I'm not sure if it would be possible to add one to a framework laptop? I'm assuming it would have to be designed into the laptop from the manufacture and can't be an extension card.
You can buy the oculink m.2 adapter kit (if you have two m.2 slots, with a compatible one), with a dremel make a hole for the connector and buy a right angle oculink cable. The Oculink supports x8 but sadly most laptops the m.2 is x4 pci-e, and you are right the laptop must have a x8 oculink.
@@Terreos Maybe not this revision of the framework since the expansion cards are just USB4 (and through the chipset?) I think, but it would be a very cool addition to a next generation or something. Or maybe they can have some sort of internal header that connects to an expansion card since backwards compatibility is such a big design element of that laptop, but still seems awfully complicated to slim down enough for a laptop (mostly the part that connects to the expansion card)
@@javierolavarria in many laptop, usb ports are on a sister board. One can print a same size custom pcb (through pcbway). And with some fine soldering skills connect a female oculink port on that pcb and oculink cable wires directly to m.2 pcie board(removing the female oculink port and still saving the back cover of laptop). But the easiest way is doing with the dremel😅
@@rixyrohaizie well, then all companies sucks esp asus, why they didn't include it by default, enclosures will be more interesting for laptop users, because now you either use m2 slot and it looks terrible or use proprietary solutions like asus's, tb is not an option because it cheaper to buy laptop with 3060-3070 than buying dock that makes your 4080 into 3070
I think AM5 will be my last desktop socket. After that I expect APUs to get good enough to go fully mobile with handheld with adding external GPU for home usage.
Oculink really takes the WM2 to the next level. I've had similar jankiness with my WM2 and eGPU over Thunderbolt. Really looking forward to trying out the WM2 refresh with their new g1 eGPU over Oculink. I think that's going to be a great setup for my needs, as the RX 7600M XT should cover the games I enjoy at 1080p/60.
Never understood why oculink never got widely supported on laptops and docks made with enclosures to look nice and sleek. The performance difference is night and day vs crappy thunderbolt
thunderbolt is more convenient, supports daisy chain, several monitors, hubs, storage etc, you could add a hub with extra monitor, add storage in the next link and lastly add an egpu at the end, with oculink you could only add pcie expansions so you get to choose between storage or gpu or network connectivity like idk 2.5gb ethernet?
@@ShhTime I know I always get super frustrated by the gigabit ethernet ports on my laptops. I always say "ugggh, gigabit ethernet? that's only 10x faster than fastethernet, which is only 10x faster than ethernet! that's just too slow for me to transfer my weird anime tentacle porn from network storage to the jerkstation thinkpad." it's a daily occurrence at this point
I would like to see a more in depth test with a heavier gpu like a 6700xt or 3070ti Also a comparison between oculink and thunderbolt. But also compared to a desktop pc with a simulair cpu.
I think this will be same as for some tests already there where they limit pcie to 8x or 4x. Techpowerup did that for 4090 and the drop on X4 for 4.0 was around 6%. But seeing how this CPU handles more efficient cards would be handy.
Oculink eGPU port would be a very nice addition to every AMD-powered laptop & desktop motherboard starting from Ryzen 6000 & 7000 series onwards despite already having USB 4.0 / TB 3.0 ports. It would also be great as well if manufacturers can add a built in Oculink port into Ryzen 5000 series desktop & laptop motherboard since they can only support USB 3.2 ports.
It will be a nice additional for any laptop, intel too. Cause it's not faster for eGPU purpose than thunderbolt - and oculink 2 even faster - x8 lanes instead of x4
Would be nice to get a comparison vs a desktop system while using a more powerful GPU like a 3080. I'm sure there's still bandwidth limitations vs a direct PCIEx16 4.0 slot but I wonder how much of a difference there really is? Also, comparisons I've seen of EGPU's vs desktop systems always have a way better CPU in the desktop which throws off the results, but the 7840U would probably match up well with something like a 5800X on desktop. I'd love to have a mini-PC with a 7840U that I can take to/from work or when I travel, then connect it to a GPU at home for when I really want to game.
So, the Occulink is not a full x16 lanes, so that is your bottleneck. You can look up articles that compare using GPUs in x4/x8/x16 lanes to find the differences you will likely experience. M.2 is at most an x4 link like he is using here, and I think that is the Occulink is as well. Remember that the x4 PCI-E 5 is as fast as x16 PCI-E 3, so your going to find for most applications it is fine.
techpowerup do PCI Express scaling benchmarks every generation. TL;DR compared with 4.0x16, 3.0x4 drops 20% (simulating using an external display) and 3.0x8 / 4.0x4 is low single digit losses. The stutters are the main issue though, not observed in a simple average fps. I wish TPU did 0.1% / 1% lows in those tests. But 4.0x4 should be good. And in the not too distant 5.0x4 (current gen high end laptops do pcie5.0 to M.2) should be fantastic.
@@greebj Something to take note of, since I relooked at the spec for OcuLink, the first version of OcuLink is PCI-E 3.0 and the newest versions OcuLink 2.0 uses PCI-E 5.0. The likelihood is this is the older version, but I could be wrong. So going from OcuLink to Oculink 2.0 really will be a quadrupling of available bandwidth.
@@greebj Oculink 4i is PCIe 4.0x4 (64Gbps) which is equal to PCIe 3.0x16 because PCIe 4.0 quadruples bandwidth compared to PCIe 3.0. Based on techpowerup's testing a 3080 loses 4% performance with PCIe 3.0x16, so I want to see if that holds true for an oculink connection vs desktop. It should "theoretically", but I haven't seen anyone test it.
FOR ANYONE WANTING TO BUY THIS- PSA: I went through ETA PRIME's link and bought what I thought was the bundle- Price went up from roughly $38.xx to $48.xx (figures gotta capitalize on the marketing). MORE IMPORTANTLY: It's no longer a bundle for $38/$50. I only got the board, no cable or adapter. Some type of marketing shenanigans that Chinese companies are not known for ;) Edit: I think they just sold out of the bundle and removed that part of the listing. The cable I bought was also removed as an option and I assume it sold out.
That's amazing! Never knew about Oculink, but now this is something I'd really want to do w/ my next APU only laptop. I had a friend who had issues w/ external AMD cards on thunderbolt too. Pretty sure he spent $400 for the dock + the GPU all to waste.
maybe 3Dprint all the enclosures for all your EGPU hardware on all vids? maybe send us aluminum blueprints to make an aluminum carpenter do it, those workshops are pretty common on all cities.
Hi ETA would you consider including PCIE Rx and Tx in your overlay for eGPU testing? It's a standard option in CapFrameX to show realtime PCIe bandwidth, and I would be extremely interesting to see what's happening there on these different setups!
Please do more videos with this eGPU solution! I think this could be a really awesome solution for the future of mobile gaming/productivity with the ability to plug into an eGPU when at home to fully unlock performance. Exploring this further could be the future for most mid level users out there! Thanks for showing this and it would be great to have future videos on competing eGPU compatible products like the OneDock and others to see what works best.
Displays with modular GPU plugin cards is the future, plenty of room for cooling/heatsinks and a single cable from big display is fine for docking with wireless input/controllers.
Man I wish you tried a more powerful GPU that's heavily bottlenecked by TB, would be interesting to see how far OCulink can push it. Also would like to see performance using the internal display if at all possible. EDIT: not sure if you said it in the video but does it support hot-swap or do you have to reboot?
dang, you definitely showed me something new today. Super cool. Really appreciate your videos, they keep me up to date on the latest hardware. Builds like this push the envelope.
After the release of PS5 my friends were joking about PCs like "Yo, when PC2 is coming out?" and now I see that "PC2" is something like this - something small, but can be easily turned into a full-fledged desktop PC once in a dock. Whether that is a Steamdeck-like device, or like this one. Looking forward to improvements of the external GPUs compatibility as the number of lanes requires an improvement. I'm sure this will be real in a year.
For those who didn't know about Thunderbolt External GPU dock, it's NEVER running at 40 gbps as advertised. It's actually running at 22 gbps (sometimes even Less). If you don't believe it, go ahead and read the documentation of Thunderbolt PCIe portion for External Graphics Card. This applies to any GPU being used regardless it's NVidia or Radeon or Intel.
@@eivis13 nope it is not. Thunderbolt or USB4 External GPU is still based on PCIe Gen 3 not Gen 4. Even the latest Razer external GPU that was launched few months ago. Also I need to remind you that Gbps isn't the same as GBps (Gigabit vs GigaByte) Because PCIe Gen 3 X4 is 32 Gbps while X16 is advertised as 16 GBps - see? Gigabit vs GigaByte, marketing like always. 1 GigaByte = 8 Gigabit
For E-GPU with only 4 pci-express lines or thunderbolt I reccomend use GPU with more VRAM like 12Gb or 16Gb - more precached data can help with stutters while gpu have only 4 pci-ex 3.0 lines
The Thunderbolt issues being less important on nvidia is because they have far better texture compression algorithms,but with so much power,Even that can't save them
We need more handheld in general coming with the oculink, At least bigger ones, like the 8 inch aokzoe, onexplayer 2, aya neo pro 2. Best solution for people who want to make the handheld their full pc
Thanks for bringing attention to OcuLink. I wonder how CXL 3 (or beyond) technology would change the interconnect playground in the near future. I look forward to seeing it in action.
Hopefully the next Steam Deck have this Oculink port support. XG mobile connector in ROG Ally/Flow is dead in the water with the docks having skyhigh prices and can't even use your own GPU with it.
Occulink is a pretty good solution but I'll just wait for USB 4.2 as it purportedly supports up to 80GBs data transfer, higher power delivery and newest display port protocols. And is backwards compatible with USB 4 which apparently current cables will support that speed and increase USB 3 to 20GBs.
Careful about using just bandwidth numbers. Thunderbolt 4 is "80gbps" but it's actually bidirectional 40gbps, not a straight "2x 40gbps TB3"... theoretically it should remove the penalty from using internal display (that happens because the display signal is competing with upstream traffic) but won't improve external monitor display much at all.
I'm baffled that this isn't more popular. Desktop is expensive and static, a laptop is extremely mobile. Even with some performance hit, it is still the better option.
the fact that the oculink supports pcie gen4 is the differential factor here (I think) otherwise I cant think of a reason why the difference in performance if not for the bandwidth limitation of thunderbolt 3 enclosures like the one you used
Thunderbolt 3 (and 4) EGPU's also have super inconsistent performance for some reason, whereas Oculink seems to give more stable framerates from what I've seen.
Oculink is like a pure pcie adapter, thunderbolt have other bus encoding, hub features and power features. I not belive than you can connect a graphics card, and usb/network device, and another external monitor, in the same oculink bus at same time, like a thunderbolt.
the word missing in the replies below is the /overhead/ associated with the tb protocol; pcie is basically like using a pci riser in a desktop build, vs a daughterchip packaging the data to send to another daughterchip on the egpu which then unpacks it and lets the gpu process it (and possibly repeat the whole process if you're using the built in monitor on a device like this). also the complication with using another protocol encapsulating the signal adds to all kinds of additional issues vs basically just using the pci bus via oculink
oculink is just like pure pcie with different connector. just link xg mobile in asus. becouse it not using thunderbolt protocol. no bi-directional occur or overhead when displaying. external display is not needed rather than tb
I wish Oculink could output a PCIe x16 port. Thunderbolt 3 and 4 seem to not be great for eGPUs. I've had the Aorus external TB3 dock with a GTX1080 I believe and sometime it just wouldn't connect or would randomly disconnect from my Dell Inspiron 7391 laptop. I ended up selling both. But even if there's only the X4 port available, I feel like the PCIe plug (that kinda reminds me of PCMCIA a bit) will definitely be much more stable.
It needs to plug into a x16 port to be x16. Laptop dGPUs are all soldered and traced through the motherboard now, there is no physical port. We're lucky we have M.2 ports that are x4 to use for this. Thunderbolt maxes at x4 because that's the limit of the DMI link between the tbolt controllers and CPU but isn't currently scaling with PCIe generation like M.2 ports are. So long as your laptop has 2 of them so it can still have a storage drive......
You should make a dedicated video on the issue of thunderbolt egpus, comparing nvidia and amd. I think a lot of people like myself would appreciate this.
The best part of this setup is: you can carry arround and the PSU is only to the GPU,so you will have all the power it can deliver -30%, instead of an desktop where you have allpower (-) 30% (-)CPU+Motherboard+ssds
Nice! The Indiegogo campaign is offering a custom oculink-powered Radeon 7600M XT eGPU, which looks great on paper, but its costs USD$650. Do you guys think it'd be better to go for a DIY solution like ETA here, or is the plug-and-play worth the extra $$ ?
I really wish this got pick up by more laptop maker. My dream setup is sth akin to a 16inch laptop with a top spec cpu, a mid range gpu (sth like a 4050ti) and an oculink doc that have enough bandwith for a top tier gpu like a 4090.
Any Framework owners here? How do their expansion cards, they use an USB-C interface, connect to the system? Is it the interface directly connected to the pci-e lanea, or do they connect through a bus (thunderbolt/usb4)? If it's the first, an Oculink module for Framework would rather interesting also for their upcoming 16".
I had the same idea, but they use usb-c ports, they should build this port on the motherboard and add more PCI lines, yes other usb-c ports speed will decrease but GPU connection speed will be better
i'm confused - the point is to have an eGPU; the GPU will have whatever video out ports you need...? or USB 4 could handle basically static usb to dp...?
Intel need to release Thunderbolt Next quickly as the bandwidth increase is exactly what we need for eGPU. Likewise USB4 V2 should be out already. We are still waiting for USB-PD 240W to be widely adopted. Unfortunately I think we have to wait until Intel Lunar lake to get Thunderbolt Next (5) integrated in the SoC.
Thunderbolt 3 and Thunderbolt 4 are the same btw. (There are "um akshually's" that come along with this like the minimums for power/data etc move up but the top end of the standard is the same - Thunderbolt 3=thunderbolt 4)
Imagine being able to use USB-4 and Oculink together on the same dock to both power the handheld, and get 8 lanes of PCIe, though, unless you're running at 4K or higher, you're probably going to be held back by the CPU long before PCIe 4x is your bottleneck
I really wish they'd make a version without the controllers. Nothing else needs to change; just get rid of the controller. Re: Oculink, one issue vs. TB is that it's not a one-cable solution.
I don't understand why don't these machines support Video In? That way, whether you want to use Thunderbolt, USB4 or Oculink, you can play only on your handheld and not get hampered by half-bandwidth.
afaik it only has 1 M2 slot? If you try to use it you'd be limited to the 64GB eMMC version as that has the storage soldered and could use this for eGPU (and then has an storage issue)
It may have taken 10 years to achieve the next level or next big thing to bring handhelds back to the main scene again but watch another 10 years from now these handhelds will be the console it's already happening has been a thing for Nintendo since 2017.
We need more 15,6" laptops with an Oculink port :)
*any* laptops with an Oculink port.
I'm not sure if it would be possible to add one to a framework laptop? I'm assuming it would have to be designed into the laptop from the manufacture and can't be an extension card.
You can buy the oculink m.2 adapter kit (if you have two m.2 slots, with a compatible one), with a dremel make a hole for the connector and buy a right angle oculink cable. The Oculink supports x8 but sadly most laptops the m.2 is x4 pci-e, and you are right the laptop must have a x8 oculink.
@@Terreos Maybe not this revision of the framework since the expansion cards are just USB4 (and through the chipset?) I think, but it would be a very cool addition to a next generation or something. Or maybe they can have some sort of internal header that connects to an expansion card since backwards compatibility is such a big design element of that laptop, but still seems awfully complicated to slim down enough for a laptop (mostly the part that connects to the expansion card)
@@javierolavarria in many laptop, usb ports are on a sister board. One can print a same size custom pcb (through pcbway). And with some fine soldering skills connect a female oculink port on that pcb and oculink cable wires directly to m.2 pcie board(removing the female oculink port and still saving the back cover of laptop).
But the easiest way is doing with the dremel😅
This is what I’m talking about! All those extra ports and storage and now we have oculink on the back
Still more io than a mac in a smaller package
i wonder whats better, oculink or Asus' XG Mobile port?
@@megakoni1I think the XG port is better
@@megakoni1 asus xg pcie 8, oculink maybe x4 or x2
@@rixyrohaizie well, then all companies sucks esp asus, why they didn't include it by default, enclosures will be more interesting for laptop users, because now you either use m2 slot and it looks terrible or use proprietary solutions like asus's, tb is not an option because it cheaper to buy laptop with 3060-3070 than buying dock that makes your 4080 into 3070
I think AM5 will be my last desktop socket. After that I expect APUs to get good enough to go fully mobile with handheld with adding external GPU for home usage.
Yeah I'm waiting to upgrade to the next gen Zen, based on leaks we're seeing a huge uplift compared to this generation vs last generation.
@Great Cthulhu world of warcraft? I miss that game but don't have anyone else to play it with so I stopped years ago
Assuming that more manufacturers have eGPU capable ports and the eGPU market is large enough to encourage manufacturers to include eGPU ports.
I'm more than fine with am4
I'm more than fine with 1151v1, even if I haven't had 1200 at the moment.
Oculink really takes the WM2 to the next level. I've had similar jankiness with my WM2 and eGPU over Thunderbolt. Really looking forward to trying out the WM2 refresh with their new g1 eGPU over Oculink. I think that's going to be a great setup for my needs, as the RX 7600M XT should cover the games I enjoy at 1080p/60.
Never understood why oculink never got widely supported on laptops and docks made with enclosures to look nice and sleek. The performance difference is night and day vs crappy thunderbolt
Because apple
thunderbolt is more convenient, supports daisy chain, several monitors, hubs, storage etc, you could add a hub with extra monitor, add storage in the next link and lastly add an egpu at the end, with oculink you could only add pcie expansions so you get to choose between storage or gpu or network connectivity like idk 2.5gb ethernet?
@@ShhTime if Laptops have both oculink and TB/USB4 ports, you can still use a hub on the TB/USB4 port and connect the egpu on the occulink port.
@@ShhTime I know I always get super frustrated by the gigabit ethernet ports on my laptops. I always say "ugggh, gigabit ethernet? that's only 10x faster than fastethernet, which is only 10x faster than ethernet! that's just too slow for me to transfer my weird anime tentacle porn from network storage to the jerkstation thinkpad." it's a daily occurrence at this point
@@ShhTime I get it but like this device they could have both?? Also the razer enclosure could add the oculink with thunderbolt?? I
I absolutely love egpu set ups like this!!
Would have loved to see some of those tests done with the 3080 ti. Maybe a video that compares oculink performance to desktop performance.
You mean 4090? 🤡
This diy graphics dock info is gold bro
I would like to see a more in depth test with a heavier gpu like a 6700xt or 3070ti
Also a comparison between oculink and thunderbolt.
But also compared to a desktop pc with a simulair cpu.
I think this will be same as for some tests already there where they limit pcie to 8x or 4x. Techpowerup did that for 4090 and the drop on X4 for 4.0 was around 6%. But seeing how this CPU handles more efficient cards would be handy.
this is what i want to see too, i want to know what the equivelent desktop CPU would be.
Oculink eGPU port would be a very nice addition to every AMD-powered laptop & desktop motherboard starting from Ryzen 6000 & 7000 series onwards despite already having USB 4.0 / TB 3.0 ports. It would also be great as well if manufacturers can add a built in Oculink port into Ryzen 5000 series desktop & laptop motherboard since they can only support USB 3.2 ports.
It will be a nice additional for any laptop, intel too. Cause it's not faster for eGPU purpose than thunderbolt - and oculink 2 even faster - x8 lanes instead of x4
Would be nice to get a comparison vs a desktop system while using a more powerful GPU like a 3080. I'm sure there's still bandwidth limitations vs a direct PCIEx16 4.0 slot but I wonder how much of a difference there really is?
Also, comparisons I've seen of EGPU's vs desktop systems always have a way better CPU in the desktop which throws off the results, but the 7840U would probably match up well with something like a 5800X on desktop.
I'd love to have a mini-PC with a 7840U that I can take to/from work or when I travel, then connect it to a GPU at home for when I really want to game.
So, the Occulink is not a full x16 lanes, so that is your bottleneck. You can look up articles that compare using GPUs in x4/x8/x16 lanes to find the differences you will likely experience.
M.2 is at most an x4 link like he is using here, and I think that is the Occulink is as well.
Remember that the x4 PCI-E 5 is as fast as x16 PCI-E 3, so your going to find for most applications it is fine.
@@NdxtremePro It's PCIe 4.0x4 (64Gbps) which is equivalent to PCIe 3.0x16 since PCIe 4.0 quadruples bandwidth per lane vs PCIe 3.0.
techpowerup do PCI Express scaling benchmarks every generation. TL;DR compared with 4.0x16, 3.0x4 drops 20% (simulating using an external display) and 3.0x8 / 4.0x4 is low single digit losses.
The stutters are the main issue though, not observed in a simple average fps. I wish TPU did 0.1% / 1% lows in those tests. But 4.0x4 should be good.
And in the not too distant 5.0x4 (current gen high end laptops do pcie5.0 to M.2) should be fantastic.
@@greebj Something to take note of, since I relooked at the spec for OcuLink, the first version of OcuLink is PCI-E 3.0 and the newest versions OcuLink 2.0 uses PCI-E 5.0.
The likelihood is this is the older version, but I could be wrong.
So going from OcuLink to Oculink 2.0 really will be a quadrupling of available bandwidth.
@@greebj Oculink 4i is PCIe 4.0x4 (64Gbps) which is equal to PCIe 3.0x16 because PCIe 4.0 quadruples bandwidth compared to PCIe 3.0. Based on techpowerup's testing a 3080 loses 4% performance with PCIe 3.0x16, so I want to see if that holds true for an oculink connection vs desktop. It should "theoretically", but I haven't seen anyone test it.
Really appreciate this OcuLink showcase video 👍
Kindest regards.
FOR ANYONE WANTING TO BUY THIS- PSA: I went through ETA PRIME's link and bought what I thought was the bundle- Price went up from roughly $38.xx to $48.xx (figures gotta capitalize on the marketing). MORE IMPORTANTLY: It's no longer a bundle for $38/$50. I only got the board, no cable or adapter. Some type of marketing shenanigans that Chinese companies are not known for ;)
Edit: I think they just sold out of the bundle and removed that part of the listing. The cable I bought was also removed as an option and I assume it sold out.
That's amazing! Never knew about Oculink, but now this is something I'd really want to do w/ my next APU only laptop. I had a friend who had issues w/ external AMD cards on thunderbolt too. Pretty sure he spent $400 for the dock + the GPU all to waste.
maybe 3Dprint all the enclosures for all your EGPU hardware on all vids? maybe send us aluminum blueprints to make an aluminum carpenter do it, those workshops are pretty common on all cities.
I would love to see how well a higher-end GPU performs over oculink compared to TB/USB.
Hi ETA would you consider including PCIE Rx and Tx in your overlay for eGPU testing? It's a standard option in CapFrameX to show realtime PCIe bandwidth, and I would be extremely interesting to see what's happening there on these different setups!
I second this! Would love to see bandwidth speeds for pcie over oculink!
Please do more videos with this eGPU solution! I think this could be a really awesome solution for the future of mobile gaming/productivity with the ability to plug into an eGPU when at home to fully unlock performance. Exploring this further could be the future for most mid level users out there! Thanks for showing this and it would be great to have future videos on competing eGPU compatible products like the OneDock and others to see what works best.
That dock + that M.2 adaptor + one of those little SBCs that has an M.2 port = Your next video!
I really wish the Steam Deck had this capability. I'd love to dock it to an eGPU so it could server as a proper higher-power gaming PC when docked.
steam deck's cpu is not as capable as the 7840U
OK now it's time for a Latte Panda Sigma + Oculink build.
Displays with modular GPU plugin cards is the future, plenty of room for cooling/heatsinks and a single cable from big display is fine for docking with wireless input/controllers.
Looks Oculink will be the answer. According to Wikipedia, the fastest oculink to date is 4.0 x8, probably still only for servers though.
its pcie 4 x4 for this one
Man I wish you tried a more powerful GPU that's heavily bottlenecked by TB, would be interesting to see how far OCulink can push it. Also would like to see performance using the internal display if at all possible.
EDIT: not sure if you said it in the video but does it support hot-swap or do you have to reboot?
From what i heard you need to reboot on oculink, it is not hot swapable like thunderbolt
dang, you definitely showed me something new today. Super cool. Really appreciate your videos, they keep me up to date on the latest hardware. Builds like this push the envelope.
After the release of PS5 my friends were joking about PCs like "Yo, when PC2 is coming out?" and now I see that "PC2" is something like this - something small, but can be easily turned into a full-fledged desktop PC once in a dock. Whether that is a Steamdeck-like device, or like this one. Looking forward to improvements of the external GPUs compatibility as the number of lanes requires an improvement. I'm sure this will be real in a year.
It's already real, but portable handheld set ups are just limited by slow speeds when connecting via thunderbot 3/4.
ok you have to start linking the wallpapers you are getting for these thumbnails and videos because Im gonna need those too lol
I can't believe I wasn't sub'd. Incredibly helpful. Thanks for the video and for the helpful sponsor! I have to get a key for my dad lol
For those who didn't know about Thunderbolt External GPU dock, it's NEVER running at 40 gbps as advertised.
It's actually running at 22 gbps (sometimes even Less). If you don't believe it, go ahead and read the documentation of Thunderbolt PCIe portion for External Graphics Card. This applies to any GPU being used regardless it's NVidia or Radeon or Intel.
so that's roughly x11 on pcie 4.0, which is still more than x16 on pcie3.0.
@@eivis13 nope it is not.
Thunderbolt or USB4 External GPU is still based on PCIe Gen 3 not Gen 4. Even the latest Razer external GPU that was launched few months ago.
Also I need to remind you that Gbps isn't the same as GBps (Gigabit vs GigaByte)
Because PCIe Gen 3 X4 is 32 Gbps while X16 is advertised as 16 GBps - see? Gigabit vs GigaByte, marketing like always. 1 GigaByte = 8 Gigabit
@@ClayWheeler ok, my bad. So baisically we're comparing pcie 3
this oculink is the future!!
this is SOOOO sick, the pricing of this solution really sells it for me :D
For E-GPU with only 4 pci-express lines or thunderbolt I reccomend use GPU with more VRAM like 12Gb or 16Gb - more precached data can help with stutters while gpu have only 4 pci-ex 3.0 lines
I always learn so much watching these videos
The Thunderbolt issues being less important on nvidia is because they have far better texture compression algorithms,but with so much power,Even that can't save them
We need more handheld in general coming with the oculink, At least bigger ones, like the 8 inch aokzoe, onexplayer 2, aya neo pro 2. Best solution for people who want to make the handheld their full pc
Thanks for bringing attention to OcuLink. I wonder how CXL 3 (or beyond) technology would change the interconnect playground in the near future. I look forward to seeing it in action.
That's amazing and would have me rethink my position on egpu's and docked laptop use.
Hopefully the next Steam Deck have this Oculink port support. XG mobile connector in ROG Ally/Flow is dead in the water with the docks having skyhigh prices and can't even use your own GPU with it.
Great solution and awesome video, thanks!
Occulink is a pretty good solution
but I'll just wait for USB 4.2 as it purportedly supports up to 80GBs data transfer, higher power delivery and newest display port protocols.
And is backwards compatible with USB 4 which apparently current cables will support that speed and increase USB 3 to 20GBs.
does is backward compatible with the current usb4 ports? don't they need a new controller chips for that 80gbps?
i think USB 4.2 will have more delay,
if use Oculink will not, it's pcie Variants.
Careful about using just bandwidth numbers. Thunderbolt 4 is "80gbps" but it's actually bidirectional 40gbps, not a straight "2x 40gbps TB3"... theoretically it should remove the penalty from using internal display (that happens because the display signal is competing with upstream traffic) but won't improve external monitor display much at all.
@@greebj its still improvement over the 22gbps in the current thunderbolt eGPU
@@johnpp21 and still have display output to external rather laptop screen . mehh
Spiderman in PCVR.
Trust me.
I'm baffled that this isn't more popular. Desktop is expensive and static, a laptop is extremely mobile. Even with some performance hit, it is still the better option.
the fact that the oculink supports pcie gen4 is the differential factor here (I think) otherwise I cant think of a reason why the difference in performance if not for the bandwidth limitation of thunderbolt 3 enclosures like the one you used
Thunderbolt 3 (and 4) EGPU's also have super inconsistent performance for some reason, whereas Oculink seems to give more stable framerates from what I've seen.
Oculink is like a pure pcie adapter, thunderbolt have other bus encoding, hub features and power features. I not belive than you can connect a graphics card, and usb/network device, and another external monitor, in the same oculink bus at same time, like a thunderbolt.
the word missing in the replies below is the /overhead/ associated with the tb protocol; pcie is basically like using a pci riser in a desktop build, vs a daughterchip packaging the data to send to another daughterchip on the egpu which then unpacks it and lets the gpu process it (and possibly repeat the whole process if you're using the built in monitor on a device like this).
also the complication with using another protocol encapsulating the signal adds to all kinds of additional issues vs basically just using the pci bus via oculink
oculink is just like pure pcie with different connector. just link xg mobile in asus. becouse it not using thunderbolt protocol. no bi-directional occur or overhead when displaying. external display is not needed rather than tb
Im personally extremely exited about this
i just looked up the oculink. it's pretty damn cool and i'm glad it became a reality that way connectors like rog's xg mobile don't have to exist
this dude doin amazing thing
This is an interesting device to say the least.
You should have flipped power supply fan side for optimal airflow
I wish Oculink could output a PCIe x16 port. Thunderbolt 3 and 4 seem to not be great for eGPUs. I've had the Aorus external TB3 dock with a GTX1080 I believe and sometime it just wouldn't connect or would randomly disconnect from my Dell Inspiron 7391 laptop. I ended up selling both.
But even if there's only the X4 port available, I feel like the PCIe plug (that kinda reminds me of PCMCIA a bit) will definitely be much more stable.
It needs to plug into a x16 port to be x16. Laptop dGPUs are all soldered and traced through the motherboard now, there is no physical port. We're lucky we have M.2 ports that are x4 to use for this. Thunderbolt maxes at x4 because that's the limit of the DMI link between the tbolt controllers and CPU but isn't currently scaling with PCIe generation like M.2 ports are.
So long as your laptop has 2 of them so it can still have a storage drive......
oculink is pcie 4 x4. so the performance is near enough with pcie 3 x16
Hello @ETAPRIME , what is the name of that dynamic wallpaper that you always use ? Thanks in advance
You should make a dedicated video on the issue of thunderbolt egpus, comparing nvidia and amd. I think a lot of people like myself would appreciate this.
Could you get your hands on the OneDock Oculink dock? Would love to see your thoughts on it!
Nice, really interesting tech!
This is what nintendo needs to do with their next console.
Can you do a video of only b-roll with that tiny plant in the background? Its a great lil guy for co-hosting your videos.
Would be nice to see your 3080ti with Oculink vs USB4
usb4/tb 32GBPS while oculink 64GBPS 😂.
The best part of this setup is: you can carry arround and the PSU is only to the GPU,so you will have all the power it can deliver -30%, instead of an desktop where you have allpower (-) 30% (-)CPU+Motherboard+ssds
Such a small and convenient setup!
I do want to see that oculink eGPU on the new LattePanda
Nice! The Indiegogo campaign is offering a custom oculink-powered Radeon 7600M XT eGPU, which looks great on paper, but its costs USD$650. Do you guys think it'd be better to go for a DIY solution like ETA here, or is the plug-and-play worth the extra $$ ?
we need more mini pcs with oculink. we have less choice.
You had room on that PVC for a water pump/res combo, so you might as well water cool the GPU 😎
Can you install Linux on this? This is so powerful that I'm thinking this as a more portable laptop instead and be docked when just at home.
Fun fact: Asus XG port literally look like Oculink-2 (8 lines instead of 4 in v1) + Type-C for other peripheria
yes. but the price. hell noo😂
If you can, disable resizable BAR when using an AMD GPU over TB/USB4.
interesting. having seen this, I think I need an interface like this for handhelds or laptops just in case i'd need graphics card gaming in the future
Friendlyelec nanopc t6 is out !!! We can buy it now.
I really wish this got pick up by more laptop maker. My dream setup is sth akin to a 16inch laptop with a top spec cpu, a mid range gpu (sth like a 4050ti) and an oculink doc that have enough bandwith for a top tier gpu like a 4090.
I've been high on OCuLink for quite a while. Seems a far simpler, more elegant, and more reliable solution than Thunderbolt.
Kind of funny when the GPU is bigger than the device it's plugged into.
I wish a 2in1 14 or 13.3 inch laptop had this occulink port with 7840u cpu
Any Framework owners here? How do their expansion cards, they use an USB-C interface, connect to the system? Is it the interface directly connected to the pci-e lanea, or do they connect through a bus (thunderbolt/usb4)? If it's the first, an Oculink module for Framework would rather interesting also for their upcoming 16".
I had the same idea, but they use usb-c ports, they should build this port on the motherboard and add more PCI lines, yes other usb-c ports speed will decrease but GPU connection speed will be better
This is why im now considering aokzoe a1 pro, and ditching my 10th gen intel desktop, but using my 3080
I thought oculink was related to VR too lol
Try disabling resizable BAR when you test with thunderbolt/usb4. It has been causing problems.
5:11 It's much, much lower than half of the bandwith that is lost.
I hate the occuling, I want more USB-A and USB-C ports!
Thanks for the info and video! What is the maximum PSU you can have for such egpu environment?
I'd love to see 2 mini DisplayPorts on one of these
i'm confused - the point is to have an eGPU; the GPU will have whatever video out ports you need...? or USB 4 could handle basically static usb to dp...?
Intel need to release Thunderbolt Next quickly as the bandwidth increase is exactly what we need for eGPU. Likewise USB4 V2 should be out already. We are still waiting for USB-PD 240W to be widely adopted.
Unfortunately I think we have to wait until Intel Lunar lake to get Thunderbolt Next (5) integrated in the SoC.
Thunderbolt 3 and Thunderbolt 4 are the same btw. (There are "um akshually's" that come along with this like the minimums for power/data etc move up but the top end of the standard is the same - Thunderbolt 3=thunderbolt 4)
Imagine being able to use USB-4 and Oculink together on the same dock to both power the handheld, and get 8 lanes of PCIe, though, unless you're running at 4K or higher, you're probably going to be held back by the CPU long before PCIe 4x is your bottleneck
Please show temps in your videos, its useful.
hope more Oculynk port in handheld gaming PC and Laptop
Honestly this is a great preview to USB4x2 and that lovely 80gbps speed. My current desktop is probably my last desktop. Crazy....
Imagine if AMD made a custom SoC that had the 7840U CPU side. But, had something like an RX 6700 10GB for it's GPU side.
Would be cool to print a enclosure for it though.
oculink is avalilble on other pcs too You just need a m.2 port adapter
this is next level shit right here
I really wish they'd make a version without the controllers. Nothing else needs to change; just get rid of the controller.
Re: Oculink, one issue vs. TB is that it's not a one-cable solution.
Would be nice if all laptops had just a 16x PCIe slot at the back :D
just simple. why would give customer upgrade their own gpu. rather buy new ones from them😂
I don't understand why don't these machines support Video In? That way, whether you want to use Thunderbolt, USB4 or Oculink, you can play only on your handheld and not get hampered by half-bandwidth.
Wow I didnt know this port exists, its much better than TB, why is this not mass adopted? Seems to be impossible to find a laptop with it
its simple. why gave customer upgrade their own gpu. rather than buy new one laptop from them 😂
1440p in FSR performance mean upscale from 720p - meh
FSR or DLSS performance test it's okay for iGPU or portable system, but not for dGpu
Can we get a video on how to setup the M.2 to Oculink in the Steam Deck? I think that would be useful.
afaik it only has 1 M2 slot? If you try to use it you'd be limited to the 64GB eMMC version as that has the storage soldered and could use this for eGPU (and then has an storage issue)
@@Michaelmaertzdorf I have seen some run the SteamOS on the SSD. Definitely not the best way to run it, but should make a fun video.
@@Michaelmaertzdorf the emmc module is on an m.2 board, he'd be stuck with the SD card
@@maverickvgc4220 not if you add a hub to the usb c and run the os from a ssd connected to the hub
When will you do review of RX 7900 XT?
It may have taken 10 years to achieve the next level or next big thing to bring handhelds back to the main scene again but watch another 10 years from now these handhelds will be the console it's already happening has been a thing for Nintendo since 2017.
Palit GeForce RTX 4060 Ti StormX is very small and single fan. Imagine That combination and a Flex 1u psu. Half the size of any E-GPU dock!
What level of graphics card would max out the effectiveness?
I don’t get it, why other companies do not use this port? Great video!
its just simple. why need gave customer upgrade their own gpu. rather just buy new laptop from them. 💲💲