Changeable GPU and Ryzen CPU

@nrp @framework-admin

-pls try to put a GPU something like Nvidia1650ti or Radeon 6600xt.
-Try making an AMD advantage edition with a Ryzen 9 5900HS.


NO, let them fix the issues with the current laptop, people are impatient.


I’m pretty sure framework can work on both. Looking at this thread, most problem should be able to get fixed with future batches and bios updates. People would also get impatient when they would have to wait 2 years for new laptop designs from framework. My guess is that in maybe half a year we could see a updated version of the current framwork laptop with 12th gen intel core processors, this would be one of the easier changes. Maybe in a year or so we could see new laptops in bigger sizes and/or support for dedicated GPUs. This is just a personal guess so don’t take it as fact or anything. Main factor in this would be the budget that framwork has to develop new laptops

As for changable GPUs, there are MXM cards which would allow you to potentially swap a gpu as long the cooling design is compatible. Though mxm cards are barely used anymore because of the thin and light trend. Laptops like the XMG Ultra 17 (2021) still offers mxm cards with nvidia rtx 3060 to 3080 GPUs

1 Like

@Josh_Cook 100% agree. I really wish people would allow them to get their feet under them before trying to request other projects. Having built gaming systems I would personally never have put the words gaming and laptop in the same sentence. No comments needed, I know some disagree…


Why does everyone want MXM? Thunderbolt enclosures are so much more versatile since they can accommodate full sized GPUs.

MXM cards are of limited use outside of laptops so they are essentially junk, which goes against part of Framework’s MO. Let Framework build up their team and stick to their core product for at least 2 years, all the while with the external GPU market supporting the people who want their awesome laptop to do more intense work. maybe 1 year out the incessant clamoring for Ryzen will make sense. but I would rather see Framework make a complimentary, thunderbolt attached, eGPU enclosure that can support the MASSIVE collection of standard GPU’s rather than trying to pump life into the shambling corpse of MXM.


eGPUs are pretty far from being the holy grail when it comes to GPUs for laptops. The biggest downside is that an eGPU enclosure has around the size of a small PC so it isn’t really portable. In other words you can only make use of it at home while laptop with an dedicated GPUs inside can be used everywhere. Another downside is the price. An eGPU enclosure with a power supply costs around 200-300$. With that price its not worth it to buy a low end GPU for sub 300$ meaning you are most likely gonna spend at least 600$ for the hole setup. While desktop GPUs are certainly powerful its still a lot of money for simply wanting to use a better GPU. The current desktop GPU prices have gotten so high because of the chip shortage that its really not worth buying one for some time. Also eGPUs arn’t as upgradable as they might seem. The bandwith of the thundebolt interface is already at its limit with the current gen of GPUs and you already have significant perfomance impact from using it. It is questionable weather a way more powerful GPU in the future could give you enough performance to make it worth upgrading. For example a GPU that should be double as fast could only give you 20-30% more performace in the end. A thunderbolt interface has around the bandwidth of 2 PCIe 4.0 lanes while desktop GPU is usually connected to 16 PCIe 4.0 lanes nowadays, thats something you need to keep in mind. On the topic of E-waste, a full eGPU enclosure probably produces more than double the amount of waste than a laptop. Adding an discrete to a potential future framework laptop would barely change the amount of E-waste in comparison.

I think i should further point the upside of using MXM modules instead of putting the GPU on the motherboard. Having the CPU and GPU be on two seperate PCBs means that if of example a component on the motherboard breaks you can buy a new mainboard and still keep the old MXM GPU module. If the GPU mailfunctions you also don’t have to replace the hole motherboard anymore. This already makes it cheaper to replace components and reduced the amount of E-waste. With MXM modules there is also the chance that you might be able to upgrade your GPU with a better GPU in the future. This way you could not only upgrade you cpu, thanks to the stadartized mainboard size of framewok laptops, but also the GPU. This would make framwork laptops almost as upgradable as a desktop PC. To solve the cooler incompatibility issue with a new MXM module it should be possible to make the cpu and gpu cooler in 2 seperate pieces. That way you would only have to replace the GPU cooler with a new compatible one, kinda like every desktop GPU also comes with a new cooler attached to it


I would like you to point out via a link to a reputable seller that is selling any generation of MXM GPUs for the MSRP of their desktop equivalent cards.

It is not economical for Framework to bear the brunt of supporting the small MXM market nearly singlehandedly at this point by pushing their design team beyond their limits to build it into a product, and their marketing/business team to go out and source these niche cards from suppliers.

next point: these cards also come barebones. its mainly BYO cooler (aka whatever the laptop has available). what if the cooler isn’t efficient enough for next-gen cards and the cards can do nothing but overheat and throttle. what if next-gen power requirements over-tax the connector (there is no auxiliary power on these cards) and cause them to under-perform when compared to desktop variants.

I have a eGPU enclosure that cost $200, fits mid-tier standard desktop cards, and is a slim enough (2.25 inches wide, 10 inches long, 6 inches tall) brick that I could easily pack it in a backpack with my laptop and use it on the go if I need the extra horsepower, or do without it and get better battery life. And I can upgrade this down the line when my desktop gets an upgraded GPU, I pass the old one down to the eGPU, and I have a GPU that is much easier to resell since the desktop market is much bigger than MXM. no eGPU enclosure generating E-waste, no unsellable GPU to collect dust in lieu of a dead market.

Your argument over bandwidth limitations is flawed in that thunderbolt 4 offers 4 lanes of PCIe 3.0 which is still plenty for reasonable GPUs. Why would anyone want to spend $800 on a card they would only use occasionally.

Again, I’ll reiterate. Let the Framework laptop do its thing as a truly upgradeable thin-and-light to appeal to the average user, while also having the connectivity to rely on existing technologies and broader markets for the enthusiast-level expandability. The Community should be more focused on “What can I do with the toolbox given to me” rather than, “I DEMAND this special tool NOW, I don’t care if it takes too much time and energy to make, or is extremely hard to find, I WANT IT.”


Thank you and well said. What’s being asked for is so niche. Let them get a foothold first.


4 PCIe 3.0 lanes have the speed of around 2 PCIe 4.0 lanes, both are pretty limiting with new GPUs like a 3060 desktop gpu. A rtx 3080 already losses around 50% of its performance. Of course it doesn’t really makes sense to put an 3080 inside a eGPU enclosure but this also shows that if you put a modern mid to upper midrange GPU like a 3060 or 3050ti inside the enclosure, you won’t be able to get significantly more performance with upgrading, the bottleneck of the thunderbolt interface is only going to get worse with future GPUs. If you already have a used GPU from a PC you own things look obviously different in terms of affordability, it also brings up the question why you need a laptop + eGPU when you have a desktop PC. I would assue that for most people who want a good GPU in their laptop do so because they don’t have a good PC. I personally see eGPUs rather as a small niche usecase for now, it is more of a compromise rather than an ideal solution. I feel like i’ve talked enough about eGPUs for, i don’t want to drag this conversation too far :sweat_smile:

On the topic if it makes sense develop a new framwork laptop already for 15" market or the gaming/workstation market depends on how much money and resources they have available. If the current framework laptop sells better than expected then it would make sense to start working on laptops for new markets earlier rather than later. It wouldn’t make sense to just sit on the excess of money and do nothing with it. They already have the groundwork done for it with their current 13.5" model, so the development cost would be way lower and thus more lukrativ. At the very least framework has to develop a new motherboard for future CPU generations like intel 12th series, otherwise the sells would drop because of the older hardware. In the end its only a question of how much money they currently make if it already makes sense to work on laptops for new markets.

The idea for the MXM module is of course only for the what if scenario that framework starts to work on a workstation/gaming like laptop. In this case it would be a good improvement in terms of repairability and potential upgradability. You have to at least admit that it would a useful feature if its possible to implement. I mean the only real downside of a MXM modul compared to soldering the gpu on the motherboard is the additional space it uses.


No, I’m digging my heels in. dGPU’s in general reduce battery life, meaning you can’t have a powerful workstation and a laptop with good battery life. since MXM looks to use all 16 PCIe lanes, once you add in the other 4 for an NVMe drive and 2 for things like I/O. you only have 2 PCIe lanes left, and no real way to do the same kind of expansion customization as what Framework has. Framework’s approach means that everything is open to choice.

In the future if a single Thunderbolt 4/USB4 connection becomes a real bottleneck you could just have two USB4 ports to an eGPU. My work laptop already does 2xUSB4/TB4 to a dock, mainly for power delivery but still.

I was about to say that we haven’t seen any recent MXM cards come out but Clevo seems to throw 30-Series in their laptops. So maybe the market isn’t as dead as I thought but I guarantee the second Framework tries to start asking for MXM GPU’s from wherever Clevo gets theirs, Clevo and all the big brands they supply laptops to are going to start picking on the little start-up.

And I’m doubling down on the fact that the MXM cards cost more new than their desktop counterparts. Sure, the used market exists, but both used and new combined are a far cry from the competitive market of the desktop GPU’s.

Final point: Thermals and power (again). my current work laptop is a Dell Precision 7550 with an i7 and a Quadro RTX 5000 Max-Q. It is as close to a “Workstation Laptop” as you can get. As I said earlier, I need to plug in TWO USB-C connectors just for the battery to stay charged while working if I don’t want to use the barrel jack DC charger. all of this amounts to a battery life of 2-4 hours max. Additionally even with a cooling solution made specifically for this hardware, and a cooling pad underneath it the CPU and GPU both run hot. I can’t imagine the design limitations of adding in GPU’s like this to a laptop meant to be as modular as Framework’s vision. you would essentially limit yourself to barrel jack DC power as well (which is a dying breed) until type C PD could match the 120W+ required in a single cable.

a modular dGPU, accounting for the range there is, would actually limit Framework to overbuilding a new laptop to essentially cater to the highest possible config, limiting overall flexibility and increasing cost of the laptop up front. killing any momentum the company is picking up from its current philosophy and market approach.


MXM have a 16 PCIe lane interface but you don’t have to connect all of them. On a desktop PC you can also have a 16 lane PCIe connector that is only connected to 4 lanes and you can still put a GPU in it without issues. The current Tigel tiger lake processors offer a total of 20 PCIe 4.0 lanes which can be use in 1x16+1x4, 2x8+1x4 or 1x8+3x4 configuration. On top of that you also have a bunch of PCIe 3.0 lanes from the chipset. You can connetct 8 4.0 lanes to the dGPU which is enough to not run into a nadwidth problem. At this point we have 3x4 PCIe 4.0 lanes left of which we can use 8 lanes for 2 M.2 slots. The framework laptop uses JHL8440 controlers for its TB4 ports which need x4 PCIe 3.0 lanes. I think the tiger lake also offers one TB4 port aswell on top. We could connect the last 4 PCIe 4.0 lanes to one TB4 contoller and connect the rest via the PCIe 3.0 lanes of the chipset. I does work :grinning_face_with_smiling_eyes:

I’m not sure if you can use 2 thundebolt ports for a eGPU but if its possible it would solve the bandwidth issue :thinking:

The thing about gaming/workstation laptops is that nvidia 3060 mobile at 120W TDP performs better than a 3080 at 90W. The laptop with the 3080 GPU would also cost around 600-800$ more. In other words if you want to get the most perfromance for you money you would rather not look at the thinnest laptops. That doesn’t mean it has to be a monster. The XMG core 2020/2021 laptops for example are still reasonably thin (20-23 mm) which also offering enough cooling for the 120W TDP mark. Trying to keep the power consumption within the 100 Watt limit of a USB C port would be a pretty bad decision. In other words barral plugs are not going to die on gaming/workstation laptops yet until the USB C port can handle 200+ Watts. I’m not entirely sure what frameworks stand on the discussion on Barrel plug vs USB C in this case. The downside of a barral plug power supply is that the connector is not universal, so maybe you could also add a USB charging port on it to so you can use it for other things too :thinking:
The tradeoff between performance and battery life is a sacrafice that a lot of people seem to be willing to take when i look at the growing gaming laptop market

I’m ot sure if a MXM module would be that limiting in flexibility. Frameworks likes to make standardised mainboard sizes which create the same limit

1 Like

but you are still skirting around the key questions, so I’ll ask again.

Where are the cheap, readily available, non-monopolized MXM cards coming from right now?

Would Framework, in it’s current state or even after two years of healthy growth down the line, even be able to utilize and rejuvenate this market to make a product in line with their idea of pricing, upgradeability, and open-source nature?

And finally:
If the ONLY OTHER BIG BRAND SUPPLIER THAT IS SUPPORTING MXM decides to drop out, what happens then? Does Framework walk back all the R&D done to support it, losing A LOT of money. Or do they push to keep the now-defunct standard in their laptops driving customers seeking easy upgrades and modern devices away and also… losing A LOT of money.

I hate to keep cutting this down but MXM is not the ideal play in the long-term, and definitely not the play in the short-term.


You make it sound like you need a another special manufacturer to make MXM cards aside from the motherboard of a laptop. Framework already has an manufacturer that is able to produce their motherboards for presumably a relativ cheap price. Its not difficult for a manufacturer that already makes an enitre laptop PCB to also produce a little daughter PCB (MXM card). Placing the GPU on the daughter PCB instead of the motherboard is not an difficult task. In the end you only pay a few dollars more for the production of the daughter PCB compared to just soldering the GPU on the main motherboard. Framework can also keep ordering more as long as all necessary parts are still being produced. Since framework can already prduce the MXM cards pretty cheap they can also sell them as spare parts themselves and/or via any other distributor

1 Like

That manufacturer would have to receive certification by Nvidia, AMD, and possibly Intel in the future to build cards based on their GPU cores that alone could take YEARS to get certified.

THEN, this manufacturer would have to sell enough cards to be able to maintain their supply lines with the “Big Three” if any of those companies see a loss in profit by sending their GPU components to this manufacturer, they would either cut them off completely, or if they are under contract to supply components, they could drag their feet. Increase prices, decrease availability to the manufacturer, etc.

To summarize and reiterate: High variability + low market volume make MXM cards cost more per model, and each model sell less volume-wise. the profit margin would be close to non-existent. As a worst case scenario, I would rather see Framework launch a CPU+GPU mainboard where both are soldered, but you get maybe 3 tiers. an i5/ryzen 5 with a 1650/3050/idk what the lower spec AMD GPU is, the mid-tier i7/ryzen 7 like the one now with a 3060/6600XT. and the high-end i7/R7-9 with 3070/6700XT.

granted this introduces 3-6 additional SKU’s to the current 3 with intel iGPU’s and the possible 3 with AMD APU’s that could come along. but that’s still 12 base products with maybe one added for a chassis with a more robust power delivery solution for the mid and high-tier configs (which you would have to do anyway for an MXM solution). So 14 big-ticket products that take up inventory space.

compared to the 3-6 mainboards with an Intel CPU/AMD APU and an MXM slot, the single next-gen chassis that would provide higher power delivery, and at least 20 different MXM boards to keep track of and sell. if Framework were to offer current and previous/next gen cards from any of the “Big Two” (or “Big Three” if Intel enters the picture). that’s 27 different, high-priced products they have to try and all sell at the same rate as their current products. If this doesn’t increase the required profit margin and therefore the cost Framework would certainly go under.

I’m not sure if a manufacturer would need a special certification to be able to produce MXM cards compared to the certification of soldering the GPU next to the CPU. The design of the motherboard needs to be already been certificated before you can even start to produce them, that goes for both MXM designs and non MXM design. I agree with you that there is always the risk that nvida or amd decide to stop giving out certifications for MXM card based laptops. I find it unlikely that they would do so out of a loss in profit, after all framwork is way too small to matter in this regard. Its not unlikely that they would stop giving out certifications because the standard is basicly dead though. Framework would have to create a very convincing product in this case


Those are the exact two reasons why they would never send components for these products to Framework. The company and the market are just too damn small when compared to the size of the desktop GPU market and the allure of non-upgradeable gpu’s in notebooks already. the former of which has already started edging unto the laptop market via more widespread acceptance of Thunderbolt and eGPUs; While the later of which is better for the businesses because consumers would have to buy entirely new products hence more money for the corporations.

There is a limit to modularity at a reasonable price. the dwindling MXM-compatible laptop market seems to speak to that. If big companies don’t want to do it anymore, why would little companies?

Framework has pushed their design right up to that limit with swappable I/O ports (and ideally other I/O components like screen, keyboard, trackpad, and speakers in the future), while maintaining upgradeable RAM and storage. Any more modularity would negatively impact Framework’s bottom line by either decreasing the already tenuous profit margin, or by driving the cost of products up and making this entire concept more unappealing to the masses of people who would consider this for their machine solely based on price and basic customization. yes it would be good for enthusiasts, but we are the (very) vocal minority that isn’t going to solely keep this company afloat. want to see what happens when a company relies too much on its vocal minority? look to Razer. They cater to a certain small group of people and their prices for similar products is more than that of the rest of the market.

1 Like

In my opinion the best outcome is AMDs new APUs. But this is coming from someone who has a desktop.


The best option if the CPU is AMD is to use the integrated AMD graphics. Hopefully we’ll have some opens to keep the firmware as open as possible.

You can’t have both

To my knowledge Ryzen is poorly supported in Coreboot and the PSP is even less well understood that the Intel ME

Besides, Intel’s Graphics Division is burning a fair amount of R&D money and their iGPU has become much more performant in the past few years

There will never be a shortage of people trying to take the steam away from a movement. It isn’t even logically coherent. AMD could release the source code or release some working source - or at least allow us to verify PSP is disabled… Or something! Would be happy if Intel helps us bootstrap their CPUs in a way that doesn’t force us to trust them.