NOTE: If you plan to use WiFi card slot for M.2. eGPU, better double check that port. Because: 1. If some cases you'll need an adapter to PCIe mini 2. Most likely bandwidth speed will be limited to x1 PCIe line
Hi. Can i ask you some questions 1. Can i use my WiFi Card Slot for my M.2 SSD 2. How much PCIe bandwidth lane is on the M.2. NVME slot My laptop only have 1 NVME slot so i want to swap my SSD to WiFi Slot and use an USB WiFi. Then i will use Oculink through my NVME slot. I just wonder the bandwidth of its to choose a GPU in my budget. Thanks for your video. Great content.
@@tuannamtran9030 I had similar questions like you few days ago. you can use your nvme ssd on wifi card slot, I guess your wifi slot is most probably a m.2 e key type with pcie 3 x1 lane, you can buy a m.2 e key to m.2 m key adaptor and use ssd in that port, Bandwidth depends on version of port if wifi slot is pcie 3 x1 it's bandwidth is around 1GB/s (which i think would be enough for your ssd if you plan to use in that slot) if nvme slot is pcie 3 x4, it's bandwidth will be around 3.5 - 4 GB/s if your slot's are pcie 4 than it's gonna be double bandwidth of pcie 3 wich I mentioned above
man, i used to plan to have a egpu but it price is unaffordable for me so i'd rather buy a whole pc and laptop is just for work. Your video explain very detail. thanks for that
And I think that's the right decision. You can find used components and build a decent PC with great upgradability. Wish you the best deals on the component market :)
So my takeaway is Thunderbolt is convenient but not so performant right now, OCULINK is less convenient but more performant. Thunderbolt is more common, but OCULINK may require an adapter. Is that it basically?
In more technical terms, TB3 can only capture about 50-70% performance of the GPU card, whereas NVME egpu or oculink can both capture about 90% GPU performance. Which is a huge huge difference. BUT Nvme/oculink you have to turn off your pc when unplug/plugging it etc. TB3 is plug and play and doesn't require a wire inside your laptop/mini PC. Therefore, TB3 should be used if you move around your setup a lot, or if you travel a lot. Because those are the advantages of TB3. I personally will go for NVME m.2 egpu dock because I want to capture as much performance as possible from my purchased card, unto my mini PC. If I buy a card for $500 and get max 70% performance (at best), then I've lost $150 and it's as if I have a lower rank card. Defeats the purpose for me. ALSO, TB3 egpu is 50% more expensive than any NVME M2 dock. (but obviously because it has a casing - you can still get a more expensive NVME M2 dock if you want casing).
Such an informative video! You're quite impressive. I learned so much from you in just 2 videos compared to hours of trying to do my own research. Really really helpful. Thanks for all the detailed facts with pros and cons, and the thorough testings done. Subscribed.
Would be cool to see the PCIe 4 x8 performance with the Framework16. We would need to adapt their thing to oculink but there are some projects on the framework forum already.
I hadn't noticed it before, and no wonder why. It's very hard to find any decent information on it even now. Apparently it can run faster on USB4 devices, but it's still a way off 4x4.0 Looks interesting though, thanks for the info
@@trysometech52 Yes it doesn’t give the full bandwidth of 4x4.0 because of the limits of USB 4 which is 40 gbps, but it give higher bandwidth than 4x3.0, which is about 30 gbps.
Happy to see you promote UGREEN, I've only had good experiences with this brand name over the years! P.S. Oculink better than Thunderbolt, but more hustle to deal with. P.P.S. Bro has an arm hooked holding his device above his head and gaming in bed, absolute G move :D
I'm using Intel ARC A750 as eGPU USB4 and it is working. but if you miss the timing of switch on for PSU it will causes VIDEO_SCHEDULER_INTERNAL_ERROR error, and also when you using GPU for first time （every time）freeze for 5 sec.
I ordered the egpu dock from aliexpress, ordered a 500w power supply, and just got the Zotac GTX 1070 today. Can't wait to try out the egpu setup. If anyone's wondering, total cost was $255. Dock was $120, Power supply was $40, and GPU was $95 used from ebay. My laptop has a 12th gen i7-1255U 32gb ram MX550 GPU, which is why I'm getting the egpu cus the mx550 is not the best.
@@mightymallet5474 oh yea I forgot I commented this. I ended up finding out that my laptop doesn't have a thunderbolt port, but rather an "always on" USB c port which is also labeled with a similar lightning bolt. I ended up returning everything ☹️
Hey, thanks for this nice video! I would like to ask 1. if you have faced any issue with Oculink since this video was released? 2. I see that OCUP4v2 is now available. Would you recommend it or a different OCULINK adapter? I'm planning to use with an RTX 4060. Thank you!
Hey, but this is OCUP4v2 :) I didn't use it much after this video, so I can't say. But the Thunderbolt eGPU from the same company was very reliable. So I think this one should also be good in the long run. In terms of recommendations, it depends on your main device. If it's a handheld, then you should go for this one. If it's some sort of mini PC/PC and you're using an m.2 slot, then you can go for a cheaper version and spend the extra money on the 4060 Ti instead of just the 4060.
@@trysometech52 I have a dell precision mobile workstation with an i7-11850H and it has one M2 PCIe 4 x4 and one PCIe 3 x 4... I will use the 4.0 for the Oculink adapter and install the SSD on the 3.0. It has a NVIDIA T1200 (GTX 1650 Ti) and I really need more Gaming performance. The Oculink is something that I'm looking forward to test.
A lot of people don’t make too much videos about EGPU maybe make more videos of it by using it to plug-in directly to a handheld because they might up to 80 gigs of speed by using USB 5
Great guide, thx! I hope future high bandwidth communication interface will be a standard for pairing your mobile or mini PC device with the external PCIE accelerator like GPU card. TB5 or future Occulink should become standard and honestly they shouldn't be bottlenecked vs full PCIE connection.
Thanks for the inputs, I'm planning to make a back up gaming rig via egpu with a Dell Latitude 5400 with i7-8665u. This video gave me ideas for the build.
Thanks for the video. I see similar increases between the 2 connections depending on resolutions (i.e. CP77 benchmark with the X1 and onexgpu (7600m XT) 33%-52% increase from TB vs oculink 1440p-1080p, X1 with a SGWzone 4060 desktop, same benchmark 28-41% increase from TB to oculink). Yes it's more convenient for TB and some games just don't need oculink bandwidth to play 120fps at 1440p or even 200+fps so I save oculink for more demanding games if I don't have access to my 3080 desktop
Man, I planning to use eGPU with mini-pc on ryzen, thats mean, I can for everyday use eGPU OCuLink, and if i need only mini-pc for trip, i just unplug OCuLink cable. For me its very good solution, and its very compact solution. Thank you for your review!
Great video, my setup right now is a Rog Ally X with a Razer Core X that has an Rx 6700xt inside and it works "OK". I can't wait in till there are some good Thunder bolt 4 EGPU docks. I think 40gbps will be pretty good for my type of setup.
Thank you, you do one of the most interesting GPU related benchmarks out there. Just want to confirm. Are you sure you have not hit any (even if little) bottleneck with 4090? 40%~ usage (for example for God of war) for CPU may be misleading based on cores utilisation. It is just weird to see below 99% for GPU in direct connection.
Hey, glad to hear that. Thanks! Regarding CPU, I have a video where I compared it to 14900K in games - ruclips.net/video/CRvw-JxhOiU/видео.html Also check pinned comment there, because I had even better results there.
Jumped on the egpu bandwagon with a mini pc and dock from Ali too. Still waiting on the dock but I already received my RX6800 and corsair RMx PSU that’s supposed to be very quiet. My first pc build in 20 years 😂 all for the sake of sim racing 😅
excellent video work, and home work! haha - don't even play games, but good info digging into the x4 pci4/3 and TB4 . gotta get my teslas into an egpu setup for basic ai.
You should have tested different Oculink eGPUs as they perform differently. I recommend SFF-8612 / SFF-8611 because that is better compared to OCuP4V2. IMHO.
Thanks for this very informative video of yours! May I ask how those benchmark numbers might compare with other GPU intensive tasks like video rendering, AI image generation or LLMs? Do they need also a lot of bandwith? If I take a look at the task manager when image generation is running, the GPU does barely anything in 3D and seemingly just using the VRAM. I assume model offloading into the VRAM will take longer with Thunderbolt though, because this is using bandwith for sure.
Everything you need to know but said nothing about the software side. What to uninstall, drivers if device already has a dedicated GPU ect, should we boot in bios to detect the Gpu, how to do it ect ect
Hey just a some question, i heard from reddit and Eta prime that AMD RX is not very optimized for egpu is this true? I've seen rtx 3060 outperforms the rx 6750 xt. Maybe you can test it in the future with the same performance bracket e.g. rtx 3060 vs rx 6600? Thank you so much!
Hi there, thanks for the video! Quick question: I want to swap out my current gpu but find it really hard to remove it. Is there any latch/mechanism holding the card into place at the Th3p4g3 eGPU (besides the one screw at the top)? Or do I just need to pull stronger?
Also tell when using thunderbolt 3 how can we connect external monitor? Do we have to use the port on graphics card to external monitor or use the port on laptop to external monitor
Hey, the OcuP4 just has a more stable stand and it works with my handheld because of the power button. The Beast has the same performance for less money, so it's up to you :)
@@trysometech52 thanks for the response! Currently I have a 2080Ti connected to my laptop via TB4. Do you think the performance will improve if I connect them via M2 with the Beast Oculink?
Yes there will be a little improvement over TB just like I showed in this video with RTX 2060. Just keep in mind that 2080Ti is also PCIe Gen 3. So you will not get full benefits of Oculink port. Also don't use Wifi m.2. slot for that, I recently figured out that it most likely will be not x4, but x1 PCIe lines.
@@trysometech52 thanks for all the info. I’ve just bought the oculink beast. Hope arrive soon. And I will connect them to an ssd m2 port. When they arrive I will send you my results! Thanks a lot
Hi! Yesterday I’ve received the Beast base. It works great with my 2080ti. It performs better than the TB4 connection. I was thinking to upgrade to a Rtx 3070 but my M2 port is Gen3. Soooo, I don’t know if it worth it. It will be limited by the bandwidth. What do you think?
You mentioned in your video that all existing eGPU docs are built on Thunderbolt 3 specs, i.e. it doesn't make sense to use them on TB 4. What about those eGPU docks which are advertised on Aliexpress as being compatible with both versions (3 and 4)? Or is this really just saying "it will work with TB 4, but with the specs/limitations of TB 3 with regards to downwards-compabitility", i.e. you won't get TB4 bandwith with these eGPU docks, even if your laptop supports it? Also, is it better to use shorter TB cables to get the most bandwith (most docks come with 0.5 to 2m cables)?
I think they're just saying that it will work on TB4/TB3 because of backwards compatibility and they have the same bandwidth limit. Regarding cables - it will work on max bandwidth on 2m cable just like on 0.5m.
Bro, nobody is talking about Thunderbolt 4 when comparing to Oculink. EVERYONE wants Thunderbolt 5 to come out and take a shit on Oculink and bury it under the sands of time.
I am very interested in the Asus rog flow series of laptops that have an external video card (rog xg mobile), but unfortunately there is only a mobile video card. Although there is a pcie 4.0 connection. If you have a chance can you do a review changing mobile to desktop version ?
Hello Sir, I am using Hp Elitedesk 800 G6 Sff with i5-10th Gen. Will this method work for me? There is some space available for low profile graphic card but its not much for Gtx 1080. I watched all of your videos related to Egpu. Can you please confirm? Anxiously waiting for your reply😊. Thank you.
as things are right now, i will buy a mini pc with the oculink port and the other things just because i can get spare parts of the same model for like 5 years if something happen to my hardware, aside that there is not a lot of margin to update the entire stack, also the laptop spare parts aside the same things and most of times having the cpu fused with the motherboard just put an extra price to the maintenance and maintanability of the full build, so for me the oculink with mini pc is better option than had to get a laptop with more things that can broke at higher price. Aaaaand i think i watched that in your other video, wich was alongside this very good because gpu's sure are expensive.
It is honestly so disappointing to see absolutely nothing regarding thunderbolt 5 so far.. After the launch of that $4000 razer laptop that already has a 4090 in it, no other laptop or device has come out with it.. The fact that they're taking this long with it makes me quite hopeless that the development and implementation of thunderbolt 5 will work as well as everyone expects it to with eGPUs
Hello! If i go for the M.2/Occulink from laptop to the thunderblot egpu chassis, will i then be able to connect my thunderbolt monitor by thunderbolt cable to the egpu board rather than then the GPU card HDMI to HDMI?
let's be clear here, most gpu activity does not use the full x16 bandwidth, it's A LOT of bandwidth and for most of the time it's just not used, it's mainly for loading textures and data onto the GPU, so basically, the only thing link bandwidth beyond a single pcie 4.0 lane will affect is load times. making sure the pcie lanes connect at the latest standard version of pcie that all your hardware can do is more important. I've run thunderbolt gpus before, the only real overhead with it is the fact that you're sending the video back over the thunderbolt cable and onto the laptop's screen, so just buy a better gpu lol.
Do you ever have problems hot plugging thunderbolt? For me, something about either my enclosure, my cable, my device, or my operating system means it frequently fails to work and I end up having to restart almost every time.
Usually everything works fine. I've had a few instances of Legion Go disconnecting when moving the handheld, but that's more of a port issue + cable bend.
@@trysometech52 Interesting. My GPD Win 4 only ever recognizes my eGPU when I plug it in during the boot-up process. If I plug it in when the machine is already running, it always fails.
What would you say could match up with a Intel® Core™ Ultra 5 125U (12 Cores/14 threads) + 16 GB LPDDR5-6400 MHz RAM running with Thunderbolt 4 (40Gbps signaling rate)?
Is this possible to make a setup with 2 types of connections? So, if needed, you connect through TB for portability but with more losses, if you need more solid performance for more demanding games you connect with Oculink with minimal losses - so it would be a base setup for TB but with possibility of Oculink. Is this possible to have such setup in one?
@@trysometech52 But I am not asking for the mobile pre-made eGPUs. Let me ask it the other way then - if YOU would be creating a setup for yourself, where you have setup like the in video (with TB) BUT you still want to sometimes squeeze more power through Oculink for demanding games, what would you do, if you have one GPU, desktop 4090, and want to get the most out of it sometimes (since TB is not giving the most)? Just make 2 separate setups for TB and OL, with screwing 4090 in and out in each? Is there a (simple) setup that would give you choice between TB and OL depending on what you want at the moment in one setup case?
Yeah, I got your question right from the first time. But you didn't specify which GPU we are talking about either :) That eGPU dock doesn't exist yet (at least I never heard of it), so it's manually re-screwing the GPU, which doesn't sound convenient.
I have a Thunderbolt egpu, and a G14 laptop with a Ryzen 7 6800 CPU and USB4. Considering RX6800, RX7800xt, and 4070 Super. May I get some recommendations on which one to get please?
In this budget I bought a 4070S for myself and used it as an eGPU in my previous video. I just don't mind using DLSS3 and AMD's FSR isn't that good yet in my opinion.
Привет. Что думаешь на счет lenovo thinkbook 14+ и внешней видеокарты по oculink? По-моему, хороший мобильный cpu способен обеспечить норм производительность в играх
Did you have trouble installing nvidia app(experience and new beta one)? Driver is installed without problem. Nvidia 4060 is ok in windows device manager but nvidia apps dont work and games cant see nvidia 4060.. only the integrated one. Thxs
They did said USB 4 port will work fine directly to external graphic card because they said it’s comfortable with Thunderball for directly to USB 4 means you can get some transfer speed really well to get some performance boost in the future future that they will fly away that USB 5 five up to 80 gigs of speed
@@trysometech52 sorry it’s just I using speech to text by using a microphone so I don’t know how to type. I have a learning disability so I meant thunderbolt 5 port
@@trysometech52 will it be possible to enable resize bar if its an oculink port connected from an m.2? Since I checked hwinfo on my surface pro 8 and im not so sure if it would have resize bar support for the m.2 slot
Does it have a thunderbolt port? I can’t find any information on it. If it doesn’t, you could use the m.2 expansion slot for oculink, but you’d have to cut a hole in the casing to plug it in.
I’m using this power bank for my Rog Ally. But it can’t activates the 30W mode of it, regardless that its OUT2 port supports 65W output. Except for that I have no complaints.
Hmm, I just checked on my handheld. I had around ~85w from OUT1 and ~55w from OUT2. But my battery on handheld is around 50%, we should also take that into count.
I am not a pro in this question, but it works for me, so it's possible. Are you sure that we have the same version? I know that there is the same looking but older one. Also if all 3 ports are used - power output is different. And last point is cable - make sure that it's good. Mine supports up to 100w of power delivery.
If your laptop has USB4, yes, you can use both TB4 and USB4 dock, BUT TB4 dock is 1gb faster, there is only ONE brand that makes such GPU docks for now and they have no charging, but they 30% faster VS the BEST TB4 dock, you basically get FULL PCIe Gen3 X3 speed TB4 dock on USB4 PC: 2500 device to host and host to device USB4 dock on USB4 PC: 3500-3700 device to host and host to device The worst speed is on Intel, TB4 on Intel will be 2200-2300MB/s
I'm very disappointed with Intel that Thunderbolt 5 is not integrated on Lunar Lake. By the time Intel get their stuff in order iGPUs will make Thunderbolt eGPUs obsolete.
@@trysometech52 I highly doubt it. Panther Lake will be integrating Thunderbolt 4 just like Meteor Lake and Lunar Lake. Many OEMs are not going to be implementing the Barlow Ridge Controller due to extra costs. Alao the performance of a Discrete Controller is not as good as when integrated. Mass adoption will only happen when it is integrated. This is the advantage Apple have with Thunderbolt, they don't have to wait for Intel. For Apple there is no really benefit of moving to Thunderbolt 5 since they don't support eGPUs anyway. So sadly no, I don't think we'll see much traction with Thunderbolt 5 until 2026,fingers crossed.
@@trysometech52 i travels a lot So i am stuck on a 4090 sff build with water cooling or a laptop ? But i want to play heavy demanding games ! So which one i choose !
@@NikhilKumar-yq1fx PC is always better in terms of performance, but even SFF will be a heavy brick. eGPU with a case will be almost SFF PC size. So if you travel a lot, I'd stick to a powerful laptop and ignore eGPU topic.
@@warlockboyburns that's total bandwidth, but for PCIe devices it will be limited. Just like it was in case of TB3. TB3/4 = 40 total => 32 for PCIe => 22-23 in reality TB5 = 80 total (120 for one side data stream) => 64 for PCIe => ?? in reality
This video is filled with tons of small errors and maybe outright misinformation. Also the benchmarks are confusing as hell, why not state what device use and not just bandwidth, bandwidth is meaningless, and tehre is no end graph to compare all games. he says there are 3 types of connection types and lists Oculink and M.2 separately. oculink and M.2 is the same exact thing, Oculink is not a protocol but connection type, extension cable. he says Oculink is Gen 4, while in reality it can be anything,Gen 1, Gen 2, Gen 3 , Gen4, Gen 5, it depends on your PCIe and where you plug it and what cable adapter you use or if its factory made then to where they soldered it. In his example of connecting to WiFi slot, he didn't said a word that WiFi connectors are PCIe Gen2 or 3 x1 [most moderns laptops usually have gen3 x1], they 4 times slower than standard m.2 and much slower than TB4/USB4. The main types of connection options are these: 1] PCIe Direct [Full Speed] 2] PCIe extensions: Oculink, M2, PCIe Extension cable. These are usually x4, but can go up to x16. Oculink comes in x4 and x8 versions, if you plug a x8 Oculink adapter into full sized PCIe slot, you'll get Oculink x8, both the adapter and the cable available on amazon/ali and other sites, if you buy long PCIe x16 extender, can be a meter [with reDrivers even longer], you'll get x16 full speed. [There are MORE options, like the cables that used for SAS and U.2: Adapter > Cable > PCIe Dock] 3] TB4 [2200Mb/s to 2700Mb/s depends on the dock and the host] 4] USB4 [3500MB/s-3700Mb/s depends on the USB4 dock and USB4 Host] Using the same brand of TB4 dock he has in his video the Host to device and device to host, on AMD platform will be 2500-2700MB/s [Legion Go]. Connected to Intel: 200-300MB/s LESS. [X1 Intel] If you use USB4 dock, you Gain 1Gb: USB4 Dock on USB4 handheld like Legion Go you get 3500-3700MB/s Host to device and Device to Host. USB4 on AMD USB4 host gives you the same performance as Gen3 x4 NVMe SSDs, you basically max the PCIe Gen3 x4 speed! The only Benefit of Oculink vs USB4 only if Oculink supports PCIe Gen4, in case its pluged to Gen3 and your system has USB4, go with USB4 for ease of use [hot plug, but right now lacks charging]. USB4 is 1Gb faster, because USB4 doesn't dedicated 1Gb for USB passtrought, TB does it all the time, even if the dock has no USB ports! Because of this protocol change even TB4 dock [like his in the video, i have the same one] that is pluged to USB4 Host [Legion Go/ X1 AMD] gains free 200-300Mb/s boost vs. Intel Host that uses native TB4. Next gen Intel like Arrow Lake should come with Native USB4 host, so everybody will be happy, but right now AMD wins twice: TB4 on AMD is 200-300MB/s faster AND AMD has USB4 docks that 1gb faster than TB4 [sold on Ali for 100-120USD]. Also as i mentioned his benchmarks are confusing for most people, he should of written TB/Oculink/PCIe x16, instead of writing bandwidth, bandwidth doesn't say anything to most people. Also it would be nice to get a slide with just numbers table that compares everything, even excel file screenshot.
I always take criticism or corrections positively, and if I'm wrong somewhere, it's certainly not to confuse people deliberately. Some of what you wrote I really hadn't heard of, but other than that you yourself are personally starting to confuse others from the beginning. For example, I didn't say there are 3 types of connectivity, I said there are 3 main types of eGPU. These are completely different things. After that, a lot of what you wrote just doesn't make sense. "benchmarks are confusing for most people" - What makes you say that? You're the only one writing about it :) The rest of us just looked up a little higher in the benchmarks and saw that it was already signed in the GPU name. And thanks to you, they will now realise that if Thunderbolt 3 has about 23gbsp, and they buy a USB4 eGPU for an AMD processor, it will work even better for them. Oh and don't you agree that showing a 2060 that runs on gen3 and just signing ‘Oculink’ wouldn't be quite right?
@@trysometech52 whats up? The right thing with Benchmarks is just say what GPU it is and what device/dock it runs on, thats how everybody does it and it will be great to have a summary so people can take screenshots of, but you do what you want to do. 3 type of eGPU means they unique, while M.2/Oculink is all the same thing, the difference is what PCI express extension cable you you use, M.2 can be Oculink with M.2 to Oculink adapter, its all same thing and makes no difference [and wifi connection is x1 so it should be mentionedy] hence its better to split them up by connectivity ie PCIe/TB/USB4 and actual real bandwidth [you can check bandwidth by using cuda benchamrk or AIDA]. if you need more info, benchmarks and tools [like fixing nvidia USB4/TB crashes aka error 43 on AMD], join iGPU.io its a great resource.
But hey, I mentioned that M.2. and Oculink are compatible and you can combine them :) x1 speed on Wifi slot was a surprise for me, I heard this option from many sources and was sure that this is legit option. I'll add that info in the pinned comment. Thanks!
@@trysometech52 Honestly, I think the simple way you described things was perfect without getting in the weeds...you demonstrated a simple relationship between connections using the most powerful Nvidia GPU available and 4k resolution as the common denominator....you basically helped set expectations in a good way. Nice video, keep it up!
NOTE:
If you plan to use WiFi card slot for M.2. eGPU, better double check that port. Because:
1. If some cases you'll need an adapter to PCIe mini
2. Most likely bandwidth speed will be limited to x1 PCIe line
Why your English so Russian?😮
Hi. Can i ask you some questions
1. Can i use my WiFi Card Slot for my M.2 SSD
2. How much PCIe bandwidth lane is on the M.2. NVME slot
My laptop only have 1 NVME slot so i want to swap my SSD to WiFi Slot and use an USB WiFi. Then i will use Oculink through my NVME slot. I just wonder the bandwidth of its to choose a GPU in my budget. Thanks for your video. Great content.
@@foggy8256 Can't you tell from his accent? It's because he's Slavic (not necessarily Russian)
@@tuannamtran9030 I had similar questions like you few days ago.
you can use your nvme ssd on wifi card slot,
I guess your wifi slot is most probably a m.2 e key type with pcie 3 x1 lane, you can buy a m.2 e key to m.2 m key adaptor and use ssd in that port,
Bandwidth depends on version of port
if wifi slot is pcie 3 x1 it's bandwidth is around 1GB/s (which i think would be enough for your ssd if you plan to use in that slot)
if nvme slot is pcie 3 x4, it's bandwidth will be around 3.5 - 4 GB/s
if your slot's are pcie 4 than it's gonna be double bandwidth of pcie 3 wich I mentioned above
@@tuannamtran9030 same thing, I'll also use oculink through nvme and swap the ssd to the wifi slot
man, i used to plan to have a egpu but it price is unaffordable for me so i'd rather buy a whole pc and laptop is just for work. Your video explain very detail. thanks for that
And I think that's the right decision. You can find used components and build a decent PC with great upgradability. Wish you the best deals on the component market :)
I actually several months ago had a razer core x egpu enclousure ended up selling it
As i switched to cloud gaming
yea, it's not ideal and stupid. the problem is, laptop already had a great performance nowadays with the same value or even more value at its price.
@@unaddan Yes! It was a issue several years ago. Now these days laptops are almost or even at toe to toe with desktop gpu variants
So my takeaway is Thunderbolt is convenient but not so performant right now, OCULINK is less convenient but more performant.
Thunderbolt is more common, but OCULINK may require an adapter.
Is that it basically?
In more technical terms, TB3 can only capture about 50-70% performance of the GPU card, whereas NVME egpu or oculink can both capture about 90% GPU performance. Which is a huge huge difference. BUT Nvme/oculink you have to turn off your pc when unplug/plugging it etc. TB3 is plug and play and doesn't require a wire inside your laptop/mini PC. Therefore, TB3 should be used if you move around your setup a lot, or if you travel a lot. Because those are the advantages of TB3.
I personally will go for NVME m.2 egpu dock because I want to capture as much performance as possible from my purchased card, unto my mini PC. If I buy a card for $500 and get max 70% performance (at best), then I've lost $150 and it's as if I have a lower rank card. Defeats the purpose for me.
ALSO, TB3 egpu is 50% more expensive than any NVME M2 dock. (but obviously because it has a casing - you can still get a more expensive NVME M2 dock if you want casing).
@@nimkal didnt you also say that rtx had to be off? so 90% of raster performance?
Mate, your works are incredible. I hope your channel becomes bigger.
Such an informative video! You're quite impressive. I learned so much from you in just 2 videos compared to hours of trying to do my own research. Really really helpful. Thanks for all the detailed facts with pros and cons, and the thorough testings done. Subscribed.
The editing of this video is really good bro
Quality content and insights here on eGPU options and comparisons. Thank you.
Perfect timing dropping this video
Would be cool to see the PCIe 4 x8 performance with the Framework16. We would need to adapt their thing to oculink but there are some projects on the framework forum already.
awesome videos😯
let's chill with simple thunderbolt connection and wait for tb5 :D
Thanks man! Sounds like a plan :)
Any info on when will tb5 come out?
Not all USB-C eGPU uses PCI 3.0, some of them uses PCI 4.0 x 4 lines like ADT-UT3G which give higher bandwidth with better performance.
I hadn't noticed it before, and no wonder why. It's very hard to find any decent information on it even now. Apparently it can run faster on USB4 devices, but it's still a way off 4x4.0
Looks interesting though, thanks for the info
@@trysometech52 Yes it doesn’t give the full bandwidth of 4x4.0 because of the limits of USB 4 which is 40 gbps, but it give higher bandwidth than 4x3.0, which is about 30 gbps.
It could reach oculink's level on pcie 3.0
@@trysometech52With Oculink or exp gdc EGPU means there is also less CPU bottleneck?
Happy to see you promote UGREEN, I've only had good experiences with this brand name over the years! P.S. Oculink better than Thunderbolt, but more hustle to deal with. P.P.S. Bro has an arm hooked holding his device above his head and gaming in bed, absolute G move :D
thunderbolt 5 is really going to take things off with 'power up' stations and mobile handheld gaming as a whole
price will also flying
Would,love to see a video about your method for that p.c. gaming setup for your bed!! Not many videos on RUclips showing how to do that well.
I'm using Intel ARC A750 as eGPU USB4 and it is working. but if you miss the timing of switch on for PSU it will causes VIDEO_SCHEDULER_INTERNAL_ERROR error, and also when you using GPU for first time （every time）freeze for 5 sec.
I ordered the egpu dock from aliexpress, ordered a 500w power supply, and just got the Zotac GTX 1070 today. Can't wait to try out the egpu setup.
If anyone's wondering, total cost was $255. Dock was $120, Power supply was $40, and GPU was $95 used from ebay.
My laptop has a
12th gen i7-1255U
32gb ram
MX550 GPU, which is why I'm getting the egpu cus the mx550 is not the best.
I'd love to hear how it goes for you. I want to do something very similar for short blender rendering.
Ur cpu is low power.
Bruh give us an update fr
@@mightymallet5474 oh yea I forgot I commented this. I ended up finding out that my laptop doesn't have a thunderbolt port, but rather an "always on" USB c port which is also labeled with a similar lightning bolt. I ended up returning everything ☹️
@@Fisheee123you could have used m.2 version or pcie of the wifi card
Hey, thanks for this nice video! I would like to ask 1. if you have faced any issue with Oculink since this video was released? 2. I see that OCUP4v2 is now available. Would you recommend it or a different OCULINK adapter? I'm planning to use with an RTX 4060. Thank you!
Hey, but this is OCUP4v2 :)
I didn't use it much after this video, so I can't say. But the Thunderbolt eGPU from the same company was very reliable. So I think this one should also be good in the long run.
In terms of recommendations, it depends on your main device. If it's a handheld, then you should go for this one. If it's some sort of mini PC/PC and you're using an m.2 slot, then you can go for a cheaper version and spend the extra money on the 4060 Ti instead of just the 4060.
@@trysometech52 I have a dell precision mobile workstation with an i7-11850H and it has one M2 PCIe 4 x4 and one PCIe 3 x 4... I will use the 4.0 for the Oculink adapter and install the SSD on the 3.0. It has a NVIDIA T1200 (GTX 1650 Ti) and I really need more Gaming performance. The Oculink is something that I'm looking forward to test.
Very well put together. Thanks!
Oculink is a great alternative for thunderbolt 4, but thunderbolt 5 has 120gbps of speed, so it's high enought to deal with a 4080 or less
A lot of people don’t make too much videos about EGPU maybe make more videos of it by using it to plug-in directly to a handheld because they might up to 80 gigs of speed by using USB 5
Great guide, thx! I hope future high bandwidth communication interface will be a standard for pairing your mobile or mini PC device with the external PCIE accelerator like GPU card. TB5 or future Occulink should become standard and honestly they shouldn't be bottlenecked vs full PCIE connection.
Thanks for the inputs, I'm planning to make a back up gaming rig via egpu with a Dell Latitude 5400 with i7-8665u. This video gave me ideas for the build.
I don't know of any mainstream laptop with oculink. Tb4 is almost everywhere.
Thanks for the video. I see similar increases between the 2 connections depending on resolutions (i.e. CP77 benchmark with the X1 and onexgpu (7600m XT) 33%-52% increase from TB vs oculink 1440p-1080p, X1 with a SGWzone 4060 desktop, same benchmark 28-41% increase from TB to oculink). Yes it's more convenient for TB and some games just don't need oculink bandwidth to play 120fps at 1440p or even 200+fps so I save oculink for more demanding games if I don't have access to my 3080 desktop
Cool. That's a smart way to use it :)
Im glad you showed the 2060 and not just the 4090 since there is much less of a cpu bottle neck. How would the results differ at 1440 p
Man, I planning to use eGPU with mini-pc on ryzen, thats mean, I can for everyday use eGPU OCuLink, and if i need only mini-pc for trip, i just unplug OCuLink cable. For me its very good solution, and its very compact solution. Thank you for your review!
Great video, my setup right now is a Rog Ally X with a Razer Core X that has an Rx 6700xt inside and it works "OK". I can't wait in till there are some good Thunder bolt 4 EGPU docks. I think 40gbps will be pretty good for my type of setup.
Thank you, you do one of the most interesting GPU related benchmarks out there.
Just want to confirm. Are you sure you have not hit any (even if little) bottleneck with 4090?
40%~ usage (for example for God of war) for CPU may be misleading based on cores utilisation. It is just weird to see below 99% for GPU in direct connection.
Hey, glad to hear that. Thanks!
Regarding CPU, I have a video where I compared it to 14900K in games - ruclips.net/video/CRvw-JxhOiU/видео.html
Also check pinned comment there, because I had even better results there.
Jumped on the egpu bandwagon with a mini pc and dock from Ali too. Still waiting on the dock but I already received my RX6800 and corsair RMx PSU that’s supposed to be very quiet. My first pc build in 20 years 😂 all for the sake of sim racing 😅
excellent video work, and home work! haha - don't even play games, but good info digging into the x4 pci4/3 and TB4 . gotta get my teslas into an egpu setup for basic ai.
Having to restart is such a pain tho :p
Maybe thunder bolt 5 will rescue us
You should have tested different Oculink eGPUs as they perform differently. I recommend SFF-8612 / SFF-8611 because that is better compared to OCuP4V2. IMHO.
Interesting, and how is it better? I had a maximum available bandwidth speed for Oculink with my eGPU
:) as always, nice video!. Soon I will see ETA prime competing with you.
Thanks for this very informative video of yours! May I ask how those benchmark numbers might compare with other GPU intensive tasks like video rendering, AI image generation or LLMs? Do they need also a lot of bandwith? If I take a look at the task manager when image generation is running, the GPU does barely anything in 3D and seemingly just using the VRAM.
I assume model offloading into the VRAM will take longer with Thunderbolt though, because this is using bandwith for sure.
Thanks! Take a look here - ruclips.net/video/R6v18s3EyD0/видео.html
@@trysometech52 thanks so much. Have my Abo, Sir!
So what better thunderbolt or oculik WHAT TO BUY?
OCuLink if possible, better performance and pricing although a bit more hassle
first you need a oculink port or external pcie oculink device
If you have an oculink port, get oculink. If not, get thunderbolt. Oculink isn't worth drilling holes into your device imo
For convenience and on-the-go gaming upgrade, Thunderbolt.
Everything you need to know but said nothing about the software side.
What to uninstall, drivers if device already has a dedicated GPU ect, should we boot in bios to detect the Gpu, how to do it ect ect
Because there is nothing to tell about. It behaves just like a regular GPU in the regular PC. You will get the image signal even without a drivers
Super interesting - thanks!
Beelink GTi Ultra with EX dock is probably what you want, PCIe x8 connection
Hey just a some question, i heard from reddit and Eta prime that AMD RX is not very optimized for egpu is this true? I've seen rtx 3060 outperforms the rx 6750 xt. Maybe you can test it in the future with the same performance bracket e.g. rtx 3060 vs rx 6600? Thank you so much!
Can't say for sure, haven't used AMD GPUs for a while.. but maybe in the future i'll get one
@@trysometech52 Thanks for the reply! I'll look forward to it!
Hi there, thanks for the video!
Quick question:
I want to swap out my current gpu but find it really hard to remove it.
Is there any latch/mechanism holding the card into place at the Th3p4g3 eGPU (besides the one screw at the top)?
Or do I just need to pull stronger?
Hey, there is no holding mechanism on the Th3p4g3, so yeah just pull a little bit stronger
Also tell when using thunderbolt 3 how can we connect external monitor?
Do we have to use the port on graphics card to external monitor or use the port on laptop to external monitor
Yes, just connect your monitor directly to the GPU
Hi! What base do you recommend? The OcuP4 or the Beast Oculink from your previous video?
Hey, the OcuP4 just has a more stable stand and it works with my handheld because of the power button. The Beast has the same performance for less money, so it's up to you :)
@@trysometech52 thanks for the response! Currently I have a 2080Ti connected to my laptop via TB4. Do you think the performance will improve if I connect them via M2 with the Beast Oculink?
Yes there will be a little improvement over TB just like I showed in this video with RTX 2060. Just keep in mind that 2080Ti is also PCIe Gen 3. So you will not get full benefits of Oculink port.
Also don't use Wifi m.2. slot for that, I recently figured out that it most likely will be not x4, but x1 PCIe lines.
@@trysometech52 thanks for all the info. I’ve just bought the oculink beast. Hope arrive soon.
And I will connect them to an ssd m2 port. When they arrive I will send you my results! Thanks a lot
Hi! Yesterday I’ve received the Beast base. It works great with my 2080ti. It performs better than the TB4 connection. I was thinking to upgrade to a Rtx 3070 but my M2 port is Gen3. Soooo, I don’t know if it worth it. It will be limited by the bandwidth. What do you think?
You mentioned in your video that all existing eGPU docs are built on Thunderbolt 3 specs, i.e. it doesn't make sense to use them on TB 4. What about those eGPU docks which are advertised on Aliexpress as being compatible with both versions (3 and 4)? Or is this really just saying "it will work with TB 4, but with the specs/limitations of TB 3 with regards to downwards-compabitility", i.e. you won't get TB4 bandwith with these eGPU docks, even if your laptop supports it? Also, is it better to use shorter TB cables to get the most bandwith (most docks come with 0.5 to 2m cables)?
I think they're just saying that it will work on TB4/TB3 because of backwards compatibility and they have the same bandwidth limit.
Regarding cables - it will work on max bandwidth on 2m cable just like on 0.5m.
@@trysometech52 Thanks!
could you in theory use a m.2 adapter to usb c and plug that into the thunderbolt port instead of losing your wifi card or nvme?
I'm pretty sure you'd lose the same amount of performance as normal thunderbolt
Bro, nobody is talking about Thunderbolt 4 when comparing to Oculink. EVERYONE wants Thunderbolt 5 to come out and take a shit on Oculink and bury it under the sands of time.
I think we're on the same page :D
But compared to what we have at the moment
mcio for egpu is way better than tbt5
@@heshijiewhat is mcio?
@@ciw8688thank is mini cool edge if I remember correctly
@@heshijiewhat’s mcio?
TB5 will be even beefier than Occulink. eGPUs will do incredibly well with Mini PCs
I am very interested in the Asus rog flow series of laptops that have an external video card (rog xg mobile), but unfortunately there is only a mobile video card. Although there is a pcie 4.0 connection. If you have a chance can you do a review changing mobile to desktop version ?
Can you tell me the song or music name when you started doing benchmarking? 😅😅
Spheres - Elliptik
Hello Sir, I am using Hp Elitedesk 800 G6 Sff with i5-10th Gen. Will this method work for me? There is some space available for low profile graphic card but its not much for Gtx 1080. I watched all of your videos related to Egpu. Can you please confirm? Anxiously waiting for your reply😊. Thank you.
Hey, can you clarify which method? I guess Oculink? If you have a free m.2. x4 slot, then sure you can go for it.
@trysometech52 Tysm :) Will definitely give it a try. Thanks again 😊
as things are right now, i will buy a mini pc with the oculink port and the other things just because i can get spare parts of the same model for like 5 years if something happen to my hardware, aside that there is not a lot of margin to update the entire stack, also the laptop spare parts aside the same things and most of times having the cpu fused with the motherboard just put an extra price to the maintenance and maintanability of the full build, so for me the oculink with mini pc is better option than had to get a laptop with more things that can broke at higher price. Aaaaand i think i watched that in your other video, wich was alongside this very good because gpu's sure are expensive.
It is honestly so disappointing to see absolutely nothing regarding thunderbolt 5 so far.. After the launch of that $4000 razer laptop that already has a 4090 in it, no other laptop or device has come out with it.. The fact that they're taking this long with it makes me quite hopeless that the development and implementation of thunderbolt 5 will work as well as everyone expects it to with eGPUs
Hello! If i go for the M.2/Occulink from laptop to the thunderblot egpu chassis, will i then be able to connect my thunderbolt monitor by thunderbolt cable to the egpu board rather than then the GPU card HDMI to HDMI?
if i only use for 1080p 60hz then will thunderbolt 3 fine?
let's be clear here, most gpu activity does not use the full x16 bandwidth, it's A LOT of bandwidth and for most of the time it's just not used, it's mainly for loading textures and data onto the GPU, so basically, the only thing link bandwidth beyond a single pcie 4.0 lane will affect is load times. making sure the pcie lanes connect at the latest standard version of pcie that all your hardware can do is more important.
I've run thunderbolt gpus before, the only real overhead with it is the fact that you're sending the video back over the thunderbolt cable and onto the laptop's screen, so just buy a better gpu lol.
Do you ever have problems hot plugging thunderbolt? For me, something about either my enclosure, my cable, my device, or my operating system means it frequently fails to work and I end up having to restart almost every time.
Usually everything works fine. I've had a few instances of Legion Go disconnecting when moving the handheld, but that's more of a port issue + cable bend.
@@trysometech52 Interesting. My GPD Win 4 only ever recognizes my eGPU when I plug it in during the boot-up process. If I plug it in when the machine is already running, it always fails.
hold up, I've heard that it works on both USB-C 4 and Thunderbolt 3 or later (without the need for AMD),
Yep, I said exactly the same. USB4 or Thunderbolt 3/4. There is also text hint: not necessary to be AMD for USB4
What would you say could match up with a Intel® Core™ Ultra 5 125U (12 Cores/14 threads) + 16 GB LPDDR5-6400 MHz RAM running with Thunderbolt 4 (40Gbps signaling rate)?
Is this possible to make a setup with 2 types of connections? So, if needed, you connect through TB for portability but with more losses, if you need more solid performance for more demanding games you connect with Oculink with minimal losses - so it would be a base setup for TB but with possibility of Oculink. Is this possible to have such setup in one?
Try to look at OneXGPU or GPD G1. But they are using mobile GPU, which you can't replace
@@trysometech52 But I am not asking for the mobile pre-made eGPUs. Let me ask it the other way then - if YOU would be creating a setup for yourself, where you have setup like the in video (with TB) BUT you still want to sometimes squeeze more power through Oculink for demanding games, what would you do, if you have one GPU, desktop 4090, and want to get the most out of it sometimes (since TB is not giving the most)? Just make 2 separate setups for TB and OL, with screwing 4090 in and out in each? Is there a (simple) setup that would give you choice between TB and OL depending on what you want at the moment in one setup case?
Yeah, I got your question right from the first time. But you didn't specify which GPU we are talking about either :) That eGPU dock doesn't exist yet (at least I never heard of it), so it's manually re-screwing the GPU, which doesn't sound convenient.
I don't have oculink port on my pc what can I do use converter to pcie or thunderbolt will it affect bandwidth?
I have a Thunderbolt egpu, and a G14 laptop with a Ryzen 7 6800 CPU and USB4. Considering RX6800, RX7800xt, and 4070 Super. May I get some recommendations on which one to get please?
In this budget I bought a 4070S for myself and used it as an eGPU in my previous video. I just don't mind using DLSS3 and AMD's FSR isn't that good yet in my opinion.
If you dont care about ray tracing and DLSS, 7800XT will be faster.
Привет. Что думаешь на счет lenovo thinkbook 14+ и внешней видеокарты по oculink? По-моему, хороший мобильный cpu способен обеспечить норм производительность в играх
В теории заучит круто, но обзоров ещё не видел на него. Да и вроде как их тяжело достать
Hello. In order to turn on the oculink egpu we just need to turn on the power suply? or the board has any kind of button? Thxs
Hey, power supply is always turned on, but there is a power button on the board itself
Did you have trouble installing nvidia app(experience and new beta one)? Driver is installed without problem. Nvidia 4060 is ok in windows device manager but nvidia apps dont work and games cant see nvidia 4060.. only the integrated one. Thxs
Hello
So there is no PCIe 16x to PCIe 16x riser??
Will e gpu work on my laptop with old gpu inside
They did said USB 4 port will work fine directly to external graphic card because they said it’s comfortable with Thunderball for directly to USB 4 means you can get some transfer speed really well to get some performance boost in the future future that they will fly away that USB 5 five up to 80 gigs of speed
Bro, sorry but I have no idea what you're saying.
Но ты можешь попробовать на русском)
@@trysometech52LMAO
@@trysometech52 sorry it’s just I using speech to text by using a microphone so I don’t know how to type. I have a learning disability so I meant thunderbolt 5 port
I bet you can't wait to get that thunderbolt 5 in your hands :P
Oh yeah, you're goddamn right :D
Wish i could enable usb4 in my Yoga 16ARP8... the 14arp8 and ideapad 16arp8 do have usb4, I don't understand that.
Is resize bar enabled automatically for the oculink port?
No, it's a part of the BIOS setting
@@trysometech52 will it be possible to enable resize bar if its an oculink port connected from an m.2?
Since I checked hwinfo on my surface pro 8 and im not so sure if it would have resize bar support for the m.2 slot
@@ausf7294 sure, the BIOS doesn't care how you connect your GPU or if it's connected at all. You can just this feature ON or OFF
thanks
That definitely looks like Spider Man at 7:23 😂
Oops 👀
@@trysometech52 just human error lol
multiple monitor setup with usb4 stutter a lot, so oculink really is the way to go for now
Only if one of them internal screen. External screens doesn't affect bandwidth as far as I know.
Idk i teat it out with one browser open 9 videos and the other running game, the game perform as bad as 780m@@trysometech52
Maybe a stupid question but would this work with Alienware alpha r2? Like the stronger version of the steam machine
Does it have a thunderbolt port? I can’t find any information on it.
If it doesn’t, you could use the m.2 expansion slot for oculink, but you’d have to cut a hole in the casing to plug it in.
@@TechMaestro01 Na it doesn’t have a one ill probably have to upgrade the case like you said
i have Viviobook 15 (X1504ZA) and geforce gt 710 2gb, can i connect them toghether using M.2 or require Oculink ?
Don't even bother. Your integrated graphics are better than the GT710
Could a Thunderbolt 4 to M.2 4.0 slot adapter use all the 40 Gbps of the port?
I can guess that it will be limited to 4x3.0 PCIe lines (32gbps)
@trysometech52 I mean connecting to the thunderbolt port of the laptop the m.2 adapter and then connecting to the m.2 port the m.2/oculink egpu
@@Stecus2704 yes, I understood that correctly. Anyway you physically will be limited by TB4 specs
Awesome, thank you so much!
Здорово, пойду и себе куплю
I’m using this power bank for my Rog Ally. But it can’t activates the 30W mode of it, regardless that its OUT2 port supports 65W output.
Except for that I have no complaints.
Hmm, I just checked on my handheld. I had around ~85w from OUT1 and ~55w from OUT2. But my battery on handheld is around 50%, we should also take that into count.
@@trysometech52 Neither works on OUT1. On their label, out1 doesn’t support 20V*3.25A 65W output, so this is expected.
I am not a pro in this question, but it works for me, so it's possible.
Are you sure that we have the same version? I know that there is the same looking but older one. Also if all 3 ports are used - power output is different. And last point is cable - make sure that it's good. Mine supports up to 100w of power delivery.
what about this EGO1-M eGPU
There are laptops with Oculink. Intel Ultra Powered Lenovo ThinkBook 14+ & 16+ 2024
I showed them on as upcoming laptops at 3:02, but I didn't find any info that they were already released. At least at the moment of preparing script.
@@trysometech52the were released in China. It is complicated to buy them as of now.
Oculink all the way.
can i use my usb c on thinkpad e590
4:23 :D
3:03 GPD Win Max, GPD Win 4 has oculink.
32 gbps is pcie 3.0?
4 lines of PCIe 3.0, yes
from my experience thunderbolt is not exactly stable, plus all sorts of little problems that its causing like sleep, windows scaling problems etc
0:07 Ahh the comment gave me "🤡" aura.
So any laptop with usb 4 can run egpu?
If your laptop has USB4, yes, you can use both TB4 and USB4 dock, BUT TB4 dock is 1gb faster, there is only ONE brand that makes such GPU docks for now and they have no charging, but they 30% faster VS the BEST TB4 dock, you basically get FULL PCIe Gen3 X3 speed
TB4 dock on USB4 PC: 2500 device to host and host to device
USB4 dock on USB4 PC: 3500-3700 device to host and host to device
The worst speed is on Intel, TB4 on Intel will be 2200-2300MB/s
I'm very disappointed with Intel that Thunderbolt 5 is not integrated on Lunar Lake. By the time Intel get their stuff in order iGPUs will make Thunderbolt eGPUs obsolete.
yeah.. it seems that TB5 will come in 2025
@@trysometech52 I highly doubt it. Panther Lake will be integrating Thunderbolt 4 just like Meteor Lake and Lunar Lake. Many OEMs are not going to be implementing the Barlow Ridge Controller due to extra costs. Alao the performance of a Discrete Controller is not as good as when integrated.
Mass adoption will only happen when it is integrated. This is the advantage Apple have with Thunderbolt, they don't have to wait for Intel.
For Apple there is no really benefit of moving to Thunderbolt 5 since they don't support eGPUs anyway. So sadly no, I don't think we'll see much traction with Thunderbolt 5 until 2026,fingers crossed.
But somehow they managed to put TB5 into 14th gen CPU in Razer 18, right? But anyway we need to wait for the first TB5 eGPU
@@trysometech52 what comedy. The Razer 18 is a desktop replacement laptop with a discrete GPU. It makes no sense to use an eGPU with it.
@@EnochGitongaKimathi Obviously no one is gonna use eGPU with it lol, I didn't mean that
I thought thunderbolt was 40gbps
Bro buy new razer blade 18 with thunderbolt 5 and then try e gpu on it
eGPUs with Thunderbolt 5 doesn't exist yet, we have to wait for it
@@trysometech52 i travels a lot
So i am stuck on a 4090 sff build with water cooling or a laptop ? But i want to play heavy demanding games ! So which one i choose !
@@NikhilKumar-yq1fx PC is always better in terms of performance, but even SFF will be a heavy brick. eGPU with a case will be almost SFF PC size.
So if you travel a lot, I'd stick to a powerful laptop and ignore eGPU topic.
thunderbolt 5 is going to be 80gbps
Most likely 64gbps. They already mentioned 4x4.0 PCIe lines in it
@trysometech52 i heard 80 and 120 best case scenario
@@warlockboyburns that's total bandwidth, but for PCIe devices it will be limited. Just like it was in case of TB3.
TB3/4 = 40 total => 32 for PCIe => 22-23 in reality
TB5 = 80 total (120 for one side data stream) => 64 for PCIe => ?? in reality
@trysometech52 oculink is 63gbps and thats like 90% of fullspeed for a 4090
At 5:27 why is the 4060 is not on the list?
Dunno, but it should be there
This video is filled with tons of small errors and maybe outright misinformation.
Also the benchmarks are confusing as hell, why not state what device use and not just bandwidth, bandwidth is meaningless, and tehre is no end graph to compare all games.
he says there are 3 types of connection types and lists Oculink and M.2 separately.
oculink and M.2 is the same exact thing, Oculink is not a protocol but connection type, extension cable.
he says Oculink is Gen 4, while in reality it can be anything,Gen 1, Gen 2, Gen 3 , Gen4, Gen 5, it depends on your PCIe and where you plug it and what cable adapter you use or if its factory made then to where they soldered it.
In his example of connecting to WiFi slot, he didn't said a word that WiFi connectors are PCIe Gen2 or 3 x1 [most moderns laptops usually have gen3 x1], they 4 times slower than standard m.2 and much slower than TB4/USB4.
The main types of connection options are these:
1] PCIe Direct [Full Speed]
2] PCIe extensions: Oculink, M2, PCIe Extension cable.
These are usually x4, but can go up to x16.
Oculink comes in x4 and x8 versions, if you plug a x8 Oculink adapter into full sized PCIe slot, you'll get Oculink x8, both the adapter and the cable available on amazon/ali and other sites, if you buy long PCIe x16 extender, can be a meter [with reDrivers even longer], you'll get x16 full speed. [There are MORE options, like the cables that used for SAS and U.2: Adapter > Cable > PCIe Dock]
3] TB4 [2200Mb/s to 2700Mb/s depends on the dock and the host]
4] USB4 [3500MB/s-3700Mb/s depends on the USB4 dock and USB4 Host]
Using the same brand of TB4 dock he has in his video the Host to device and device to host, on AMD platform will be 2500-2700MB/s [Legion Go].
Connected to Intel: 200-300MB/s LESS. [X1 Intel]
If you use USB4 dock, you Gain 1Gb: USB4 Dock on USB4 handheld like Legion Go you get 3500-3700MB/s Host to device and Device to Host.
USB4 on AMD USB4 host gives you the same performance as Gen3 x4 NVMe SSDs, you basically max the PCIe Gen3 x4 speed!
The only Benefit of Oculink vs USB4 only if Oculink supports PCIe Gen4, in case its pluged to Gen3 and your system has USB4, go with USB4 for ease of use [hot plug, but right now lacks charging].
USB4 is 1Gb faster, because USB4 doesn't dedicated 1Gb for USB passtrought, TB does it all the time, even if the dock has no USB ports!
Because of this protocol change even TB4 dock [like his in the video, i have the same one] that is pluged to USB4 Host [Legion Go/ X1 AMD] gains free 200-300Mb/s boost vs. Intel Host that uses native TB4.
Next gen Intel like Arrow Lake should come with Native USB4 host, so everybody will be happy, but right now AMD wins twice: TB4 on AMD is 200-300MB/s faster AND AMD has USB4 docks that 1gb faster than TB4 [sold on Ali for 100-120USD].
Also as i mentioned his benchmarks are confusing for most people, he should of written TB/Oculink/PCIe x16, instead of writing bandwidth, bandwidth doesn't say anything to most people.
Also it would be nice to get a slide with just numbers table that compares everything, even excel file screenshot.
I always take criticism or corrections positively, and if I'm wrong somewhere, it's certainly not to confuse people deliberately. Some of what you wrote I really hadn't heard of, but other than that you yourself are personally starting to confuse others from the beginning.
For example, I didn't say there are 3 types of connectivity, I said there are 3 main types of eGPU. These are completely different things. After that, a lot of what you wrote just doesn't make sense.
"benchmarks are confusing for most people" - What makes you say that? You're the only one writing about it :) The rest of us just looked up a little higher in the benchmarks and saw that it was already signed in the GPU name. And thanks to you, they will now realise that if Thunderbolt 3 has about 23gbsp, and they buy a USB4 eGPU for an AMD processor, it will work even better for them.
Oh and don't you agree that showing a 2060 that runs on gen3 and just signing ‘Oculink’ wouldn't be quite right?
@@trysometech52
whats up?
The right thing with Benchmarks is just say what GPU it is and what device/dock it runs on, thats how everybody does it and it will be great to have a summary so people can take screenshots of, but you do what you want to do.
3 type of eGPU means they unique, while M.2/Oculink is all the same thing, the difference is what PCI express extension cable you you use, M.2 can be Oculink with M.2 to Oculink adapter, its all same thing and makes no difference [and wifi connection is x1 so it should be mentionedy] hence its better to split them up by connectivity ie PCIe/TB/USB4 and actual real bandwidth [you can check bandwidth by using cuda benchamrk or AIDA].
if you need more info, benchmarks and tools [like fixing nvidia USB4/TB crashes aka error 43 on AMD], join iGPU.io its a great resource.
But hey, I mentioned that M.2. and Oculink are compatible and you can combine them :)
x1 speed on Wifi slot was a surprise for me, I heard this option from many sources and was sure that this is legit option. I'll add that info in the pinned comment. Thanks!
@@trysometech52 Honestly, I think the simple way you described things was perfect without getting in the weeds...you demonstrated a simple relationship between connections using the most powerful Nvidia GPU available and 4k resolution as the common denominator....you basically helped set expectations in a good way. Nice video, keep it up!
You do realize a 1+ minute sponsorship post means over 10% of your entire content is shilling?
It's a good video, and 10% isn't that much.
Hey man I think I have the same laptop as you the ThinkPad E14 gen 5 with 13700h can you make a egpu video on it or any ocuilink modification for it ?
Nah, mine is gen1. It's pretty old already and I gave it to my parents already :) That video fragments were from my older videos.