Indeed, a 5090 with oculink 4i makes little sense, it is the reason why I opted against a new GPU for now. One question, why do you say Oculink 8i is not reliable, do you have any more detail, sources? I am thinking of building my own Oculink 8i EGPU inside a Formd T1 case once we have an adapter of some sort (although I am not sure anymore that it will actually be built). Would be interesting to know a bit more about your concerns! Thanks a lot!
Actually i am not saying Oculink 8i itself is unrealiable. It is just most of the 8i compatible Pcie board are garbage. For example, I think you can only find the 8i to PCie board made by NFHK (or chengyang, which is just NFHK) the material, trace designs are not good enough for 4.0 speed. I have tried it with a 4i to 8i cable to test its signal integrity. It is horrible. When using Nvidia’s PCI error checker, the error can go very high. And more error means worse performance, and less stability. Thus, I will still try the Oculink design, but in the end, I will check something else, like Slim-SAS or MCIO… SlimSAS are usually designed to handle SAS4@24g, which is higher than Pcie4.0 @16g, and for MCIO, it is now used in Servers as PCIe5.0 and 6.0 interface, so it is a better option for stability.
3 Likes
Great info to know, thanks!
It is really crappy it is so difficult to find actually good hardware for this sort of thing vs. just random garbage made seemingly just for the lolz of chaos it brings.
I haven’t seen much from @Josh_Cook lately. I know he had some real-life stuff going on before. Josh, hope all is going well!
No wonder @Josh_Cook cannot get answers, these guys are just busy making new, cheaper products. I wonder if the 16" was a sensible purchase reading all this.
I for one was rather sad the FW16 got no love in the announcements. I’ve been ready for a new laptop for a year now, but:
- I can’t really justify buying the older 7840/7940 ryzen for ~$2500usd loaded new to replace my 11th gen i7 with a rtx3080m (other than crappy nvidia drivers in linux perpetually)
- There’s been no deals on ebay for even a used FW16 at a justifiable discount enough
- There’s not even a hint of an announcement if/when we might see an 8000 let alone a 9000 ryzen new/replacement board (or helping Josh with his module)
Threads like these are what got me excited to want a FW16, but watching the vendor involvement enthusiasm fizzle out like like this, I too question if maybe it just didn’t sell well enough for them to care to continue making/supporting it beyond this.
I imagine it’s a case that the 16 is still fairly recently released, and they’re focusing on improvements for their older models. They do have limited staff, they’re not a huge company yet, so they have to pick and choose their battles. Give them some time, I’m sure they’ll have something impressive coming for the 16 as well, they just can’t do everything at once.
7 Likes
I still don’t like this excuse.
They are small team is no excuse. The did a FW12 and a Desktop Computer nobody asked for.
It’s fine if they have enough people for doing it all but the did it and they ignored the FW16.
For sure my next laptop won’t be a FW and I told many people they should buy one which I won’t recommend anymore until it gets better. No BIOS support, No driver support, no QMK upstream patches. Doesn’t look good in my opinion. Instead if cleaning up this old crap they decided to produce new products.
5 Likes
This !
I have the exact same point of view.
Framework is gonna be the new AMD (yeah recently it’s getting better, I know, but still): great hardware, crap drivers.
This why, on other hand, Nvidia is so successful: they took care of the software stack and then everyone is happy and working with their hardware despite their bad/strategical hardware choices.
I know they aren’t ignoring the 16 but it definitely seemed that way at the event. I know their answer is we won’t announce until the product is 100% ready. But I feel like the motherboard upgrade roadmap should definitely be announced, or something. Like the 13 is our flagship baby because that’s how we launched and it will get new processors in Feb, and based on design flow and 13 history we except to make 16 upgrades… (2 months behind the 13 or in Q2, or some time frame). I feel like GPU updates would be nice to know but those could run into all types or problems (managing heat, board size, etc…) and mobile GPU update announcements are few and far between. So I’m ok with not knowing (I’m counting on Josh coming threw with the oculink anyway making the GPU modules kinda not a concern). Basically I just appreciate more transparency on the mainboard roadmap, because those shouldn’t be secret or scrapped for some error, since if it works in the 13, I wouldn’t see why it wouldn’t work in the 16 (meaning mainboard upgrades should be the only product with 100% delivery rate, since that’s the whole concept of framework). You can keep OLED or touchscreen 100% secret and announce only after it’s 100% in production, that’s cool with me.
1 Like
I agree with your point generally, but I am also disappointed at how it all went. But ultimately I do understand, the CEO wanted to branch out new product lines asap, and ultimately Framework is now also simply a bit dependant on AMD (not sure to what extent they are collaborating with Nvidia). One thing I do however have to say is that if there is no new GPU module for the Framework 16 in the next 2 years, then that is simply a failure and break of promise to the community. I get it that reacting to AMD saying they have a new chip and bringing out a 4.5L Desktop is good for the long-term business case. But in all honesty, I would have so, so much preferred an announcement of a RTX 5090 laptop module with 24gb of VRAM… Just my point of view.
A lot of companies don’t release roadmaps for their products such as Apple. This is because developing new products takes time and involves a lot of back and forth between partners, especially as Framework is a smaller company. In addition, challenges can arise during the process such as in the case of the AMD Framework 13. It is better to deliver only what you can promise than to overpromise and underdeliver. People may see the Framework 13 as a more mature platform however, the reality of things is that dev work for any platform can be challenging in different ways. In addition, there are a lot of instances of media reporting companies delaying road maps and that can affect their stock prices. Now imagine how it could affect Framework. It could shake investor confidence. I know a lot of people are trying to time their purchases for various reasons however, as technology continuously advances, what was once new will eventually become older. I believe that for tech, if it’s something that you need in the moment, then buy it and enjoy it. If it’s something you don’t need to buy, then just wait.
1 Like
I think Framework has done well by at least the FW13 in being able to iterate year on year new hardware releases, but so far it doesn’t seem like either they’re capable of, or care to do so anymore with the 16. This is a bit sad 2 years later without any real quantifiable product enhancement release beyond the ssd module only recently.
This leads me to believe there is something preventing an update, as this should be a purely iterative work to slap in a new chip and connect traces mostly. So said the simpleton perhaps, but that was the whole point of the product, no? Now 2 years later without an update to the motherboard, why?
The largest sacrifice with only 1 full and a only another half m.2 drive slot I was hoping they’d overcome in a v2, but at this point I’m just wondering if there will ever be a v2.
Same can be said for the GPU module updates, but as others said prior, I had hope to use an external module for this. Though egpu’s are depressing in their own right that the industry won’t move beyond a dumb 4x occulink connector for 8x or 16x extendable busses, seems an utter lack of vision vs. clone and reproduce economy.
I was hoping this forum might be one with energy to work past those limitations with individual ingenuity such as Josh Cook and other contributors to the conversation here, and still do.
Unfortunately for Framework, I’m not ready to drop cash on a new Framework 16 until I see an update now, not sure why anyone else would buy one at this point either. I can’t imagine they’re producing any new either, so it would be nice to hear of some roadmap for those of us still excited by the (near) future possibilities.
second update on the adapter. traces for oculink are done. both are linked together for a 1x8 connection. as long as the PCIe board is capable of doing it. I have already got my hand on a 2 oculink 4i to mcio 8x cable and a mcio egpu board. now just need to finish the thing…
9 Likes
There was an announcement in the forums by Nirav Patel, that the difference between the 7000 and 8000 series APUs is too small, so the 8000 series got ruled out.
Nirav announced in the forums, that they’ve recently got a dedicated software team, so this should get addressed soon.
Framework never had any kind of roadmap and I don’t see this happening in the future either.
2 Likes
Recently? Wasn’t that like 4 or 6 months ago? Not recent at all.
I’m seeing a mismatch in trace length and even via count that would, as far as I know, completely decimate the signal integrity. Or maybe I’m missing that, in which case don’t mind me.
Also I hope it’s temporary but your board dimensions seem to exit the chassis - worse, they don’t just go through the slit intended for IO, but the whole length of the card, meaning you’d have to cut into the plastic to add this board. I don’t see a reason for that so probably temporary?
I’ve made a 4-layer board for dual 4i before (though never tested, I think there might have been some known issues even).
The two inner ground planes are not displayed.
See how traces wiggle around to match length?
I also started work on a 8i version using a 6-layer board since I think I wasn’t happy with the tolerances on the 4-layer board.
Sadly had to abandon it as it took up way too much time and Josh started working at that time (and even planned to add a redriver which I never planned). And seeing he had issues, I am not confident I wouldn’t have inevitably faced the same issues anyway.
1 Like
Ok so apparently that version was already public:
However I apparently hadn’t yet used custom DRC rules yet which means this is super early. DO NOT USE.
If you produce with JLCPCB I have a cutom rules setup that should ensure proper impedances for both PCIe pairs and DP pairs. Setup for 6 layers right now, and I’d recommend you use 6-layers too, but could adapt. Not sure how the current kicad version fares but 2 years ago the interactive router didn’t respect these rules so I made a custom version that fixed it and made routing this actually viable. Sadly never got merged.
Coincidentally I am in need for Oculink expansion pretty soon so may start working on this eventually if Joshs version takes much longer yet (@Josh_Cook please at least be forward with the problems that you’re facing so parallel efforts don’t start when there’s fundamental issues)
4 Likes
Guys, awesome work. Can’t wait to see the final board.
Given your collective expertise on PCI-E boards, check out this thread, I’m sure you’ll be able to monetize such a board once the 7700S gets a successor… Just a hint! 