I don't anticipate anyone using Oculink on devices that would have PCIe3.0 but know that bandwidth would only be slightly better than Thunderbolt 3/4 in that arena. ALSO, older cards will see diminishing returns as well between the two. So something to be mindful of.
Why not just use m2 to full Pcie slot adapters which can be had for cheap. Not sure why go m2 to oculink. Oculink only out performs thunderbolt because it’s 4.0 m2 pcie slot.
@@Jimmys_TheBestCop I think the main reason is how easy it is to disconnect and reconnect as compared to a m.2 to PCIe solution - shutting the device down and unplugging the oculink cable is way more convenient than removing a m.2 connection. You get the easy disconnect and go that Thunderbolt provides, but you DON"T get the hotswap capability.
I made an oculink egpu based on yours and ETA prime’s design. It works wonderfully for my wm2, you have saved me from spending way more money for much more inferior thunderbolt egpus
This One Dock solution caters to Win Max 2 users since it has the most easily accessible M.2 2230 slot. But for other USB4 handheld devices or TB3/4 laptops, the thunderbolt eGPU solution still triumphs because of convenience.
I just wished Valve released the SteamDeck back in Q4 2019 alongside the RDNA-2 announcement. They could've stolen a lot of the "pandemic sales" from the Nintendo Switch. And it would've given them enough time to come up with the successor. Perhaps one built on AM5 with a powerful Zen4 CPU and a modest RDNA-3 iGPU. Then also release an eGPU Dock for it. Which basically Fast-Charges the unit, provides Active Cooling, speeds up the CPU, disables the iGPU, and runs the eGPU. Being consumer-friendly the dGPU in the unit would also be replaceable even with a behemoth sized card. It could just support AMD cards as they work as partners, but the community would be welcomed to make it compatible with Nvidia cards.
@@ekintekoa deck with a USB4 eGPU is probably the most user-friendly solution atm - would be such a perfect console honestly. Only costs what you put down and already works, gets more powerful as you save up for gpus. OCulink is totally blowing my mind as someone that just heard about it, and I’m already very happy on a Thunderbolt rig
@@TheCBoysDotCom Not the mention that it's viable. A modern 8-core CPU in the main unit is all that's needed for this decade. The quadcores have fun their course, and we're seeing a lot of instability, crashes, and bottlenecks. Even if Valve doesn't upgrade the hardware, the original SteamDeck as a "weak handheld" is still overpowered. Like the PSP of it's day, or even the retro Sega Nomad, simply more capable than the norm.
@@ekinteko so true. With my eGPU rig, I’m using a 6 core mobile cpu Intel 8th gen. Even with what I’m sure is a hardware bottleneck I enjoy good frame rates and very high settings, makes me all the more excited for what’ll happen as mobile chips get more powerful. AMD and Apple in particular given their power efficiency and good temps.
@@TheCBoysDotCom If we say Intel is 1-generation behind AMD, then AMD is 3-generations behind Apple. The Apple M2 Pro is a potent beast, despite it still being based on the ageing A14 chipset. Their next architecture is going to debut at end of this year, for the iPhone, and it will be massive improvement. This will trickle down to their iPads, and later the Macs. The only reason to not get your hopes up, is that Apple doesn't want to enter that space. There's antitrust. Besides they are the largest gaming corporation in history, because Apple makes the highest revenue and highest profits from Video Games. Yes, it's all from their AppStore but you wouldn't ruin a good thing for yourself would you? No, Apple isn't going to change. The silver lining is that Android follows Apple. So the upcoming QC 8g3 is going to be a beast. And we may see it in Windows tablets, laptops, and Mini PCs, and it might push big-AAA developers to transition from x86 to support ARM. So we might get a killer console eventually, not from Apple, not from Intel, but through the overall competition and likely from AMD and Qualcomm.
What I wish you would have done at 6:00 is do the comparison with TimeSpy or something like that, where it would be a performance-oriented metric first. But, I get that you are showing gaming, but I'd like to see what numbers you get without or without eGPU setup.
Always wondered about just having a custom cover to allow easy access to the M.2 slot, cool. That performance is super impressive and any difference vs desktop might just be from a weaker cpu instead of occulink.
good video! I myself have a 15s eq 5700u HP laptop and would like to connect an egpu to it. Which slot is best suited for this? the MPcie slot from the WiFi or the m2 slot from the SSD? And if the m2 slot is better for the egpu, can I use the m2 SSD again with an adapter on the mpcie slot? I would then get WiFi via USB. Greetings
Now the real question... can this be done to the GPD Win4? I somehow doubt that it has an M.2 slot that is convenient enough to get to, but I would be more than willing to pay someone to mod the shell and add a port for this. The Win4 and my current 3090 as an eGPU is EXACTLY what I have planned as a spare/mobile setup to my current Dr. Zaber Sentry 2.0 build that I have as a main PC.
The M.2 is right there when you remove the back; its right on the outside. But a little recessed under the cooler. Still very accessible. But if you gave up that slot for Oculink, you'd be limited to just booting from the MicroSD. Which you could but I am not sure if you should o.o That said, it should absolutely be possible.
@@IngwiePhoenix_nb I meant a second M.2 slot in the Win4. I wouldn't be willing to give up my M.2 boot drive. I'm wondering if it would be possible to integrate an Oculink connector into the Win4 in some other way if possible.
@@IngwiePhoenix_nb And that is exactly what I am afraid of. But if it IS possible, I am not afraid to literally throw money at whoever gets it figured out to mod my unit. The gains from using Oculink vs USB4 are mindblowing, and since I already planned on using an eGPU anyway, I believe that the gains would be work the effort.
You don’t need oculink. Get a m2 raid card which has multiple m2 slots. M2 adapter to handheld to m2 card. 1 slot for your nvme drive and another m2 adapter to full pcie slot. There are videos of people doing this to the steam deck. Oculink is only beating thunderbolt cause it is accessing the m2 slot which is pcie 4.0 and it’s faster then thunderbolt. Only difference is oculink has a nicer cable then m2 to full slot😊
I didn t get the part where you get an Oculink port on your laptop., i ve never seen an Oculink port on a laptop, you can barely find Oculink port on the brain new MiniPC models.
Thanks for these vids exploring new or more performant tech! Can you do an analysis of a lapwtop with camm2 ram? It's thinner and faster than soldered ram, it needs promoting so OEMs look to use it, that will get more sales from peeps like myself, I'm avoiding soldered ram Hope you can, if so, thanks
Phawx, stop showing us you've been hitting the gym 3:46 some of us still have to live with ourselves you know? 😄 Great video, I'll give it a day or two before FOMO sets in, thanks!
Usb4v2 and thunderbolt 5 are coming with more bandwidth than occulink! And you will have power delivery, ethernet, usb, and displayport tunelled via thunderbolt/usb4. Occulink is a good solution for eGPUs but thunderbolt will proliferate because tunnels the protocols that consumers need/care about. Perhaps occulink should consider doing the same.
Oculink is currently close to full desktop performance so waiting a year for thunderbolt 5 devices to come out and the buying a +$300 Thunderbolt 5 eGPU enclosure/adapter isn't really worth it.
@@AlexanderPavel Read the updated usb4v2 standard. Released in June 2023. Usb 4 v2 must support pcie 4.0x4 yes, but the standard has been updated to permit controllers that can tunnel pcie 5.0x4! It is explicitly stated in the updated docs. I think it’s possible that thunderbolt 5 may indeed be on pcie5 because intel is talking about 120gbps mode in asymmetrical mode. Pcie4.0x4 doesn’t have enough bandwidth to support 120 Gbps! But pcie 5.0x4 does!
@@qdaniele97 the problem with occulink is that it doesn’t have power delivery. So you will need a two cable solution when using occulink. Vs usb4/thunderbolt which can tunnel Ethernet pcie DisplayPort and power delivery over 1 cable. Also, I don’t believe that occulink supports hotplugging unlike thunderbolt, which does. But I will admit that occulink is a great invention/technology. I wish more pc motherboards had occulink as thunderbolt can be hit or miss depending on controller firmware, BIOS version, and other undocumented quirks.
I'm getting my GDP winMax 2 2023 in a couple of weeks, 100% going the Oculing eGPU route, I'd like to use my own power supply with no limits maybe with a 4090 though. I'll strap the card under my desk, no case (zip ties screwed into the underside of the desk).
How come it’s taking so long for this kind of tech to be normalized? Extending those PCIe lanes over a cable to enable good eGPU solutions seems like such a beneficial thing.
@@kovspbru It's definitely about data rate and has nothing to do with latency. Thunderbolt/USB 4 is only 4 lanes of PCIE 3.0 (40gbps) where as USB4 Gen 2 is 80mbps or 8 lanes of PCIE 3.0. If the card can't push enough data into the processor that's what causes the frame drop. The reason Oculink works so well is it's bandwidth is equivalent to 8 lanes of PCIE 3.0 of in thunderbolt terms 80mbps. Sorry if this is a rant, but latency has absolutely nothing to do with it as thunderbolt is PCIE based and therefore has little to no latency as it is. Hope this explains it.
@@JC7119 plz read his comment. PCI-E tunneling via TB's problem is not Bandwidth. PCI-e Tunneling via TB3 Controller in eGPU Box makes A LOT OF LATENCY while Oculink doesn't. Because Oculink is just External Cabling PCI-E, not Tunneling.
@@ITrueNewbieI if latency was the issue there would be perceived lag. The definition of latency is literally delay of data. If you wanna understand why an egpu can't seem to hit full speed on usb4/thunderbolt do yourself a huge favor and Google what happens when you run a graphics card on pcie 3.0 on 4x then 8x the 16x on a traditional motherboard with no tunneling or cable. It scales exactly 1:1 as the a data rate between pcie 3.0 4x and tb3/4 are identical.
I really need help. I just brought the Gigabyte gaming box 3080 for the Win max 2 but it just won't connect to thunderbolt 3 cable nor usb4 cable. I've tried reinstalling the usb 4 drives, the thunderbolt 3 drives, ive tried installing the AMD auto scan adreneline addition, resetting the WIn max 2 BIOS... nothing is working... all i can think of is reinstalling windows 11, or downgrading to windows 10. Some people have said this works for them. How do you get the egpu to work for yours?
I got a server where I wanted too connect GPUs in an external case. It has tons of oculink support on the mainboard for 24nvme drives at full X4, was searching for this for long. It's pcie so I don't get why this wasn't on the market yet.
Over the last few years there have been 2 or 3 laptops with a native oculink connector, which served PCIe 3.0x8. USB4/Thunderbolt 4 is simply too slow for modern eGPUs. OCuLink-2 can do PCIe 4.0x8, but no hardware maker seems interested in offering this for a gaming laptop. Why can't consumers get nice stuff? It's always reserved for servers…
So tempted to add an oculink port to my laptop now... I might even have space for a very clean install. What's the maximum cable length and would an internal cable to a port cause issues?
1m is generally the most I'd go. I don't know of any active cables that are longer. Internal m.2 to oculink is fine and is what is often used for laptops.
@@ThePhawx Thanks, I wonder if anyone has tested length? I just keep thinking about how PCIE is weirdly okay with long extension cables, it does seem 1M is listed max though.
don't forget, occulink is like buying a noose to hang someone, without the tree. What good is occulink if the laptop you want to use it on, or any future laptop coming out, doesn't support it??
does any one know if a rtx 3090 usb type c port can be used for power and data on gpd win max 2's usb c for up to a total of 8 lanes? i like to see if its even possible and if so by how much? or does just run like [kvm] design
@@post-leftluddite Yes, thunderbolt 4 is like a 4x PCIE 3.0 and the Oculink is 8x PCIE 3.0 (It's really a 4x PCIE 4.0). At least from what I know. Otherwise same rules as thundebolt, an external display connected directly to the GPU will get better FPS than on the WIN Max 2 monitor.
Hmm, never got eGPUs before but I have a winmax 2 and the ONLY reason I turn on my desktop PC at this point is for VR. I don't have the new one with a Oculink Port but I don't use the second SSD anyway so heck, I might as well and then just sell my desktop PC. All my desktop PC is for anyway is to hold my 3090 for VR, my winmax 2 plays everything else.
Yes, you can use a m.2 to OCuLink port (SFF-8612) adapter. Although that isn't shown in this video, I believe one is shown in Phawx's One Dock review. In this video, at 1:46 you can see a Win Max 2 with a modified access panel for the m.2 2230 SSD socket/bay, that modification being a cut-out through which an OCuLink cable (SFF-8611) is plugged into such an adapter.
Is this TB3 ReBAR performance hit specific to AMD CPUs or is it with intel CPUs too? Specifically with the 11th gen intel, win max 2021 and win3? Also is it specific to certain games Like 2077 and Metro Exodus? I noticed the Nvidia Inspector mentioned those specifically. I use a 3060Ti with the win max 2021, 1195g7, with ReBAR on in the bios. But if i'm leaving performance on the table, I'd want to know.
What are the, if any, disadvantages of disabling REBAR? This is my first time hearing of it and, to me, it seems like something that should never be enabled by default. Why does it exist?
I only vaguely understand it. Rebar is used for some backward compatibility with old versions of DirectX. The Intel ARC GPUs actually rely on it for DirectX 9 support I believe. Turning it off with an AMD or Nvidia GPU should be fine.
@@FlexibleToast REBAR is a memory sharing technology. It allows the CPU and GPU to share VRAM. Specifically it allows the CPU to have complete access to the GPU's memory buffer. Usually it's limited to a small fraction. What Intel is doing is using software to emulate DirectX9 similar to Linux using Proton as a compatibility layer because they don't have hardware capable of running DX9 natively. This is not related to REBAR.
Love the idea, and I'm glad it's getting more feasible, but it's not for me. I'm just not that big on an eGPU - I have a nice desktop for when I want to be stationary, and I really only use my GPD WM2 when I'm away from it. I do appreciate having thunderbolt, but not for an egpu - I just occasionally connect my WM2 to the dock that's normally reserved for my work macbook - then I get an awesome display and a full keyboard and mouse.
The only devices with a native oculink port are from GPD right now. So not a ton of support. But GPD is the only company making an eGPU with oculink to support those devices.
Personally this is useless, I use the internal screen with egpu 60fps at 800p in 99% of the games on win max 1 with a 6 feet cable.. I dont know why anyone would want to use an oculink+ an external screen when you could simply game stream at 4k on it without wires. Why buy a handheld and not use the screen built in and carry mini egpu if you travel around the world.
Don't care Been donw that road before with proprietary or not universal connections. Sticking with USB-C/thunderbolt connections for their ubiquity and besides USB-C 4.2 or whatever is purported to support 80Gbs data transfer speed and with p.d 3. Whatever will support over 100w charging. Thunderbolt 5 is supposedly going to have 160Gbs data transfer speed. I will inly jump on to a new connection IF it is a new Universal port/connection and only then when it is the clear replacement of USB-C and is adopted on to all devices that have used USB-C
Cool story bro. Also TB5 is 120Gbs which is still a bit slower than oculink. Go ahead and use your universal adapter that for some reason you can do the same for oculink. As long as you have a m.2 slot, it’s just as universal as thunderbolt lol
If it would be a Intel Platform we would have pcie5 that would be even more awesome.. pcie6 is around the corner .. imagine how massive it will be then!
I don't anticipate anyone using Oculink on devices that would have PCIe3.0 but know that bandwidth would only be slightly better than Thunderbolt 3/4 in that arena.
ALSO, older cards will see diminishing returns as well between the two. So something to be mindful of.
Does this offer much benefit over just pcie/nvme egpu Solutions? Like the ones all over AliExpress?
Why not just use m2 to full Pcie slot adapters which can be had for cheap. Not sure why go m2 to oculink. Oculink only out performs thunderbolt because it’s 4.0 m2 pcie slot.
@@Jimmys_TheBestCop I think the main reason is how easy it is to disconnect and reconnect as compared to a m.2 to PCIe solution - shutting the device down and unplugging the oculink cable is way more convenient than removing a m.2 connection. You get the easy disconnect and go that Thunderbolt provides, but you DON"T get the hotswap capability.
What cable do you recommend?
Link 2 buy
That transition from thumbnail to video really is smooth art , great job ❤️
Thanks!
It blew my mind. I loved it.
Reciver HestiA work stAtion
Other than the idiot arrow
i got an ad first up so didn’t see that
This is my setup at home, too. Literally my dream computer. Win max 2 with occulink & 3090. Portable power, desktop power. No compromises.
I made an oculink egpu based on yours and ETA prime’s design. It works wonderfully for my wm2, you have saved me from spending way more money for much more inferior thunderbolt egpus
This One Dock solution caters to Win Max 2 users since it has the most easily accessible M.2 2230 slot. But for other USB4 handheld devices or TB3/4 laptops, the thunderbolt eGPU solution still triumphs because of convenience.
It is usable with Steam Deck?
no and yes, you should run the OS from a USB drive since the M2 drive port is use to connect the eGPU
One dock has usb4/tb3 too
I want to see this as standard on the next iteration of the Steam Deck.
I just wished Valve released the SteamDeck back in Q4 2019 alongside the RDNA-2 announcement. They could've stolen a lot of the "pandemic sales" from the Nintendo Switch.
And it would've given them enough time to come up with the successor. Perhaps one built on AM5 with a powerful Zen4 CPU and a modest RDNA-3 iGPU. Then also release an eGPU Dock for it. Which basically Fast-Charges the unit, provides Active Cooling, speeds up the CPU, disables the iGPU, and runs the eGPU. Being consumer-friendly the dGPU in the unit would also be replaceable even with a behemoth sized card. It could just support AMD cards as they work as partners, but the community would be welcomed to make it compatible with Nvidia cards.
@@ekintekoa deck with a USB4 eGPU is probably the most user-friendly solution atm - would be such a perfect console honestly. Only costs what you put down and already works, gets more powerful as you save up for gpus. OCulink is totally blowing my mind as someone that just heard about it, and I’m already very happy on a Thunderbolt rig
@@TheCBoysDotCom Not the mention that it's viable. A modern 8-core CPU in the main unit is all that's needed for this decade. The quadcores have fun their course, and we're seeing a lot of instability, crashes, and bottlenecks.
Even if Valve doesn't upgrade the hardware, the original SteamDeck as a "weak handheld" is still overpowered. Like the PSP of it's day, or even the retro Sega Nomad, simply more capable than the norm.
@@ekinteko so true. With my eGPU rig, I’m using a 6 core mobile cpu Intel 8th gen. Even with what I’m sure is a hardware bottleneck I enjoy good frame rates and very high settings, makes me all the more excited for what’ll happen as mobile chips get more powerful. AMD and Apple in particular given their power efficiency and good temps.
@@TheCBoysDotCom If we say Intel is 1-generation behind AMD, then AMD is 3-generations behind Apple.
The Apple M2 Pro is a potent beast, despite it still being based on the ageing A14 chipset. Their next architecture is going to debut at end of this year, for the iPhone, and it will be massive improvement. This will trickle down to their iPads, and later the Macs.
The only reason to not get your hopes up, is that Apple doesn't want to enter that space. There's antitrust. Besides they are the largest gaming corporation in history, because Apple makes the highest revenue and highest profits from Video Games. Yes, it's all from their AppStore but you wouldn't ruin a good thing for yourself would you? No, Apple isn't going to change.
The silver lining is that Android follows Apple. So the upcoming QC 8g3 is going to be a beast. And we may see it in Windows tablets, laptops, and Mini PCs, and it might push big-AAA developers to transition from x86 to support ARM. So we might get a killer console eventually, not from Apple, not from Intel, but through the overall competition and likely from AMD and Qualcomm.
I was waiting for this sorcery for a longg time.
Thank you for your work! Great video. Please make video on detailed comparison 8x lanes oculink vs thunderbolt vs desktop 🙏
It would be nice if GPD offered OcuLink upgrade kit for original Max2 - after all it would slot into m2 2230 slot at the bottom.
What I wish you would have done at 6:00 is do the comparison with TimeSpy or something like that, where it would be a performance-oriented metric first. But, I get that you are showing gaming, but I'd like to see what numbers you get without or without eGPU setup.
Always wondered about just having a custom cover to allow easy access to the M.2 slot, cool. That performance is super impressive and any difference vs desktop might just be from a weaker cpu instead of occulink.
That's why he benchmarked with insanely taxing GPU settings at 4k. The weaker CPU is not bottlenecking at all.
good video! I myself have a 15s eq 5700u HP laptop and would like to connect an egpu to it. Which slot is best suited for this? the MPcie slot from the WiFi or the m2 slot from the SSD? And if the m2 slot is better for the egpu, can I use the m2 SSD again with an adapter on the mpcie slot? I would then get WiFi via USB. Greetings
now i wanna see oculink on a aya neo or steamdeck style device.
GPD massive in da house...
The Winmax 2 2023 has oculink right on the back of the case.
Now the real question... can this be done to the GPD Win4? I somehow doubt that it has an M.2 slot that is convenient enough to get to, but I would be more than willing to pay someone to mod the shell and add a port for this. The Win4 and my current 3090 as an eGPU is EXACTLY what I have planned as a spare/mobile setup to my current Dr. Zaber Sentry 2.0 build that I have as a main PC.
The M.2 is right there when you remove the back; its right on the outside. But a little recessed under the cooler. Still very accessible. But if you gave up that slot for Oculink, you'd be limited to just booting from the MicroSD. Which you could but I am not sure if you should o.o That said, it should absolutely be possible.
@@IngwiePhoenix_nb I meant a second M.2 slot in the Win4. I wouldn't be willing to give up my M.2 boot drive. I'm wondering if it would be possible to integrate an Oculink connector into the Win4 in some other way if possible.
@@ApacheVR-4 Technically, it should have sufficient PCIe lanes... but I am sure it would take some hardmodding to get this working. :/
@@IngwiePhoenix_nb And that is exactly what I am afraid of. But if it IS possible, I am not afraid to literally throw money at whoever gets it figured out to mod my unit. The gains from using Oculink vs USB4 are mindblowing, and since I already planned on using an eGPU anyway, I believe that the gains would be work the effort.
You don’t need oculink. Get a m2 raid card which has multiple m2 slots. M2 adapter to handheld to m2 card. 1 slot for your nvme drive and another m2 adapter to full pcie slot. There are videos of people doing this to the steam deck. Oculink is only beating thunderbolt cause it is accessing the m2 slot which is pcie 4.0 and it’s faster then thunderbolt. Only difference is oculink has a nicer cable then m2 to full slot😊
They need to put an out put like this on the need GPD Win!
I didn t get the part where you get an Oculink port on your laptop., i ve never seen an Oculink port on a laptop, you can barely find Oculink port on the brain new MiniPC models.
Why is this not more popular and well known and produced by the big companies?
Thanks for these vids exploring new or more performant tech!
Can you do an analysis of a lapwtop with camm2 ram?
It's thinner and faster than soldered ram, it needs promoting so OEMs look to use it, that will get more sales from peeps like myself, I'm avoiding soldered ram
Hope you can, if so, thanks
Hi Phawx, does the FPS go up, when Resolution Scaling is used? (FSR,DLSS)
Or is this part of the bandwidth?
Phawx, stop showing us you've been hitting the gym 3:46 some of us still have to live with ourselves you know? 😄 Great video, I'll give it a day or two before FOMO sets in, thanks!
The future is here(almost!), and it's amazing.
I'm curious how this test would do with a 3050, and where the best matched GPU to thunderbolt bandwidth would be. Can you do some experiments?
Usb4v2 and thunderbolt 5 are coming with more bandwidth than occulink! And you will have power delivery, ethernet, usb, and displayport tunelled via thunderbolt/usb4. Occulink is a good solution for eGPUs but thunderbolt will proliferate because tunnels the protocols that consumers need/care about. Perhaps occulink should consider doing the same.
Oculink is currently close to full desktop performance so waiting a year for thunderbolt 5 devices to come out and the buying a +$300 Thunderbolt 5 eGPU enclosure/adapter isn't really worth it.
Not really. Oculink 2 supports PCIe 4.0x8, while USB4 v2 will only support PCIe4.0x4.
Oculinkoculink is just pcie, you xan attach whatever you like to it, ethernet nic, gpus, usb hubs etc
@@AlexanderPavel Read the updated usb4v2 standard. Released in June 2023. Usb 4 v2 must support pcie 4.0x4 yes, but the standard has been updated to permit controllers that can tunnel pcie 5.0x4! It is explicitly stated in the updated docs.
I think it’s possible that thunderbolt 5 may indeed be on pcie5 because intel is talking about 120gbps mode in asymmetrical mode. Pcie4.0x4 doesn’t have enough bandwidth to support 120 Gbps! But pcie 5.0x4 does!
@@qdaniele97 the problem with occulink is that it doesn’t have power delivery. So you will need a two cable solution when using occulink. Vs usb4/thunderbolt which can tunnel Ethernet pcie DisplayPort and power delivery over 1 cable. Also, I don’t believe that occulink supports hotplugging unlike thunderbolt, which does.
But I will admit that occulink is a great invention/technology. I wish more pc motherboards had occulink as thunderbolt can be hit or miss depending on controller firmware, BIOS version, and other undocumented quirks.
Why did you use Thunderbolt 3 for testing instead of thunderbolt 4??????
could you put the card in an enclosure? are there any you'd recommend if so? also, which m.2 adapter did you use?
I'm getting my GDP winMax 2 2023 in a couple of weeks, 100% going the Oculing eGPU route, I'd like to use my own power supply with no limits maybe with a 4090 though. I'll strap the card under my desk, no case (zip ties screwed into the underside of the desk).
Great, but the Oculink is not available in all countries? In my country, I only see thunderbolt or USB so far.
the question in 2024 is : oculink or thunderbolt 5?
Seems it is oculink as tb5 is nowhere here yet.
just a head's up. Thunderbolt 5 will be released in 2023/24 with transfer speeds up 120gbs. Twice as fast as T bolt 3&4
How come it’s taking so long for this kind of tech to be normalized? Extending those PCIe lanes over a cable to enable good eGPU solutions seems like such a beneficial thing.
Where can you buy one of these? I’m only finding a taobao link, and that’s a lot of logistics to get something shipped to the us
Well this is aging extremely well, hopefully this port becomes more mainstream
I may missed it but the oculink + eGPU only works on Windows or it works with Linux as well?
Can you please post a link to the PCIe 4.0 adapter you used? Gen 4 adapters seem pretty much non-existent
Fair to say the GPD win4 7840u plus the G1 is gonna be pretty close to desktop performance?
how can i buy the M2 to oculink adapter designed for the GPD winmax 2 and the cable ?
Where did you get that single eGPU mount with space for PSU?
0:00 en efecto es cine 🚬🗿
You should share this with the Framework community, for the FW16.
This is why I can't wait for USB 4.0 Gen 2. 80gbps over a type-c cable.
I beleive it's not about speed, but about latency.
But if you're right - I will spend my money on eGPU (once more =( )
@@kovspbru It's definitely about data rate and has nothing to do with latency. Thunderbolt/USB 4 is only 4 lanes of PCIE 3.0 (40gbps) where as USB4 Gen 2 is 80mbps or 8 lanes of PCIE 3.0. If the card can't push enough data into the processor that's what causes the frame drop. The reason Oculink works so well is it's bandwidth is equivalent to 8 lanes of PCIE 3.0 of in thunderbolt terms 80mbps. Sorry if this is a rant, but latency has absolutely nothing to do with it as thunderbolt is PCIE based and therefore has little to no latency as it is. Hope this explains it.
@@JC7119omments like this is why I haven’t completely abandoned UA-cam comments
@@JC7119 plz read his comment. PCI-E tunneling via TB's problem is not Bandwidth. PCI-e Tunneling via TB3 Controller in eGPU Box makes A LOT OF LATENCY while Oculink doesn't. Because Oculink is just External Cabling PCI-E, not Tunneling.
@@ITrueNewbieI if latency was the issue there would be perceived lag. The definition of latency is literally delay of data. If you wanna understand why an egpu can't seem to hit full speed on usb4/thunderbolt do yourself a huge favor and Google what happens when you run a graphics card on pcie 3.0 on 4x then 8x the 16x on a traditional motherboard with no tunneling or cable. It scales exactly 1:1 as the a data rate between pcie 3.0 4x and tb3/4 are identical.
what Thunderbolt egpu dock this in the video the with the atx power supply can you link thanks in advance
as you have gpd win mini, how about a test with cheap amazon £35 oculink adapter ;)
I really need help. I just brought the Gigabyte gaming box 3080 for the Win max 2 but it just won't connect to thunderbolt 3 cable nor usb4 cable. I've tried reinstalling the usb 4 drives, the thunderbolt 3 drives, ive tried installing the AMD auto scan adreneline addition, resetting the WIn max 2 BIOS... nothing is working... all i can think of is reinstalling windows 11, or downgrading to windows 10. Some people have said this works for them.
How do you get the egpu to work for yours?
Thank you for an amazing video! Do you think such type of an eGPU enclosure could work with GPD Win4?
I honestly haven't looked, does the Win 4 have a secondary M.2 port? I would be kind of surprised if it does just based on its form factor.
My server has a couple of those on it. Epyc CPUs have so many PCIe lanes I think they just figured why not?
I got a server where I wanted too connect GPUs in an external case. It has tons of oculink support on the mainboard for 24nvme drives at full X4, was searching for this for long. It's pcie so I don't get why this wasn't on the market yet.
Over the last few years there have been 2 or 3 laptops with a native oculink connector, which served PCIe 3.0x8. USB4/Thunderbolt 4 is simply too slow for modern eGPUs. OCuLink-2 can do PCIe 4.0x8, but no hardware maker seems interested in offering this for a gaming laptop. Why can't consumers get nice stuff? It's always reserved for servers…
Which laptops are these?
Why use a 3090 for the bottom test with different REBAR? Compare like with like
Same for the Steam Deck please 😘
Will the Oculink be still better than the USB 5 ?
So tempted to add an oculink port to my laptop now... I might even have space for a very clean install. What's the maximum cable length and would an internal cable to a port cause issues?
1m is generally the most I'd go. I don't know of any active cables that are longer. Internal m.2 to oculink is fine and is what is often used for laptops.
@@ThePhawx Thanks, I wonder if anyone has tested length? I just keep thinking about how PCIE is weirdly okay with long extension cables, it does seem 1M is listed max though.
Any link for a oculink to tb 3 cable ? I can't find nowhere ? Please help !
don't forget, occulink is like buying a noose to hang someone, without the tree. What good is occulink if the laptop you want to use it on, or any future laptop coming out, doesn't support it??
This is really interesting. Is this Oculink 2 with 16gbps per lane?
Could this be used in a micro form factor PC?
Do you need disable REBAR with oculink?
Amazing!
Does the WIN 4 have Oculink?
please do more videos like this
does any one know if a rtx 3090 usb type c port can be used for power and data on gpd win max 2's usb c for up to a total of 8 lanes? i like to see if its even possible and if so by how much? or does just run like [kvm] design
Is it possible with steamdeck through m.2
Meanwhile, my new Mac Mini Pro is going to have PCIe 5 with Thunderbolt 5.
Does this work in the steam deck?
will this work on mac s?
So is occulink the same deal as connecting your egpu through an m.2 nvme socket which uses pcie 4x 3.0 or 4.0 ?
4.0
@@AMZZZMA So this oculink connection is using PCIe 4.0 x4? So it has double the bandwidth of Thunderbolt 3/4?
@@post-leftluddite Yes, thunderbolt 4 is like a 4x PCIE 3.0 and the Oculink is 8x PCIE 3.0 (It's really a 4x PCIE 4.0). At least from what I know. Otherwise same rules as thundebolt, an external display connected directly to the GPU will get better FPS than on the WIN Max 2 monitor.
So the winmax 2 has a second nvme slot?
Yes, that slot (m.2 2230) having its own dedicated access panel is one of its distinguishing features over every other handheld to date.
Hmm, never got eGPUs before but I have a winmax 2 and the ONLY reason I turn on my desktop PC at this point is for VR. I don't have the new one with a Oculink Port but I don't use the second SSD anyway so heck, I might as well and then just sell my desktop PC. All my desktop PC is for anyway is to hold my 3090 for VR, my winmax 2 plays everything else.
Can I use this in my 6800h laptop which doesn't have an oculink port with some adapter?
Yes, you can use a m.2 to OCuLink port (SFF-8612) adapter. Although that isn't shown in this video, I believe one is shown in Phawx's One Dock review. In this video, at 1:46 you can see a Win Max 2 with a modified access panel for the m.2 2230 SSD socket/bay, that modification being a cut-out through which an OCuLink cable (SFF-8611) is plugged into such an adapter.
@@chortlesg.3641 seems like a hassle 😭
Does anyone know if this oculink setup is able to run VR games like HL Alyx?
It depends on the graphics card not the Oculink connection
I wish we didn't have to use the Discord for all this stuff, its impossible to navigate :(
Yes it is, it's not user friendly at all
Is this TB3 ReBAR performance hit specific to AMD CPUs or is it with intel CPUs too?
Specifically with the 11th gen intel, win max 2021 and win3?
Also is it specific to certain games Like 2077 and Metro Exodus? I noticed the Nvidia Inspector mentioned those specifically.
I use a 3060Ti with the win max 2021, 1195g7, with ReBAR on in the bios. But if i'm leaving performance on the table, I'd want to know.
I'm in the same case of you, I had the i7 win max 2021 but with a 3070Ti TB egpu.
You say green line is rebar disabled, your graph text says rebar is enabled for green?!?!??!!
What are the, if any, disadvantages of disabling REBAR? This is my first time hearing of it and, to me, it seems like something that should never be enabled by default. Why does it exist?
I only vaguely understand it. Rebar is used for some backward compatibility with old versions of DirectX. The Intel ARC GPUs actually rely on it for DirectX 9 support I believe. Turning it off with an AMD or Nvidia GPU should be fine.
@@FlexibleToast REBAR is a memory sharing technology. It allows the CPU and GPU to share VRAM. Specifically it allows the CPU to have complete access to the GPU's memory buffer. Usually it's limited to a small fraction.
What Intel is doing is using software to emulate DirectX9 similar to Linux using Proton as a compatibility layer because they don't have hardware capable of running DX9 natively. This is not related to REBAR.
I can hear your monitor crying for a clean up
Isn't this just what Minisforum B550 already kinda does?
can someone dump the links so i dont have to join yet another discord server?
Love the idea, and I'm glad it's getting more feasible, but it's not for me. I'm just not that big on an eGPU - I have a nice desktop for when I want to be stationary, and I really only use my GPD WM2 when I'm away from it.
I do appreciate having thunderbolt, but not for an egpu - I just occasionally connect my WM2 to the dock that's normally reserved for my work macbook - then I get an awesome display and a full keyboard and mouse.
I mean why don't they just double up and add 2 thunderbolt cables? :P
wink wink GPD :D
Oculink 1.0 is already obsolete, Let's hope Oculink 2.0 has longer life and development.
Thunderbolt 5 coming next year should close the gap with twice the bandwidth of TB3/TB4/USB4
Wow!
But how many devices have a oculink port? It seems to rare now as of 2023. If that's only present on already gaming devices...
The only devices with a native oculink port are from GPD right now. So not a ton of support. But GPD is the only company making an eGPU with oculink to support those devices.
I feel like this connector will be dead in 2 years time. Given how bad USB 4 is right now, it will eventually improve thankfully.
Personally this is useless, I use the internal screen with egpu 60fps at 800p in 99% of the games on win max 1 with a 6 feet cable.. I dont know why anyone would want to use an oculink+ an external screen when you could simply game stream at 4k on it without wires. Why buy a handheld and not use the screen built in and carry mini egpu if you travel around the world.
how's the latency when streaming competitive games like COD?
Pcie hotplug
Don't care
Been donw that road before with proprietary or not universal connections.
Sticking with USB-C/thunderbolt connections for their ubiquity and besides
USB-C 4.2 or whatever is purported to support 80Gbs data transfer speed and with p.d 3. Whatever will support over 100w charging.
Thunderbolt 5 is supposedly going to have 160Gbs data transfer speed.
I will inly jump on to a new connection IF it is a new Universal port/connection and only then when it is the clear replacement of USB-C and is adopted on to all devices that have used USB-C
Cool story bro. Also TB5 is 120Gbs which is still a bit slower than oculink. Go ahead and use your universal adapter that for some reason you can do the same for oculink. As long as you have a m.2 slot, it’s just as universal as thunderbolt lol
Wink wink lol
what about buying a desktop instead.....
Portability and optional egpu when trynna game is looking alot better than a beast gaming PC that you can't bring anywhere
If it would be a Intel
Platform we would have pcie5 that would be even more awesome.. pcie6 is around the corner .. imagine how massive it will be then!
PCI card digonse card
Hecate