but really the problem now is that. I am kinda losing faith in framework now🥹
I am just trying to improve on a obsolete platform. the CPU bottleneck and just the overall design of ram position and peripheral are all significantly hinder the performance. you are really not getting any significant performance increase with a better gpu. I have a 5090 and in many game I am seeing just 10% increase from the 7700s. starting to think about alternatives already…
I’ve been super annoyed with nvidia for the past few years, and looking for a new laptop brought me to check out where the framework16 was at to look at amd more than intel/nvidia cartel. Framework 16 was idea for a number of reasons, this thread more than most, but the platform hasn’t moved on sadly, and doesn’t seem like it will.
If I were more ambitious I’d make my own mobo and take over their platform as at least a defined specification with commodity things like cases, battery, display, etc. I find myself later in life and care less, hoping other rational humans might with more will, but still a distant thought with still no better options lately.
I really do wish the rabid/rapid OEM’s like Minisforum did the same sort of form-factor vs bespoke one-off things every time they do, but I still have some hope for companies doing rapid integration like them in a platform like framework created, but seems to abandon.
I plan to reuse my old 1070, which is still way better than the iGPU, so Oculink would still be useful.
I unfortunately also agree on this as mentioned before. The Framework 16 options are getting a bit limiting now and in all fairness, are starting to get old for the price. We urgently need an update and my expectations are justifiably high. I would expect a solid state-of-the art mainboard with tb5, as well as a very capable GPU module with at least 16gb of VRAM. But on the other hand, I am also aware Framework as a company may be struggling with the current tariff situation and resulting uncertainty. Still, they had ample time - years - for product development before all this started. It is now the time to really provide an update for the Framework 16 line soon.
I’m not sure what you’re seeing with your 5090, but I am having an excellent experience with my 6950XT. Now I have no modern desktop PC to compare against, nor are there good benchmark suites for linux (and enough public info to compare against), so I’m not sure how many %s I’m leaving on the table compared to a full 16x link on a modern desktop.
But I can finally play modern, graphically intensive titles again, and it’s solid, and the CPU is plenty powerful to drive it.
Now I’m not discounting the issues you’re experiencing, I’m just trying to put it in perspective for anyone reading this: This very much is a solution, and while it still LOOKS jank (and frankly, it is), that does not show for me in the day-to-day use. The cables are all inside, the DEG-1 provides a good eGPU mount. It’s just not the excellence you might have expected buying the FW16, imagining 8i Oculink with custom boards and all that glamour, which I agree is disappointing.
So I’m still going to buy a card the moment someone here has them in hand and puts them up for sale:)
Edit: Yeah I think we’re expecting different things, not fair to compare a very high end GPU to a older, still high end one, and I agree an upgrade would be great, but idk how many people are expecting performance on par with the very best desktops - I certainly didn’t, and was positively surprised. I see that given the potential, you might expect better.
To me all this just proves that FW16 is an unloved product by its manufacturer. They put all their effort to new cheaper products.
I do worry whether they keep their initial promise and deliver motherboard upgrades - one with USB4v2. If not, this was just very expensive and moot undertaking.
1 Like
This is not my experience at all. Baldur’s Gate 3 stutters and gives about 40 FPS on the internal screen when plays on the dGPU, while my 3080 in a Thunderbolt 3 dock hits 60 FPS on a 4K screen.
Another comparison is that Final Fantasy XV runs at 47 FPS on my 2080 Ti desktop (with a 3950X) while it gets to 53 FPS on Framework 16 with a 3080 in a Thunderbolt 3 eGPU.
Those are worthwhile upgrades to me. I can keep Laptop 16 while getting rid of my (hopefully, less upgradeable!) Corsair One. And I both have a better GPU to play at home, and a competent dGPU to use outside home.
On the other hand, the Thunderbolt implementation gives such a good performance that I’m drawn away from OcuLink with its signal integrity issues and inability to hot un-plug.
all these cases are 4k. CPU performance doesn’t matter in 4k.
What is the client base that is:
A: Buying a $3,000 GPU.
B: Playing on a Laptop.
C: Choosing to frankenstein these together but not play at 4K.
The entire premise of playing on a laptop is itself an exercise in compromise..
1 Like
Ah, I see! So the competitive stuff at lower resolutions is more important for you?
Well i got a significant uplift, from usb4 6800xt to usb4 9070xt to oculink 9070xt. The USB4 eGPU Setup was alot more bandwith limited. My external Screen is a uwqhd 3440x1440 and from dGpU to any of the eGPUs the avg gain was more than double, but smotheness/stutterwise the dGPU and Oculink Setup are the best.
3 Likes
What is the client base that is:
A: Buying a $3,000 GPU.
B: Playing on a Laptop.
C: Choosing to frankenstein these together but not play at 4K.
This is kinda my use case: Reusing my existing desktop GPU for LAN parties I fly to.
Well, modify this as follows:
A: Have a desktop GPU more powerful than a 1060.
B: Be acceptable with a cold swap.
C: Choosing to not get a desktop PC
Oculink is realistically a extremely affordable $150 solution. Cheaper than a motherboard, even. For this price you get a m.2 to Oculink board, a Oculink cable, and a Oculink to PCIe slot board. Get yourself a power supply for like $75 (or reuse a old PC power supply) and you are good.
I think someone used a modified dual m.2 carrier for oculink, and just use a x4. It’s “not bad”.
2 Likes
Sure, it can obviously be modified to whatever hyper-specific requirements you have. One of the fortunate things is that the 7700S GPU module is substantially faster than a 1060. Somewhere between a Desktop 1080 and a 1080Ti.
For my use case (I’m playing at 4K), USB4 has been perfectly serviceable as an occasional gaming rig when my home Desktop isn’t feasible. When I’m hyper-mobile, I just use the 7700S GPU Module as it’s more than capable for the kind of gaming I’d do in only a laptop body on a laptop screen. As someone who’s on the move when they’re traveling, I value the hotswap ability of USB4 more than the performance loss at 4K.
That’s why I was curious as to what what different groups are serviced by this solution. The entire premise being discussed is that some sort of Expansion Bay module is needed, and that has struggled to come to market to have an Oculink alternative. If it does, I would be highly surprised if anyone is bringing a low volume accessory like that forward for less than $100. You’ll also need the interposer if you never got the GPU Shell for another $30.
At this point, it’s fairly safe to say that all the technology needed to bring this to market exists now, and its implementation is already very common in the enterprise industry with deep pockets and high volume. The question of whether or not this can come to Framework 16 is purely a matter of cost at this point. Someone could pay $20K and probably get a one-off commissioned / made. But no one is interested in paying that to get Oculink to their Framework 16. 
1 Like
It’s really expensive, though. Like, really expensive.
A Thunderbolt to PCIe PCB cost north of $150. Full enclosures are approaching $300. You can get the RX 7700S for that money.
What if I don’t buy the 7700S? It’s $400. I can buy a Oculink solution (for $150) and a $250 GPU. Or, better yet, save those $250 because I already have a GPU.
It’s basically half-price USB 4 / Thunderbolt 3.
Correct. But not quite. A simple PCB, somewhere in the $20 territory, to “adapt” the expansion bay signal into a Oculink port that you can easily plug and unplug is needed.
This PCB will just sit in the Expansion Bay “Shell”, which now also take the dual m.2 SSD carrier. Which I use. So its no longer just a shell.
more Oculink Expansion Bay Technical ramble
Why Framework don’t give you the option to include this from factory, no clue. I think this is great way to just eliminate the fan PCB and dedicated cable, but perhaps it’s cheaper that way.
If I really wanted to I can probably come up with this in 3 days. Take the mechanical from the dual SSD design (the github seems outdated) Framework publish all the pinouts, and Oculink definition is also out there. Though, I have a feeling there need to be capacitor and other signal-integrity bits. Those will take way longer.
You will need the GPU interposer, that you are not getting away with.
Or you can “adapt” one of the m.2 slots on the dual SSD carrier into a Oculink, but it’s x4, not x8. Which, oh right. Thunderbolt is only x4. Plenty of Chinese supplier make them in a bewildering variety for cursed hardware setups like “gaming on Google Meet”, or some half-disassembled laptop. Except Framework’s expansion bay shell have a built-in cover you can remove, so hopefully you can source one that dont need those madness.
But those wont come with signal repeaters.
Granted, Oculink introduce its own problems, like signal integrity (need signal repeaters), cold-plug only, etc. But it’s significantly cheaper, “faster”, and simpler. You can also hook up any other PCIe device and it should “just work”. You don’t need Thunderbolt / USB4. You dont even need the GPU to support being USB4 or Thunderbolt, because Oculink is not. The only problem is you might not be able to feed into the internal laptop display.
3 Likes
Hopefully it’s clear in some of these comments just how much of this is a discussion of compromises. There’s not a single right answer. We’d like to “have it all”, and we’re discussing why that’s not working out too well so far.
I’ll point out that directly to your point about expensive, the AOOSTAR AG02 is $220 and has both Oculink and USB4. My Framework 16 was like $2,800 going off of memory. This gets extremely subjective given the context because quite frankly I do not see $220 as even remotely “expensive” in the context of what we’re talking about in this thread. Which is the Framework 16. Even the most scrawny configuration is going to cost you over $1,500. This whole thing is not budget stuff. But again, it’s subjective, and I’m not going to tell people what that limit should be. If it’s expensive for you and others that deem it as such, that’s fine, put it on the list as a detractor for USB4.
If you don’t want the 7700S the best part is that it’s completely optional. You just take compromises from one place, and apply them to another. For me, with my travel schedule, the idea that I need to bring an external GPU Controller and GPU packed in my luggage and deployed every time I want to play a game is a complete non-starter. As in, if that was my only option, I wouldn’t have even bought this thing. But that said, I could totally see people who are buying this as “a desktop, but for some reason I don’t want a desktop” could totally use it in a Dock Only for Gaming or GPU-Compute design.
I’m honestly not going to comment too much around the Simple PCB for Oculink stuff because that’s honestly getting so far out of my wheelhouse of acceptance, that it’s for other people to scope / define. I personally would expect any system using an Oculink design that will leave the system to use a proper Retimer. “Repeater” is fuzzy nomenclature so I avoid it per the PCI-SIG. It’s going to either be a redriver or a retimer, and redriver’s in the PCIe 4 era are on their way out. But a retimer in low quantities are going to be approaching $50 for something like a TI DS160PT801. That’s also going to require external EEPROM’s, Firmware Management, and all the things Framework already does today with the retimers built into their laptops. Again, for those trying for ultra-budget methods willing to make those compromises, I hold no ill will towards them. I would not be interested.
In this thread alone, several people have claimed how easy producing this Oculink system will be, and yet 2 years later here we are. I have been using USB4 in the interim for over a year.
If you take all the things you care about, ignore the bits that you don’t, then it’s easy to take two standards, compare them, and declare one standard as “cheaper, faster, and simpler”. But ignoring those compromises as part of that declaration doesn’t guarantee there is enough mindshare in the market to make it successful. This is all a set of compromises.
What’s cheaper, significantly faster, and simpler than all of this? A Desktop. Yet for some reason or another that seems to be a compromise too far. For what it’s worth, I’d love to see Framework themselves bring an Oculink shell module with a proper retimer, bifurcation, and firmware management integrated into the rest of the work Framework already does there. 3 ports, 2 Oculink 4i, and 1 Oculink 8i port. If they charged $200-$250 for the whole Shell / Interposer / PCB, I’d buy it.
Unfortunately, Oculink comes so late in the cycle, we’ll only be able to rely on it for this generation. Things go to CopprLink for the future, but unfortunately the SFF-TA-1032 connector is 3 times the thickness of Oculink, so it’s going to be much more difficult to integrate in mobile designs. But until PCIe 5 GPUs go mainstream, we’ll generally accept the compromise of non-ideal SSD Storage access
5 Likes
Since when is the USB4 Board expensive? The adtlink ut3g is about 129€ from aliexpress and it works like a charme.
Thunderbolt is bullcrap when you have a native usb4 usb computer. And you don’t need not any usb4 compatibility from the gpu, which isnt even a thing. I switched to oculink as it adds further bandwith for my rx9070xt, but with anything less performant than an 6800xt you don’t loose any really noticable performance.
3 Likes
Now that’s not bad, considering a OcuLink 4i board with retimer (like the DEG1) also costs around 100€ (though it comes with a cable, not sure how expensive a proper USB4 cable is - plus cost of OcuLink 4i mod for the laptop).
Small correction to a great post, you can totally play on the internal screen using a dGPU over OcuLink. That’s the main way I play since the performance impact is minimal and the internal screen is the best I have for high-refreshrate, competitive play.
Which is where the additional bandwidth of OcuLink does come in handy, on USB4 you’ll see a much larger hit.
Also, people mention signal integrity as a main issue of OcuLink - I believe it’s the opposite. I’ve had absolutely no issue with my 0.5m OcuLink connection to my DEG-1 (which notably, DOES include a retimer, which you DO need from testing in both oculink threads). It’s rock solid, and feels like a desktop, which is not something I generally hear from USB4-eGPU users (seen reports of flakyness and microstutters as an additional protocol layer has to correct errors). That’s why I refer to my OcuLink setup as a dGPU, and a USB4 setup as an eGPU - they are in completely different classes in terms of stability and software support.
So what you’re after depends on your use case. If you travel more and also expect to play heavier games not suitable for an iGPU on the go, and have a bit more to spend, 7700S+USB4 eGPU makes perfect sense, though you miss some performance in desktop mode.
For me, playing less-demanding games using the iGPU on the go, and the performance-intensive ones requiring peak performance (with a used 6900XT and used PSU) at home, OcuLink is a cost-effective, beautiful setup.
1 Like
Well most of the hard part of usb4 is very similar to what a pcie redriver does (and you already payed for the one on the laptop side with the laptop).
Ultimately occulink is just better performance wise and can be done cheaper but it does not have hotplug which is a pretty big drawback.
I do hope we’ll eventually get something like modern expresscard, just a pure pcie (plus maybe an usb and pd aux channel if we want to get fancy) with hotplug on a connector. The handheld gaming device market is growing where something like this would be very useful so it may happen.
1 Like
The thing is, OcuLink COULD be hotplug. It just needs to be supported across the stack.
BIOS/FW, Kernel PCIe Driver, Graphics Driver, etc.
On Linux, PCIe driver does support hotplug. AMD Graphics drivers might as well, not entirely sure, though I’m sure it can be worked around by manually disabling the GPU before unplugging or something like that if it doesn’t already support it.
The main problem is the BIOS/FW (not entirely sure which one is responsible for this) on the FW16, it does not support even basic standard protocols for rescanning the PCIe Bus, which I’ve tried.
If the firmware is responsible for that, we might be able to fix that. If it’s the BIOS, as I’ve heard before, we’re probably out of luck, as even Framework doesn’t directly develop the BIOS, just license it from Insyde.