News Framework moves into desktops, 2-in-1 laptops at Second Gen event

oh hey look, it's a Ryzen 395+ mITX mobo that everyone and their moms have been nagging about.
I gotta admit, that 4.5L total package size is hard to beat. It looks similar to a Velka 3.

The lack of LPCAMM2 is a bummer, but AMD's top engineer said it couldn't be done, so looks like we're stuck with soldered on RAM.
 
The article said:
The mainboard on its own will be available from $799. Pre-orders are open now, and Framework expects to ship sometime in Q3.
Out of curiosity, does the motherboard support a standard heatsink, by any chance? I doubt it - and that's the main issue I have with a lot of these mini-ITX boards based on laptop CPUs (which seem to be more common in Asian markets, BTW).

I also wonder if you can swap out that 120 mm CPU fan for a 140 mm or 150 mm fan, which I also doubt.

If they were serious about customizability, they'd have really tried to give us more flexibility in cooling that beastly APU. Also, a full-length PCIe slot would've been nice, even though I'm sure it would've had no more than x8 connectivity.
 
Out of curiosity, does the motherboard support a standard heatsink, by any chance? I doubt it - and that's the main issue I have with a lot of these mini-ITX boards based on laptop CPUs (which seem to be more common in Asian markets, BTW).
I'd be really curious if the memory needs to be cooled, because if so I'm not sure how good a regular cooler would be for this. You can see the memory layout in the first picture and no matter how you mount a standard cooler something would be uncovered.

At first glance I thought maybe LGA 1700 mount compatibility, but the top to bottom appears to be slightly longer than side to side so most likely proprietary. Their cooler has 6 heatpipes, measures 123.75 x 123.15 x 54.58mm and is intended to use a full sized 120mm fan so I don't think cooling will be an issue. They say nothing about the cooler mounting on the mainboard product page: https://frame.work/products/desktop-mainboard-amd-ai-max300?v=FRAMBM0002
 
"The Framework Desktop is a 4.5L Mini-ITX machine using AMD's Ryzen AI Max "Strix Halo" chips with Radeon 8060S graphics. While this is a mobile chip, Framework says putting it in a desktop chassis gets it to 120W sustained power and 140W boost "while staying quiet and cool." Framework says this should allow 1440p gaming on intense titles, as well as workstation-class projects and local AI."

Lol, I mean, Strix Halo isn't REALLY a mobile chip. Is it going into "mobile" devices? Yeah. Where will it really shine? mITX and NUCs.

I'm not one to get that bummed out about it not having LPCAMM2 -- just is what it is. Being able to game at 1440p with good fps on an APU is still crazy to me. Some called it way back when but this is probably AMD's end-game as far as mainstream graphics (mid-end gaming GPU's that is) and certainly budget gaming graphics go. I mean, it's scaled up to 120W, dual CCD's, and a big ol' GPU die, so yes, AMD just needs to keep the successor models fed in terms of memory using LPCAMM2 in the future.
 
LPCAMM exists but not even Framework can get it working with Strix Halo. That is extremely disappointing and moves Strix Halo from the most interesting mobile processor I've ever seen to one I will probably pass up.
Why? Do you really expect you'd upgrade to more? I also wonder what the prospects are for having any higher capacity LPDDR5X in the next few years.

I share your dislike of soldered-down RAM. However, I'm willing to accept it, in exchange for performance benefits that wouldn't otherwise be possible. That presumes the pricing is reasonable enough that I can buy as much as I expect I'll need.

BTW, what I expect will happen in the next 5 years or so is that the market will transition to mostly on-package memory and your only option for expansion will be via CXL.
 
Last edited:
  • Like
Reactions: Loadedaxe
LPCAMM exists but not even Framework can get it working with Strix Halo.
I think this is the wrong takeaway from what they said. Strix Halo would require two CAMM2 sockets equal distance from the SoC. Now consider that they're doing an ITX board and ask yourself where they would put them.
That is extremely disappointing and moves Strix Halo from the most interesting mobile processor I've ever seen to one I will probably pass up.
I'm pretty sure 64GB is the highest capacity LPCAMM goes since they're limited on the number of packages. That means 128GB would be the most you'd likely see any time soon which matches the highest SKU. Sure it's bad from the standpoint of you have to buy the SKU that matches your expected maximum needed now, but it's not crippling maximum capacity like on package with LNL did.
 
I'd be really curious if the memory needs to be cooled, because if so I'm not sure how good a regular cooler would be for this. You can see the memory layout in the first picture and no matter how you mount a standard cooler something would be uncovered.
Yeah, if you look at the heatsink, it has wings over the memory. It also has thermal pads covering the RAM, just like in a GPU.
I'd guess the mounting hole spacing is proprietary due to the 8 RAM chips taking up a bunch of room.
It even has extra wings and mounting points to cover the VRM between the CPU and PCIe slot.
It looks like it uses 6 screws to mount.
I think this is the wrong takeaway from what they said. Strix Halo would require two CAMM2 sockets equal distance from the SoC. Now consider that they're doing an ITX board and ask yourself where they would put them.
I am guessing it wasn't possible to stick 2x LPCAMM in there due to the way the traces come out from the CPU.
Instead of 4+4 chips on the opposite ends like a threadripper, it's 2+4+2 bracketing a single side making for a poor signal path without soldered on memory.
 
LPCAMM exists but not even Framework can get it working with Strix Halo. That is extremely disappointing and moves Strix Halo from the most interesting mobile processor I've ever seen to one I will probably pass up.
You have to learn to live with it, if the price is right. $2k for the 16-core chip and 128 GB is certainly moving in the right direction.
Why? Do you really expect you'd upgrade to more? I also wonder what the prospects are for having any higher capacity LPDDR5X in the next few years.
Strix Halo starts with 32 GB, which is a bit low considering it's shared. Some AI applications might be fine with the 8-core, 32 CU model and 128 GB even though that will be a rare or non-existent pairing. Barebones could help you get to the amount of RAM you want, and might help you save some money.

On the capacity, it's unclear to me why 48 or 96 GB couldn't be done, or even 192-256 GB instead of the apparent 128 GB limit. I'm pretty sure there are 24 GB LPDDR5X packages on the market which could be used to get to 192 GB. 128 GB should be sufficient for the 70B LLMs though.
 
  • Like
Reactions: bit_user
On the capacity, it's unclear to me why 48 or 96 GB couldn't be done, or even 192-256 GB instead of the apparent 128 GB limit. I'm pretty sure there are 24 GB LPDDR5X packages on the market which could be used to get to 192 GB. 128 GB should be sufficient for the 70B LLMs though.
48GB and 96GB should be possible, but the highest density 32-bit LPDDR5X packages are 128Gb (Samsung and Micron at least, SK Hynix has horrible public part data but I'd be surprised if they were higher) which makes the maximum capacity 128GB.
 
Why? Do you really expect you'd upgrade to more?
Framework computers are supposed to be repairable. RAM failure is a real possibility. Soldered RAM is not repairable. I recognize options are growing more limited so I may have no choice but it's certainly a large strike against any Strix Halo device. And given that running LLMs locally is an appealing idea right now, I won't consider a Strix Halo device with less than 128 GB of memory.

Also if I may indulge in a tangent, this particular device from Framework sounds remarkably disappointing. It's much less repairable and upgradable than the desktop I use today and I wasn't shopping by those criteria when I bought its parts.
 
I think this is the wrong takeaway from what they said. Strix Halo would require two CAMM2 sockets equal distance from the SoC. Now consider that they're doing an ITX board and ask yourself where they would put them.
Under the board. If that doesn't work, don't use ITX. I honestly haven't read what they were going for but ITX is an attempt to make pieces for large form factors fit in a slightly smaller space. It's pretty useless which is why all of the most compact cases that support dGPUs have to use a riser cable with the sandwich layout. When I first saw the words "Framework desktop" I was hoping for a new form factor that comes closer to those sandwich cases but without the riser cable.
 
Under the board. If that doesn't work, don't use ITX. I honestly haven't read what they were going for but ITX is an attempt to make pieces for large form factors fit in a slightly smaller space. It's pretty useless which is why all of the most compact cases that support dGPUs have to use a riser cable with the sandwich layout. When I first saw the words "Framework desktop" I was hoping for a new form factor that comes closer to those sandwich cases but without the riser cable.
I doubt under the board would be viable because they're using the back for WiFi and M.2 already. On the LTT video the Framework rep claimed signal integrity problems when using CAMM2 which honestly only makes sense if there is no choice about where the memory traces go as @Notton noted above.
qudmB3q.jpeg
I definitely agree that it would have been better if they used a larger form factor since the cooling height is going to kill the ability to use a low profile case with it (aside from their custom designed case). If they wanted to stick to standards and have a small board Mini-DTX and Flex ATX could have been options or even something like ASRock Rack uses "deep" mITX would have given more space.

edit: I'm not sure anything like those low profile cases works without at the very least a daughter board providing the PCIe (though I'd absolutely prefer this to a riser cable).
 
Last edited:
Under the board. If that doesn't work, don't use ITX. I honestly haven't read what they were going for but ITX is an attempt to make pieces for large form factors fit in a slightly smaller space. It's pretty useless which is why all of the most compact cases that support dGPUs have to use a riser cable with the sandwich layout. When I first saw the words "Framework desktop" I was hoping for a new form factor that comes closer to those sandwich cases but without the riser cable.
If you look at how many PCIe lanes the 395+ has, you'd quickly realize why they didn't go this route.
https://www.techpowerup.com/cpu-specs/ryzen-ai-max-395.c3994
 
The only thing more annoying than an ITX mobo is an ITX case... I don't have large hands and I can barely fit them in to do anything in those tiny things.

This all just seems more like a toy for very rich kids or something.
 
  • Like
Reactions: Loadedaxe
The only thing more annoying than an ITX mobo is an ITX case... I don't have large hands and I can barely fit them in to do anything in those tiny things.

This all just seems more like a toy for very rich kids or something.
I tried it...I am guilty. I let the ITX bug bite!
I bought a 5700G, 32GB 3600MT/s memory, a Asus B550i Gaming Strix MB and a Goodisory A09 Black case with a Silverstone Flex 350W PSU.

I did this as a project for fun. (Not so much)
I had to take the case completely apart, I have small hands as well and when I got done...I swear I will never do it again!

It does work well, I just havent found a use for it. I may sell it or stick it in my closet and use it for a backup PC.
 
The only thing more annoying than an ITX mobo is an ITX case... I don't have large hands and I can barely fit them in to do anything in those tiny things.

This all just seems more like a toy for very rich kids or something.
it is metter of tasts.
I use as rock deskmini for 6 years and it is nice to have good performance in small and silent case (2L).
I wish framework was smaller, around same size as deakmini with external PSU.🤔
 
Framework computers are supposed to be repairable. RAM failure is a real possibility. Soldered RAM is not repairable.
It's not end-user repairable, but you can send it somewhere like these guys. For them, replacing a bad RAM chip is no problem:

Much easier than on-package memory!

Also if I may indulge in a tangent, this particular device from Framework sounds remarkably disappointing. It's much less repairable and upgradable than the desktop I use today and I wasn't shopping by those criteria when I bought its parts.
It's a tough comparison, because it uses an APU that you can't get in a desktop like you built. If it used a more standard CPU, then you'd have a point. However, I think they might've been looking specifically for market niches that weren't already being served by the DIY market.

I had been in the market for an Alder Lake N board, for quite some time, and it took me over a year to find one I could live with. Most of them appeared in mini-PCs with proprietary form factors. Of the few using mini-ITX boards, the specs were generally disappointing. It's not every day that you find the BGA CPU you want, in the form factor you want, with all the specs you want.
 
  • Like
Reactions: TheSecondPower
I tried it...I am guilty. I let the ITX bug bite!
I bought a 5700G, 32GB 3600MT/s memory, a Asus B550i Gaming Strix MB and a Goodisory A09 Black case with a Silverstone Flex 350W PSU.

I did this as a project for fun. (Not so much)
I had to take the case completely apart, I have small hands as well and when I got done...I swear I will never do it again!
I had long lusted after the Silverstone PT13 - a 1.3 liter Thin mini-ITX chassis. I wanted to use it for a passively-cooled micro-server, as sort of an upgrade to the way I had eventually started using my Raspberry Pi.

I found the seemingly right motherboard - a N97-based industrial board.


Well, it was quite an ordeal. First, the motherboard's heatsink turned out to be absolute garbage. To their credit, the reseller sent me a passive heatsink for it and I attached a 92 mm case fan to the case's side grille, blowing directly onto it (although it covered only about 70% of the heatsink). Because the case wasn't designed for a fan, that meant I had to remove the front panel USB headers. The fan also intersected the front panel header for things like the power button, so I had to get creative, pulling the wires out of their connector blocks, shrink-tubing them, and then bending them out of the way.

Next, I did some airflow mods, to reduce recirculation and direct more of the air over the CPU heatsink. Those mostly involved blocking some of the vent holes by taping bits of black paper from the inside. Also, I folded a piece of plastic over part of the SODIMM slot, next to the CPU cooler, that deflected more air towards it. It formed a chute for some air to cool the SODIMM and exit out the side/top.

Finally, I tried putting another heatsink under the CPU, to draw away some heat and transfer it directly to the bottom of the case.

The final issue was the stand, which has holes in the bottom but sits flush with the table, blocking air from reaching them. So, I had to cut and glue little pieces of plastic so I could attach rubber feet, in order for there to be some vertical airflow.

This is the most case modding I've ever done. If I knew how much work it'd be from the outset, I probably wouldn't have tried. Looking back, I don't really regret it. Even though it's still noisier than I wanted, it's silent at idle and has worked very well for its intended purposes.
 
Last edited:
  • Like
Reactions: Loadedaxe
I'm not sure anything like those low profile cases works without at the very least a daughter board providing the PCIe (though I'd absolutely prefer this to a riser cable).
If a new form factor was used, the PCIe slot could be moved to the top of the board and turned 90 degrees so that the graphics card will lay flat and the case can be very slim. Or it could be put on the back to support only the sandwich layout.
 
The 395 is great for laptops, but it's not my first choice for a desktop today.
It has most of the same architecture as a 9950X, except more memory bandwidth and a bit lower clock & power limits. I think it should perform a little better, in some cases.

The main downside is if you want a dGPU. There, it seems the most they could do is PCIe 4.0 x8. That's still enough for mid-range GPUs.
 
It's a tough comparison, because it uses an APU that you can't get in a desktop like you built. If it used a more standard CPU, then you'd have a point. However, I think they might've been looking specifically for market niches that weren't already being served by the DIY market.

I had been in the market for an Alder Lake N board, for quite some time, and it took me over a year to find one I could live with. Most of them appeared in mini-PCs with proprietary form factors. Of the few using mini-ITX boards, the specs were generally disappointing. It's not every day that you find the BGA CPU you want, in the form factor you want, with all the specs you want.
That's fair. As an alternative to a mini ITX desktop with a desktop CPU and dedicated graphics, it's pretty limited. But compared to most mini PCs it has more flexibility and far better performance. (Except many mini PCs have removable RAM.)