I think the most of the performance loss in rtx 2060 come From the fact the card only supports pcie gen 3 while the oculink support pcie gen 4 so the rtx 4090
@@trysometech52 what about rtx3080ti? i got acer helios 300 3060 and 12700H, and i'm suffe playing forbiden west with FHD, i dont want to build a new rig. Just want to buy a 3080ti and use m2 to oculink, cut through a small hole under the laptop, whenever i travel far, i still bring my laptop on the go. and when i sit on sofa i plug it in 3080ti with 4k tivi. back to the topic, would oclink have performance loss with 3080ti?
Of course the native direct connection won the benchmark. However I'm surprised by how significantly reduced the performance gap between it vs the Oculink-connected 4090. It's a lot better than the thunderbolt eGPU version, and even the XG Mobile with a 150W RTX 4090 Laptp version. I wish more laptop brings us this connection, and more manufacturers make more baseline oculink eGPU chassis.
@@icaroleal7351 that doesn't matter. If you connect 4090 via Thunderbolt - the bandwidth will be the same as 2060, because Thunderbolt has PCIe 4x3.0 connection for PCIe devices like eGPU :)
Oculink port can carry a pcie 5.0 signal without major problems... But there isn't a pcie 5 GPU or adapter yet. Future egpu adapters wouldn't have performance lost, on pcie 5.0 x4.
did you try checking the bandwidth with cudaz? in my testing, it seems aida64 and cudaz have different results cudaz shows higher bandwidth in *MiB/s* than aida64 *MB/s* just a weird thing
One last thing - when the egpu setup is complete (GPU+Adapter +PSU ) and you plugin into the mini PC / oculink device via Oculink port: 1. Do you Disable the iGPU before / or after ? Or you just choosed the Nvidia GPU whe available in games / aplications ? 2. Did you had any issue such as error 43 and had to use a script ? 3. Do you just install Nvidia drivers alongside AMD drivers or DDU the AMD drivers in favor of Nvidia drivers ? Because I'm planning to test myself with my Minisforum UM 780XTX and an RTX 4070 TI Super (preety excited btw) Awesome video, just saw that there's a much better looking sleek newer GPU doc with Oculink by Minisforum called the DEG1 - around 100 euro I think ?
1. No, eGPU works by default 2. Yes 3. Just installed Nvidia drivers and it worked fine on my handheld Yeah I saw that dock, looks awesome :) But together that dock with mini PC will be like a disassembled SFF sized PC
Just ordered the exact same model from amazon... I'll see how it works... if it works maybe I'll try it with some pcie-cards with dual oculink ports (I can't find any with just 1 oculink port), ideally I don't want to waste an M.2 slot.
One big question that I have: what if you were to plug in the adapter over one of the M.2 slots instead? How much performance would you lose from that? Other than that, you video was awesome and super informative, keep up the good work!
Thanks! I'm not sure I understand your question. I had it connected via an m.2. adapter, but it also PCIe 4x4.0, so it won't affect performance compared to Oculink.
I can't wait for a true 16x connection that all Laptops support. Perfect for on the go and @home, just plugin a GPU for max performance... Well: a guy can dream, though. A guy can dream... 😥
what Pc do you have in this video ? i see multiple graphic cards Pcie ports/slots.... which could be optimal for building/making/creating Ai LLMs.... let alone optimal for running Ai like auto1111 or comfiyui or running Llama LLMs etc
Note that in Thunderbolt 5 or Thunderbolt 4 you don't have full bandwidth only for GPU ... Where in Oculink you have 100% for GPU (or you can connect storage :))
Curious as to why the oculink have lower tdp? Almost 40-50w different, that could be where the performance gone. Also it would be great to also show GPU clock speed since it could suggest if it’s run at lower speed as well.
@welly4986 Sorry, how do you think if it is an oculink adapter for a SSD M.2 PCI-E 4.0 x4 slot on motherboard, results will be the same? I don't have an oculink in my laptop, but i have Intel core 12700h with 2 ssd 4.0 slots and i want maximum usage of my graphic card. Thanks!
Awesome video thanks dude, just wondering about my mini pc it has 4070 12g and I want to use egpu 4090 connect them using USB c from 4090 to DP mini pc is it going to work ? I have no idea how to connect that I have DP, HDMI and USB 3.2 ports
Was that D4 Result with or without recently added Ray Tracing? I'm assuming with, as my 4090 nets me around 100-120 @ 3440 x 1440, but I'd like to confirm.
Thats actually not that much less, surprising makes you wonder how much bandwidth is actually needed for these cards in reality. Perhaps it would be different if it was running at 1440p or 1080p.
In any case, this is the best that is now for those who want to play on portable devices from time to time, or do not want a full-size PC, or simply change devices often, and the video card remains, or during the transition period, when the entire amount for a global upgrade is not available.
Hi ive watched your video about oculink and ordered the same one like you but im having game crash after setup, i was using a thunderbolt 3 egpu before with same setup as oculink with no problem, using a mini pc but oculink is giving problems with DXGI_ERROR_DEVICE_HUNG (0x887a0006). Do you have anymore information on the oculink setup please.
Hey, not really sure that if it's eGPU related problem. I did some googling and it seems that there could be a lot of reasons, including XMP profile on RAM in the BIOS settings. Do you face this problem only with some specific game or it's a global issue? I never had this error so I am not sure that I could help you man
I'm thinking of using this to power a 1650 and connect that to my laptop. My job requires some travel so this sounds like a good alternative and is small enough to carry around. Or does anyone have a recommendation for a small enclosure that supplies 75W?
El vídeo muestra que hoy en día, ya no hay diferencia entre una laptop y una PC de escritorio. Ya estamos en abril de 2024 Y que bien, me alegra que avance a pasos agigantados la tecnología
Was cyberpunk 2077 running in path tracing mode? If not on the 4090 could you please test that. Also can you test Control with RT at 4K and Alan Wake 2 with full Ray Tracing at 4K with as high of settings as you can run? I spy a OneXplayer X1, could you rerun the tests including that? I'd appreciate it. I'm considering the ONEXPLAYER X1 with 64GB of ram myself.
I sold my 4090 so I can’t run those tests now, but in a week or two I’ll release my review on OneXplater X1. Man…that device has a lot of troubles. Avoid it at all costs.
@@trysometech52 They've released new bios updates recently in the last 2 weeks or so. You have you get from their discord, which is shady and stupid. There's a video of the trick required that you need to do to update it. I'll be watching your review either way. I don't like how the controller module to link them together and use them via bluetooth is still a "pre-order" thing. A channel called Barely Average Dude and he didn't like the controller. I'm interested in the tablet style function more than the controllers as I would use a TB hub and KB and mouse. If TB5 wasn't coming out at the end of the year I would have purchased the X1, but considering it will be faster than current OCulink devices I'll used that as I want a portable that can give me at least 8x PCI-e 4.0 and TB5 will do that, getting the EGPU bay for it will be a wait. Intel already has a demo of it on their yt channel.
I’ve updated BIOS two times already, there is something more than just a software. Also AMD version of X1 will be released soon, but even that will not fix some other problems. Isn’t Thunderbolt 5 also has 4x4.0? I think performance will be similar, but with hot swap and other features
@@trysometech52 There's a 120 Gbps option that uses PCI-e 4.0 x8. The standard mode is the x4 same as prior TB, but with PCI-e 4.0. Since the tube hates links just search for thunderbolt 5 120Gbps and Intel has videos on it and there's plenty of articles on it.
I think that’s just a marketing material. In reality TB3/4 is not 40gbps for eGPU connection (I have a video on this topic). Also they claimed x2 performance, which is exact difference between PCIe3 vs 4. So technically they didn’t lie.
@@trysometech52 it would be interesting to try a gen 5 m.2 slot and see the difference if any, because even at the x4 bus width, should be faster on a gen 5 slot. pci express 4.0x4 vs 5.0x4
that egpu adapter (beast) is the cheapest one you can buy. if it has the wifi m.2 connector pcie x1 putting a rtx 4090 on it is a joke. the nvme m.2 connection is at least pcie x4. get an rg43s
I think the most of the performance loss in rtx 2060 come From the fact the card only supports pcie gen 3 while the oculink support pcie gen 4 so the rtx 4090
You're absolutely right! I forgot about that
Have tested or just guessing
@@jussikankinen9409 yes I tested
@@trysometech52 what about rtx3080ti? i got acer helios 300 3060 and 12700H, and i'm suffe playing forbiden west with FHD, i dont want to build a new rig. Just want to buy a 3080ti and use m2 to oculink, cut through a small hole under the laptop, whenever i travel far, i still bring my laptop on the go. and when i sit on sofa i plug it in 3080ti with 4k tivi.
back to the topic, would oclink have performance loss with 3080ti?
@@TungNguyen-rg1sc apart from CPU bottleneck in some games, you should be fine.
I expected much more loss in performance, at least 30%. But this is cool. Keep it up my guy.
Saw a comment you posted on his vid a year ago when he had 17 subs, you stuck with him😎
Awesome video. I'm glad I bought an Oculink board with a metal support for the GPU to stand straight. Glad to see the review of this one.
which one did you get? I saw one with a big black base and swapable brackets for PSU but the ETA was too long...
Of course the native direct connection won the benchmark.
However I'm surprised by how significantly reduced the performance gap between it vs the Oculink-connected 4090.
It's a lot better than the thunderbolt eGPU version, and even the XG Mobile with a 150W RTX 4090 Laptp version.
I wish more laptop brings us this connection, and more manufacturers make more baseline oculink eGPU chassis.
Thank you. This was an awesome video showing the value of Oculink!
Please try out the new Adt-Link R43UH pcie 4.0 with a much stronger gpu. Including all the software process as well.
Oops, comments were accidently disabled D:
Can you also make a video when it's connected through thunderbolt?
@@ChaosA1T If you hadn't skipped to the beginning of the video, you'd know that I already have it :)
Hey bro. The RTX2060 is pcie 3.0. The bandwich is the half compared ti RTX4090. 6:31
@@icaroleal7351 that doesn't matter. If you connect 4090 via Thunderbolt - the bandwidth will be the same as 2060, because Thunderbolt has PCIe 4x3.0 connection for PCIe devices like eGPU :)
I mean tree at once @@trysometech52
Oculink port can carry a pcie 5.0 signal without major problems... But there isn't a pcie 5 GPU or adapter yet. Future egpu adapters wouldn't have performance lost, on pcie 5.0 x4.
did you try checking the bandwidth with cudaz?
in my testing, it seems aida64 and cudaz have different results
cudaz shows higher bandwidth in *MiB/s* than aida64 *MB/s*
just a weird thing
Nice i was looking for that. Love from Brasil.
They came out with a new one that can hold the big cards
OCuP4V2 OCuLink GPU Dock w ReDriver Chip
Yeah this one looks good, I got their Thunderbolt variant. And seems like I can fit it in the same case
1080p would also be interesting here it's where we see the bandwidth limitations more
Thanks for the result.
You can order a board and make a platform for it. Way cheaper than kits. Amazon has the better choices.
One last thing - when the egpu setup is complete (GPU+Adapter +PSU ) and you plugin into the mini PC / oculink device via Oculink port:
1. Do you Disable the iGPU before / or after ? Or you just choosed the Nvidia GPU whe available in games / aplications ?
2. Did you had any issue such as error 43 and had to use a script ?
3. Do you just install Nvidia drivers alongside AMD drivers or DDU the AMD drivers in favor of Nvidia drivers ?
Because I'm planning to test myself with my Minisforum UM 780XTX and an RTX 4070 TI Super (preety excited btw)
Awesome video, just saw that there's a much better looking sleek newer GPU doc with Oculink by Minisforum called the DEG1 - around 100 euro I think ?
1. No, eGPU works by default
2. Yes
3. Just installed Nvidia drivers and it worked fine on my handheld
Yeah I saw that dock, looks awesome :) But together that dock with mini PC will be like a disassembled SFF sized PC
What is the performance loss on a desktop? For the purpose of freeing up pcie slots
Just ordered the exact same model from amazon... I'll see how it works... if it works maybe I'll try it with some pcie-cards with dual oculink ports (I can't find any with just 1 oculink port), ideally I don't want to waste an M.2 slot.
did some tests with furmark, chaos vantage, vray gpu, performance loss was around 7% only
One big question that I have: what if you were to plug in the adapter over one of the M.2 slots instead? How much performance would you lose from that?
Other than that, you video was awesome and super informative, keep up the good work!
Thanks! I'm not sure I understand your question. I had it connected via an m.2. adapter, but it also PCIe 4x4.0, so it won't affect performance compared to Oculink.
Hej hej, where I can buy this dell adapter ? I see , )) thank you for this video.
And where I can buy this cables from 6 pin to videocard pins ?
it's amazing tho. the value of the oculink is great.
Good job! Thanks for video!
I can't wait for a true 16x connection that all Laptops support. Perfect for on the go and @home, just plugin a GPU for max performance...
Well: a guy can dream, though. A guy can dream... 😥
what Pc do you have in this video ? i see multiple graphic cards Pcie ports/slots.... which could be optimal for building/making/creating Ai LLMs.... let alone optimal for running Ai like auto1111 or comfiyui or running Llama LLMs etc
Whats the backround song you've used during the Benchmark section ?
oh dear this helped me with a huge decession
I wonder what it's going to be like when Thunderbolt 5 egpu docks come out. Thunderbolt 5 vs Oculink Hype!
Note that in Thunderbolt 5 or Thunderbolt 4 you don't have full bandwidth only for GPU ...
Where in Oculink you have 100% for GPU (or you can connect storage :))
Curious as to why the oculink have lower tdp? Almost 40-50w different, that could be where the performance gone. Also it would be great to also show GPU clock speed since it could suggest if it’s run at lower speed as well.
Less load on the GPU - less power consumption
@welly4986 Sorry, how do you think if it is an oculink adapter for a SSD M.2 PCI-E 4.0 x4 slot on motherboard, results will be the same? I don't have an oculink in my laptop, but i have Intel core 12700h with 2 ssd 4.0 slots and i want maximum usage of my graphic card. Thanks!
Awesome video thanks dude, just wondering about my mini pc it has 4070 12g and I want to use egpu 4090 connect them using USB c from 4090 to DP mini pc is it going to work ? I have no idea how to connect that I have DP, HDMI and USB 3.2 ports
random but when you said youre too tired, what were you using?
DeepL translator
I may missed it but the oculink + eGPU only works on Windows or it works with Linux as well?
Oculink port does it means HDMI?
no, it's a totally different port and connection
hi, thanks for sharing. this is what I search for. So the Oculink has better performance over Thunderbolt 3 or 4 right?
Yes, but it has some flaws too. I’ll make a direct comparison video soon
I'm an editor so I'm more focused on workflow dedicated gpu's, like for instance the Sparkle/Intel Arc Series. Will this eGPU work for me?
The real bottleneck is the protocol itself which only has 4x pci-e lanes allocated.
do you recommend any PCIE oculink card instead of the m.2 version?
Was that D4 Result with or without recently added Ray Tracing? I'm assuming with, as my 4090 nets me around 100-120 @ 3440 x 1440, but I'd like to confirm.
All benchmarks was done with Ray Tracing and without DLSS in 4K. I mentioned that
good test
Would be good compare to pci express loss.
awesome video
Is there any latency difference between egpu and regular gpu?
Thats actually not that much less, surprising makes you wonder how much bandwidth is actually needed for these cards in reality. Perhaps it would be different if it was running at 1440p or 1080p.
Resolution doesn't affect performance, I tried to explain it here - ruclips.net/video/_UE_2NyaAas/видео.html
In any case, this is the best that is now for those who want to play on portable devices from time to time, or do not want a full-size PC, or simply change devices often, and the video card remains, or during the transition period, when the entire amount for a global upgrade is not available.
Where you bought from 6 pin out - to gpu ?
Hi ive watched your video about oculink and ordered the same one like you but im having game crash after setup, i was using a thunderbolt 3 egpu before with same setup as oculink with no problem, using a mini pc but oculink is giving problems with DXGI_ERROR_DEVICE_HUNG (0x887a0006). Do you have anymore information on the oculink setup please.
Hey, not really sure that if it's eGPU related problem. I did some googling and it seems that there could be a lot of reasons, including XMP profile on RAM in the BIOS settings. Do you face this problem only with some specific game or it's a global issue? I never had this error so I am not sure that I could help you man
I'm thinking of using this to power a 1650 and connect that to my laptop. My job requires some travel so this sounds like a good alternative and is small enough to carry around. Or does anyone have a recommendation for a small enclosure that supplies 75W?
You'll probably get the same performance with the iGPU 780M, IMO it'll be useless to use Oculink with a 1650
Occulink vs direct pcie is like pcie port 4×4 vs pcie 4×8 or pcie 4×16
While I generally agree; removing the GPU from the case has thermal implications that make for some results that won't always align perfectly.
El vídeo muestra que hoy en día, ya no hay diferencia entre una laptop y una PC de escritorio.
Ya estamos en abril de 2024
Y que bien, me alegra que avance a pasos agigantados la tecnología
alv
Was cyberpunk 2077 running in path tracing mode? If not on the 4090 could you please test that. Also can you test Control with RT at 4K and Alan Wake 2 with full Ray Tracing at 4K with as high of settings as you can run? I spy a OneXplayer X1, could you rerun the tests including that? I'd appreciate it. I'm considering the ONEXPLAYER X1 with 64GB of ram myself.
I sold my 4090 so I can’t run those tests now, but in a week or two I’ll release my review on OneXplater X1. Man…that device has a lot of troubles. Avoid it at all costs.
@@trysometech52 They've released new bios updates recently in the last 2 weeks or so. You have you get from their discord, which is shady and stupid. There's a video of the trick required that you need to do to update it. I'll be watching your review either way.
I don't like how the controller module to link them together and use them via bluetooth is still a "pre-order" thing. A channel called Barely Average Dude and he didn't like the controller.
I'm interested in the tablet style function more than the controllers as I would use a TB hub and KB and mouse. If TB5 wasn't coming out at the end of the year I would have purchased the X1, but considering it will be faster than current OCulink devices I'll used that as I want a portable that can give me at least 8x PCI-e 4.0 and TB5 will do that, getting the EGPU bay for it will be a wait. Intel already has a demo of it on their yt channel.
I’ve updated BIOS two times already, there is something more than just a software. Also AMD version of X1 will be released soon, but even that will not fix some other problems.
Isn’t Thunderbolt 5 also has 4x4.0? I think performance will be similar, but with hot swap and other features
@@trysometech52 There's a 120 Gbps option that uses PCI-e 4.0 x8. The standard mode is the x4 same as prior TB, but with PCI-e 4.0. Since the tube hates links just search for thunderbolt 5 120Gbps and Intel has videos on it and there's plenty of articles on it.
I think that’s just a marketing material. In reality TB3/4 is not 40gbps for eGPU connection (I have a video on this topic). Also they claimed x2 performance, which is exact difference between PCIe3 vs 4. So technically they didn’t lie.
Oculink is the best
Brother I have a question
I have 2 ports of USB 3.2 ...
Can I connect egpu to my laptop??
no
How can i connect the m.2 adapter to an office laptop which I can not open?
Thunderbolt to M.2 then M.2 to OCulink.
what was the m.2 slot speed? gen 4?? Gen 5 should be fast enough to run the 4090 at full speed on oculink.
We need Oculink v2 for that, cuz currently max speed is 64gpbs, which is PCIe 4.0x4
@@trysometech52 it would be interesting to try a gen 5 m.2 slot and see the difference if any, because even at the x4 bus width, should be faster on a gen 5 slot. pci express 4.0x4 vs 5.0x4
can you trying 1080 or 1440p to please
Seems like the 1% lows are the main problem.
Still more powerful than a 4090 laptop
Is there any advantage to M.2 gen 4 vs OCuLink?
No, because Oculink also using PCIe gen 4x4
So, we could do VR with oculink?
Sure, VR headset is just like en external monitor for the GPU. I tried that with Thunderbolt eGPU and it worked just fine
@@trysometech52 oh ok, I heard it can deep up to 30% of perfs. Which headset are you using?
Oculus Rift S. It's pretty old now, but still gets the job done :)
that egpu adapter (beast) is the cheapest one you can buy. if it has the wifi m.2 connector pcie x1 putting a rtx 4090 on it is a joke. the nvme m.2 connection is at least pcie x4. get an rg43s
Nope, this adapter was PCIe 4.0. I literally started from the bandwidth benchmarks