So we may be offbase here thinking we want a redriver. We may actually want a retimer.
In fact, since retimers participate in the PCIE protocols rather than just passing them on, they look like they might be the solution for the OCulink 4x and NVME expansion cards as participating in the protocol might allow for properly splitting lanes. I remember there was a problem regarding that earlier…
@Josh_Cook could this help your 4x OCulink 4x NVME cards and become the standard for both 4x and 8x cards?
I haven’t kept up to date so I apologize in advanced, but I’m always eager to learn more. Could someone tell me what the term “redriver” means and how it applies to this thread?
The “redriver” or “retimer” is basically to clean up the signals for PCI-e and allow longer cable length or for it to function properly at all. It is yet to be seen as far as I know whether it will be required for this setup.
I believe he’s currently prototyping to see how much that all matters for this particular purpose along with proving his design functions at all. If he’s where I think he is with it I’d be surprised if it was shipping out this year. Considering the low sales/production volume and reasonable prices I’m assuming this isn’t really a primary job so he probably doesn’t have infinite time to spend on it.
Prices will be increased in 24hrs to accommodate higher component prices, both models will be increased to $90AUD.
Still working on the designs, due to a couple issues that came up. I will hopefully send out another update soon, but from the 31st of October to the 14th of November I wont be active at all
I’d be glad to increase what I paid once they’re shipping out if it costs more than originally intended, especially with how it seems you’re putting in quite a bit to verify it’s a quality product once it’s complete. Just let us know what the plan is there.
Hey man, I really love the work you are doing. I have seen you’re planning to take a bit of a break, which is always very important and I’m glad you’re doing that, even if it is not for the most ideal of circumstances. I hope you get some rest.
Anyway, I was just curious about one thing but I think others will be able to answer it for me: So, I am planning on buying a FW16 for college but wanted to use an eGPU for the limited gaming I do. Anyway, because we have to modify our non-GPU expansion bay module, would we need to buy a second module for non-GPU usage? Like, would I be able to just unplug the Oculink cord and use it as a regular laptop for my regular day? I only ask because I haven’t seen any pictures (which I understand with all of the documentation issues), making it hard to visualize. Thanks for any help anyone can provide!
For this use case, it would be possible. But when you unplug it, you have to shut down the laptop first before unplugging. Unplugging an OCuLink connector while the laptop is still on is basically like unplugging a graphics card from a motherboard while the PC is running, so don’t do that.
Aside from that, the OCuLink connector also has a limited amount how many times you can unplug and replug it. I don’t know exactly how much it is, but it’s not the same amount as a USB-C connector.
I am unsure about the last part, so someone might need to correct me on that.
Regarding my prior comment about disabling the M.2 socket on the 4x + NVMe board, I have decided to have 2 different firmware versions, with the version that will ship with it will have the M.2 socket disabled, but the DP input enabled, with the other version having the m.2 enabled, and the DP disabled, still working out the exact procedure for reprogramming, but the official modules do support firmware upgrading over i2c, so it should be similar.
Okay. At this rate then do you think it would be better for me to just deal with USB 4 until thunderbolt 5 comes out? I have a 3060ti but I have had trouble finding estimates for the performance drop. Sorry I know this is not exactly pertinent to this discussion and if you don’t know the answer then I would completely understand.
My experience with thunderbolt/usb4 was disappointing. The performance drop on a 3070ti was noticeable - it was around a 30% fps drop compared to my normal desktop (it was hard to tell exactly because I couldn’t use the same CPU on both setups, but neither rig was using 100% cpu resources while I was benchmarking). You don’t get the full 32gbit/s throughput (pcie 3.0 x4). I measured pcie speed to be right around 19gbit/s using the 3DMark pcie bandwidth test. Oculink, on the other hand, gets you pretty close to the full 64gbit/s of pcie 4.0 x4.
It matters a lot more on higher end hardware. I’m wanting to use a 7900xtx and I’m pretty confident that USB4 will bottleneck it. I’m hoping occulink 8x will give me similar to desktop performance.
It looks like the 8x cards won’t be available in the next couple of weeks, but I still need to put together a mobile solution in the meantime.
Can someone run me through the 16" laptop config I’ll need for the 8x card setup, and something that I can use now with my graphics card until the 8x card is ready?
yeah. Thunderbolt will work and plug and play for now. just the performance will be bad tho. I have a 7900xtx and it performs just a tiny bit better than 7700s because of the bandwidth bottleneck…
I don’t mind waiting longer if needed. Take your time…