything is made for me.
i don't think this was a huge sacrifice of their time, it is just an itx case at the end of the day. i personally don't care in the slightest for LLMs or other AI work but that audience will throw money at anything, and i'm glad some of that will be going towards framework and AMD
I'm actually really excited about this. I'm hoping that an affordable hardware platform like this might be able to kickstart interest in ROCm, and hopefully will help break the stranglehold Nvidia has on gpu compute.
It's always been a catch-22 for ROCm, hopefully by providing large amounts of vram at a fraction of the price (a 40gb a100 gpu by itself is like 8k), maybe we can break out of the the carch-22.
Afaik all of the other computers that use this apu simply suck for compute - ones a tablet and the other is a laptop.
But being able to cluster these, and have over 500gb of vram for less than 10k? That's really exciting for me, not from a personal perspective, but for research and companies.