GEM12 only has two rear USB2.0 ports. If you need to connect a USB3.0 device behind the GEM12, you can only use a TYPE C hub or docking station. After using GEM12 for about a week, I found that the rear power supply TYPE C port has a serious problem.: The thickness arround TYPE C port is about 1.6mm, and the length of the TYPE C male plug of the GEM12 original power supply is 7mm. However, for most commercially available TYPE C hubs or docking stations, the length of the uplink TYPE C port is 6.5mm. This results in the TYPE C hub or docking station being unable to stably connect to the rear TYPE C female port of GEM12. In addition, the original power supply of GEM12 is only DC, but the DC interface is changed to TYPE C style. Therefore, if the original power supply of GEM12 is connected to the power input port end of the USB HUB or docking station, it will not be able to supply power at all, and you must purchase another one power supply of PD3.1 that can stably provide 120W power to the GEM12. It is difficult to find expansion devices and power supplies that meet both of these requirements。 So currently the rear TYPE C of GEM12 can only be connected to the original power supply. Unless the manufacturer can solve this problem.
the rear usb-c is full futured (power supply, data transmission, display) but we are forced to always use it as power supply? can you enlighten me on this?
I have two Gem12s, both with the 8845HS; one setup as my Plex/HomeLab server, running UNRAID, and the other as my primary gaming PC, running Windows 11 Pro. Both units are setup with the same DIY eGPU dock (RTX 4060s in both) with a pico PSU for the Plex machine and the HDPLEX 250W GaN PSU for my gaming PC. I have been beyond impressed by the performance of this little machine and I really think these mini PCs are the future for gaming and workstation use. The price to performance ratio is unparalleled. Power consumption hits 200W under absolute MAX load, and it requires some serious effort to achieve that amount of draw. When sitting at idle, they draw just under 40W at the wall. The only "flaw" would be the lack of USB 3.0 I/O on the back of the machine, but I haven't found that to be an issue in my use-cases.
“You know, I like what this Prius can do. But keep in mind that it’s just a Prius. Right, so let’s see what it can do when we strap a fighter jet engine to it.”
using my 8845HS GEM 12 with an old 1070 GPU for 100+ FPS easily at 1080p Gaming on Ultra or High settings in all the games I play.. fun times on the Mini's !
Using the RTX 4070 TI Super it's a better use case example than using anything higher that than. I think most people stop at about the 4070 Ti Super price point. Being it's pretty decent at entry 4K, great at 1440p & under $1000. Should do more videos with that card connected to these mini pc's as the peak card of use.
Modularity and repairability. Also, you'll bulk up bigger than Arnold Schwarzenegger. It's all about those gains. But seriously, not everyone wants to travel with a battery, carry waste keyboard, screen, or other parts. It is a very niche use but there for those that are tech savy enough to want it.
@@Bajicoy I work in tech, I would never carry an itx build, chunky psu cable, spend more on a portable usb c monitor, cables and adapters, mousepad, mouse, keyboard, microphone, headset on top of my stuff if I knew I'd be traveling.. It makes 0 sense, it's so much weight, you need a desk to set everything up, there's the risk of damaging pcie slots and cooling components, you lose a ton of reliability..
@@NukaOrQuantum I also like building computers, but I would never put myself through the inconvenience of carrying and using that, also you'll be paying more for luggage each time, plus all the headaches I meantioned in my other comment. For $650 you're not building a sff pc with a dedicated gpu that's worth while, and you're conveniently not including all the peripherals that you're taking with you, lets not kid ourselves..
There is also 65W-75W TDP Performance mode setting on the AAOSTART GEM12 in BIOS. While I'd only expect a few % improvement on the CPU, it should also offer a small boost on the 780m iGPU.
Your mini PC reviews are awesome and have really got me into buying one. Especially with everything they can do. I also bought an Asus G22CH and upgraded it to a 4070 Super and your review of it really sold me on the purchase. Im sure Asus Appreciates it 😁. Thanks for all you do. Semper Fi!
The Gem10 has the CPU fan permanenty stuck on, even when in sleep mode. Somthing to be aware of if you want a completely silent PC when using it on a low power mode.
100%.noise levels extremely important for such mini devices and honesty my main concern that keeps me away. But if there was some tests in this regard i might change my mind. 🎉
Pretty good in comparison to what else is out there. When mentioning things like Ryzen AI, I wonder how many reviewers know exactly what it is and how the numbers stack up in terms of how it affects real word usage.
Some of us may not have high budget for an external A-GPU, perhaps you can show some affordable ones featured along an equally affordable Mini PC? I am looking for an entry level but good-for-gaming Mini PC right now, having my eyes on the new SZBOX Cyberpunk themed Mini PC. At least it comes with USB-4.0 so it can support an A-GPU, right?
With Fallout 4, You should Have Gone To The Downtown Part Of The City. That's Where You'd Want To Benchmark. As The Outskirts Isn't A Good Representation.
Nice system. I do like that the oculink isn't an m.2 converter with a hole in the side of the case . It loses points for not specifying oculink type (I assume 4x since the 8x port is way wider).16 TOPS AI is also nice.
These chipsets are fully capable of gen 4x16 .. they make riser cabled that are gen 4x16 for regular pcs .. you telling me they can't make a connection and a cable that can do the same ? At least gen 4x8 if cabling tech is not there yet. That would get you way less bottleneck on mid tier gpus.. even with occulink i wouldnt go above a 4060
So correct me if i'm wrong, but my understanding is at higher resolutions, limitations of OCuLink bandwidth is less noticeable. it's only when FPS get near triple digits the bottleneck becomes much more apparent? Im looking for a semi portable system (i travel weekly from home. I remote desktop to my tablet or Quest3 to work environment. ) with eGPU capable of driving VR (ideally approx equiv 4k 90fps) so hoping something like 4090 at 4k would be able to drive cyberpunk 2077 to 60fps at 4k, on the GEM12.
ETA Prime could you please link the video on how to make your own OCulink? I tried searching your videos and I was having trouble finding it. Thanks for all your videos!!! I bought a mini pc from one of your videos last December and I would love to try and hook up and external video card using OCulink. Thanks again for everything!!!
There is another mini that AOOSTAR make, with a Ryzen 9 and three nvme slots and the occulink interface. With the Amazon coupon it comes close to the same price as this unit under review. Imagine a 12 tb NAS using nvme ssd, and a great igpu. Plus dual 2.5gb Ethernet to fully realize max transfer rates. Then imagine getting a TB to 10gb Ethernet box, unfortunately running around 200 usd, and the theoretical transfer rates over Ethernet go from 300 to 1250 MB/s. Not bad for a tiny box.
Can you do me a big favor? Can you test these mini PCS that have oculink? With Star citizen 3.23 and see what the difference is both with and without the GPU. That would be awesome, thank you!
Quite a bit more performance is lost with these setups using pcie 3.0 as they operate with x4 channels instead of x16 (normal desktop), so a pcie 3.0 setup of this type would be at 1GB/s x 4 = 4GB/s vs. 2GB/s x4 = 8GB/s for pcie 4.0
@@marcgtsr Well!, from what i can tell!, the pcie 3.0 performance should be similar to results of those using thunderbolt enclosures!, older pcie3.0/oculink adapters!, or usb 4.0! So find someone using a "2060super" on one of those setups!
@onizuka4534 Alienware laptops were the only devices that ever were capable of this, AFAIK. They had a proprietary port and sold an external GPU box and cable as an accessory. It even included a dedicated power supply. At the time, I think they went for around $300 USD, so very reasonable. It outperformed iGPUs by 20-30%. But the newer laptops didn't include this feature. I think it was too niche, and the laptops with a dedicated GPU were much higher performance by then. But in theory, you could still get a pretty decent, although heavy by modern standards, used laptop with a 4c-8t CPU that is a respectable PC, and get the accessory. You'd have a portable PC (laptop) with the option of running a modern GPU. Though with some games, but not many, you could have a bit of a CPU bottleneck. Though that bottleneck certainly wouldn't be worse than the sacrifice of PCIe lanes.
@@shodan6401 OCulink has as much bottleneck as running the card on a slot with less lanes has since it doesn't have to convert the signal to something else or make use of some proprietary technology. I wouldn't be surprised if the Graphics Amplifier is actually OCulink but using a proprietary cable.
Nice little box. Would be good for a home game server for streaming to a handheld while in bed or on the couch. With Thunderbolt 5/USB4 80Gbps finally hitting machines (so far only the 2024 Razer Blade 18), I don't see a bright future for Ocilink.
has anyone tried eGPU's on Thunderbolt 5? eGPU's were a thing since TB3 and while TB4 didn't improve on raw speed, both are inferior for eGPU's compared to USB 4 and USB 4 is inferior to Oculink all in actual gameplay testing. Oculink is also much cheaper to implement vs Thunderbolt not to mention you can have Oculink on AMD computers. Right now AMD is favored for gaming cpus and portable gaming handhelds (Intel's ARC GPU drivers are a complete shitshow, I had one) so I kinda don't see Thunderbolt 5 being that special because those mobile platforms are the ones that would make the most use of an external GPU.
Thunderbolt 4 40gbs possui cerca de 20gbps de transferência real através dos PCIE Lanes. Na melhor das hipóteses, Thunderbolt 5 80gbps terá a mesma velocidade do Oculink. Porém, é mais provável que o Thunderbolt 5 seja bem mais lento. One thing is theory, other is the reality. Maybe Thunderbolt 6 be faster than Oculink PCIE 4.0. But not today with TB5.
@@icaroleal7351 The current version of OCulink (OCulink-2) is the equivalent of PCIe 4.0 x8, which has a max speed of 16 GB/s. Thundebolt 5 bi-directional has a max speed of 10 GB/s and 15 GB/s in uni-directional mode. But OCulink has the potential of becoming even faster because it was already used in the server world as an interconnect between devices before it became available to us end users.
ETA PRIME, Could you please, please, please let me know what base did you use to put the GPU and PSU together? I have a similar setup but can't find a mounting base to put them together.
What i wanna see is AMD's finest GPU the RX 7900 XTX with the Oculink setup on this mini PC, i am thinking of getting ASRock Radeon RX 7900 XTX Taichi OC for this setup and if the 4070 or 7800 XT is the same performance then i might go with the 2 lower options, gonna go with AMD as i am a Linux user and AMD is better at Wayland on Linux, if i was more into x11 i would go with an Nvidia card as they are not opensource the drivers and Nvidia runs really well on Linux with the newest drivers(40 series only through)
It would be cool to see this mini-pc's performance on the GPD portable GPU. I likely won't be going out to buy a 4070 anytime soon and the portable GPU has the advantage of well ... being PORTABLE. LOL
*@ETA PRIME* If you have a RTX dual fan GPU. Like a 4070 TI Super. It'd be nice to see a tiny build (like the single fan 4060 tiny builds you've done). Whether it be in a super small sff case or a mini workstation style case. Backpack size 4070 build. Or even a setup like what you have here in this video but in a small enclosure or sff case to house all these components.
Hello ETA Prime, hope you are well. I have a question regarding the GEM12 & GPD G1 egpu. I bought GEM12 6900HX, I connected with the GPD G1 Oculink. My problem is, when I shutdown the GEM12, it reboots automatically. Is there a way that I can shutdown the GEM12 normally with the Oculink connected? Thank you.
...does it support USB4/TB eGPU...? My GMKTec K6 doesn't :( ...no PCIe Tunneling function... maybe this test can be added to your future Mini PC reviews :)
Great question. Is there a easy way to figure out if a system supports USB PCIe Tunnelling, or have to plug in ans try? Especially as most of these minipc lack a proper manual, not that all the usual Mobo manu properly list it
Benchmark request: could you run “llama.cpp” with Llama3-8B (Q4 or Q8 , maybe both) eg get it to generate a paragraph and report tokens per second . You might be able to do this through the ollama UI. This handy little LLM at 4bit runs on an entry level MacMini M1 8gb at 11toks/sec , fast enough for conversation. I bet these latest Ryder’s could do pretty well..
Even running on CPU on an old AMD 3100 cpu, I’m getting 8.86 told/second. Even if they don’t have GPU support it’ll be great to know how the CPU copes .. should be at least double given the DDR5 and core count
How do you properly power on the egpu setup? I know you need to power down the pc and egpu then plug in the oculink cable. Do you need to turn on the pc then the egpu or the egpu then the pc or does it matter which you do first? I just don't want to fry my unit by doing the process incorrectly 😅
You need to provide external power to the egpu, as the OCulink cable is just a PCIe lane on a cable. Ideally, you have to turn the GPU first, then the PC. Or, if you use USB C instead, the setup is hot swappable.
Many of the pcie to oculink adapter boards will have a jumper or a switch thus giving a choice to turn the gpu on manually or automatically with the host as one would any other desktop.
No checking if you can boost the CPU to higher power in the BIOS ? No overclocking options ? Your attention to detail has gone down in the recent months ... these reviews are no longer usefull ...
It's more like a showcase channel. These are probably all paid content. And he's also shilling shady key resellers, so not exactly high scruples there. I watch because it shows me what's coming down the pike, but I would never trust it as my only review if I ever considered buying one of these.
If you need to use OCULINK minisforum um780 xtx will be occupied by an M.2 slot The GEM will not take up a slot, i.e. a real OcuLink mini pc with two M.2 slots. And according to reliable information, the GEM12 series is selling very well, that's why the GEM12 series is always out of stock on their website
ETA; I always dig your videos. Question: I'm looking for an Oculink based GPU enclosure (RTX 3000 or 4000 series based). Not something DIY but a good clean solution that could sit in the living room next to something like this Mini PC. Do you know of any good options? I've been looking but it appears that only USB4 options are available and they're usually pretty expensive considering their lower bandwidth potential as compared to oculink. Keep up the great work sir. 🫡
I was looking for the answer to that question several months back and have found that unfortunately there is not. If you got any diy in you, could make something... cab etc. The only cheap, practical way is to use the pcie/oculink adapter $40-60, a $500-600 gpu and a $500-600 ryzen mini pc and you will get similar price/performance to a desktop - I personally like the exposed look with proper cable management, but to each their own.
@@ahilker1 Hey thanks for the feedback. I don't mind the DIY look but I have little ones. I can imagine my son losing a finger or cheese ending up in the fans 🤣🤣
You shouldn t mind that much about those beautifull overpriced egpu and buy those DIY because they are the most effective ones, like ADT -Link K43SG with PCIE 4 x4 and Oculink port, as you said all beautiful expensive egpu case are only PCIE 3 with thunderbolt port and sometime it s not even thunderbolt 4. With the DIY egpu you ll lose 0 to 10% max in performance unlike your beautiful egpu with worst spec where you ll lose lot more performances
These mini PCs are like pokemon cards, but you'll go broke trying to catch em' all.
For real! lol.
So just like Pokémon TGS then 😁
not if you realize most are the same. they are ALL crazy fast.
Still cheaper than laptops
GEM12 only has two rear USB2.0 ports. If you need to connect a USB3.0 device behind the GEM12, you can only use a TYPE C hub or docking station. After using GEM12 for about a week, I found that the rear power supply TYPE C port has a serious problem.: The thickness arround TYPE C port is about 1.6mm, and the length of the TYPE C male plug of the GEM12 original power supply is 7mm. However, for most commercially available TYPE C hubs or docking stations, the length of the uplink TYPE C port is 6.5mm. This results in the TYPE C hub or docking station being unable to stably connect to the rear TYPE C female port of GEM12. In addition, the original power supply of GEM12 is only DC, but the DC interface is changed to TYPE C style. Therefore, if the original power supply of GEM12 is connected to the power input port end of the USB HUB or docking station, it will not be able to supply power at all, and you must purchase another one power supply of PD3.1 that can stably provide 120W power to the GEM12. It is difficult to find expansion devices and power supplies that meet both of these requirements。
So currently the rear TYPE C of GEM12 can only be connected to the original power supply. Unless the manufacturer can solve this problem.
Thanks for pointing this out.
Stuff like this is why everyone should look for flaws online before they make a purchase. This will help a lot. Thank you.
the rear usb-c is full futured (power supply, data transmission, display) but we are forced to always use it as power supply? can you enlighten me on this?
I have two Gem12s, both with the 8845HS; one setup as my Plex/HomeLab server, running UNRAID, and the other as my primary gaming PC, running Windows 11 Pro.
Both units are setup with the same DIY eGPU dock (RTX 4060s in both) with a pico PSU for the Plex machine and the HDPLEX 250W GaN PSU for my gaming PC.
I have been beyond impressed by the performance of this little machine and I really think these mini PCs are the future for gaming and workstation use. The price to performance ratio is unparalleled.
Power consumption hits 200W under absolute MAX load, and it requires some serious effort to achieve that amount of draw. When sitting at idle, they draw just under 40W at the wall.
The only "flaw" would be the lack of USB 3.0 I/O on the back of the machine, but I haven't found that to be an issue in my use-cases.
“You know, I like what this Prius can do. But keep in mind that it’s just a Prius. Right, so let’s see what it can do when we strap a fighter jet engine to it.”
Eta Prime is a ricer now. Lol.
using my 8845HS GEM 12 with an old 1070 GPU for 100+ FPS easily at 1080p Gaming on Ultra or High settings in all the games I play.. fun times on the Mini's !
Using the RTX 4070 TI Super it's a better use case example than using anything higher that than. I think most people stop at about the 4070 Ti Super price point. Being it's pretty decent at entry 4K, great at 1440p & under $1000. Should do more videos with that card connected to these mini pc's as the peak card of use.
$650 ain’t bad for this. Would make for a solid travel PC, but I’d still personally build my own mini PC with a dedicated GPU
What's the point of a travel pc when laptops exist
Modularity and repairability. Also, you'll bulk up bigger than Arnold Schwarzenegger. It's all about those gains.
But seriously, not everyone wants to travel with a battery, carry waste keyboard, screen, or other parts. It is a very niche use but there for those that are tech savy enough to want it.
@@fran2911Because I like building PCs. I’m not saying it’s for everyone. Most laptops for $500-$650 are gonna be pretty lacking in terms of specs.
@@Bajicoy I work in tech, I would never carry an itx build, chunky psu cable, spend more on a portable usb c monitor, cables and adapters, mousepad, mouse, keyboard, microphone, headset on top of my stuff if I knew I'd be traveling..
It makes 0 sense, it's so much weight, you need a desk to set everything up, there's the risk of damaging pcie slots and cooling components, you lose a ton of reliability..
@@NukaOrQuantum I also like building computers, but I would never put myself through the inconvenience of carrying and using that, also you'll be paying more for luggage each time, plus all the headaches I meantioned in my other comment.
For $650 you're not building a sff pc with a dedicated gpu that's worth while, and you're conveniently not including all the peripherals that you're taking with you, lets not kid ourselves..
We need someone to create an affordable oculink JBOD enclosure so you can make these into a NAS.
11:30 you have the fortnite tag in fallout gameplay :)
I like that a lot of new models are realizing how important cooling DDR5 and Gen 4 SSD is from the last generation. Competition is good.
£2,238.25... No thanks
On link different brand and different mini PC, lol
💀
@NathOnGames Your comment made me skip the video.😂😂😂
It's 650€ what do you mean 2k?
@@valentinoohp It's £2,238 in the UK.
There is also 65W-75W TDP Performance mode setting on the AAOSTART GEM12 in BIOS. While I'd only expect a few % improvement on the CPU, it should also offer a small boost on the 780m iGPU.
Your mini PC reviews are awesome and have really got me into buying one. Especially with everything they can do. I also bought an Asus G22CH and upgraded it to a 4070 Super and your review of it really sold me on the purchase. Im sure Asus Appreciates it 😁. Thanks for all you do. Semper Fi!
They send You these computers like candy 😀
Glad to see the GEM12 here finally. This PC is my dilemma between this and the Minisforum V3 tablet.
Nice Video! There is so many miniPC brands out now that it's hard to keep up with them all. Can someone post a list of their favorite MiniPC brands?
Minisforum
Minisforum and now Aoostar, i would love to get this GEM 12 with an oculink port.
only one for me, Intel.
The Gem10 has the CPU fan permanenty stuck on, even when in sleep mode. Somthing to be aware of if you want a completely silent PC when using it on a low power mode.
You should make us hear the fan noise in every single of your Mini PC reviews.
100%.noise levels extremely important for such mini devices and honesty my main concern that keeps me away. But if there was some tests in this regard i might change my mind. 🎉
Pretty good in comparison to what else is out there. When mentioning things like Ryzen AI, I wonder how many reviewers know exactly what it is and how the numbers stack up in terms of how it affects real word usage.
Great video but it would have been nice to see you test the same games on both gpus rather than different ones.
Ryzen 9 PCs with 128 GBs RAM need more reviews please.
the 4x link doesnt seem to be affecting the performce too much.. . that cyberpunk fps is pretty much on the button at 1440p
Would you mind sharing the 3D print you use for the Oculink dock?
Thank you for now including COD in your benchmarks. :)
Would most definitely like to see comparisons with this to the gem 10... Trying to decide and there's not much info out there..
I love that you upload almost every day. I like to hit play all most days while I'm multitasking or about to pass out
Some of us may not have high budget for an external A-GPU, perhaps you can show some affordable ones featured along an equally affordable Mini PC? I am looking for an entry level but good-for-gaming Mini PC right now, having my eyes on the new SZBOX Cyberpunk themed Mini PC. At least it comes with USB-4.0 so it can support an A-GPU, right?
As I understand it, the Super 4070Ti for the Oculink port is superfluous! Can you tell me the most suitable graphics card for this Oculink port?
With Fallout 4, You should Have Gone To The Downtown Part Of The City.
That's Where You'd Want To Benchmark. As The Outskirts Isn't A Good Representation.
Nice system. I do like that the oculink isn't an m.2 converter with a hole in the side of the case . It loses points for not specifying oculink type (I assume 4x since the 8x port is way wider).16 TOPS AI is also nice.
These chipsets are fully capable of gen 4x16 .. they make riser cabled that are gen 4x16 for regular pcs .. you telling me they can't make a connection and a cable that can do the same ? At least gen 4x8 if cabling tech is not there yet. That would get you way less bottleneck on mid tier gpus.. even with occulink i wouldnt go above a 4060
More interested in turning this into an OPNsense box. What chips are running the NICs?
I'd have loved to see that iGPU with Battlefield.
Thanks for the video, very interesting. Could you make a video testing OneXGPU from Oculink?
So correct me if i'm wrong, but my understanding is at higher resolutions, limitations of OCuLink bandwidth is less noticeable. it's only when FPS get near triple digits the bottleneck becomes much more apparent?
Im looking for a semi portable system (i travel weekly from home. I remote desktop to my tablet or Quest3 to work environment.
) with eGPU capable of driving VR (ideally approx equiv 4k 90fps) so hoping something like 4090 at 4k would be able to drive cyberpunk 2077 to 60fps at 4k, on the GEM12.
ETA Prime could you please link the video on how to make your own OCulink? I tried searching your videos and I was having trouble finding it. Thanks for all your videos!!! I bought a mini pc from one of your videos last December and I would love to try and hook up and external video card using OCulink. Thanks again for everything!!!
There is another mini that AOOSTAR make, with a Ryzen 9 and three nvme slots and the occulink interface. With the Amazon coupon it comes close to the same price as this unit under review. Imagine a 12 tb NAS using nvme ssd, and a great igpu. Plus dual 2.5gb Ethernet to fully realize max transfer rates. Then imagine getting a TB to 10gb Ethernet box, unfortunately running around 200 usd, and the theoretical transfer rates over Ethernet go from 300 to 1250 MB/s. Not bad for a tiny box.
Using USB-C for 19V fixed power seems irresponsible. Someone is going to get a surprise if they plug that into their phone
USB-C has built-in voltage management to avoid this (I learned about it last week)
@@MarcosCodas Listeten to what he says in the video. It's not using PD (Power Delivery) It can not adjust the voltage down to 5V for you phone
@@motionthings my bad of that’s the case. And what an oversight!
Think this may end up banned in EU.
As it does not follow the proper standard.
Can you do me a big favor? Can you test these mini PCS that have oculink? With Star citizen 3.23 and see what the difference is both with and without the GPU. That would be awesome, thank you!
Would be nice to see a test with an older GPU, a PCIe 3.0! Like a 2060super ou something like that.
Quite a bit more performance is lost with these setups using pcie 3.0 as they operate with x4 channels instead of x16 (normal desktop), so a pcie 3.0 setup of this type would be at 1GB/s x 4 = 4GB/s vs. 2GB/s x4 = 8GB/s for pcie 4.0
@@ahilker1 Yes, that's why I would want to see a test! And a fell games need more bandwidth, specially these new PS ports to PC!
@@marcgtsr Well!, from what i can tell!, the pcie 3.0 performance should be similar to results of those using thunderbolt enclosures!, older pcie3.0/oculink adapters!, or usb 4.0! So find someone using a "2060super" on one of those setups!
It would be so awesome if once upon a time we could get something like OcuLink but without losing performance - external GPU with 16 PCieX lanes
OCulink is not a protocol. It's just a PCIe lane in cable form.
@onizuka4534
Alienware laptops were the only devices that ever were capable of this, AFAIK.
They had a proprietary port and sold an external GPU box and cable as an accessory. It even included a dedicated power supply. At the time, I think they went for around $300 USD, so very reasonable.
It outperformed iGPUs by 20-30%. But the newer laptops didn't include this feature. I think it was too niche, and the laptops with a dedicated GPU were much higher performance by then.
But in theory, you could still get a pretty decent, although heavy by modern standards, used laptop with a 4c-8t CPU that is a respectable PC, and get the accessory.
You'd have a portable PC (laptop) with the option of running a modern GPU. Though with some games, but not many, you could have a bit of a CPU bottleneck.
Though that bottleneck certainly wouldn't be worse than the sacrifice of PCIe lanes.
@@shodan6401 OCulink has as much bottleneck as running the card on a slot with less lanes has since it doesn't have to convert the signal to something else or make use of some proprietary technology. I wouldn't be surprised if the Graphics Amplifier is actually OCulink but using a proprietary cable.
Quick question. What docking station are you using, if you don’t mind me asking? Thanks!
I wish there was an actual Mni itx or mATX mobo availble with full PCIE5.0 X146 slot and desktop DDR5
Would be nice to see it tested with GPD G1 with Oculink.
Power usage? Linux support?
Nice little box. Would be good for a home game server for streaming to a handheld while in bed or on the couch.
With Thunderbolt 5/USB4 80Gbps finally hitting machines (so far only the 2024 Razer Blade 18), I don't see a bright future for Ocilink.
has anyone tried eGPU's on Thunderbolt 5? eGPU's were a thing since TB3 and while TB4 didn't improve on raw speed, both are inferior for eGPU's compared to USB 4 and USB 4 is inferior to Oculink all in actual gameplay testing. Oculink is also much cheaper to implement vs Thunderbolt not to mention you can have Oculink on AMD computers. Right now AMD is favored for gaming cpus and portable gaming handhelds (Intel's ARC GPU drivers are a complete shitshow, I had one) so I kinda don't see Thunderbolt 5 being that special because those mobile platforms are the ones that would make the most use of an external GPU.
Except USB is a protocol with extra overhead while OCulink is just a PCIe lane in a cable.
Thunderbolt 4 40gbs possui cerca de 20gbps de transferência real através dos PCIE Lanes.
Na melhor das hipóteses, Thunderbolt 5 80gbps terá a mesma velocidade do Oculink.
Porém, é mais provável que o Thunderbolt 5 seja bem mais lento.
One thing is theory, other is the reality.
Maybe Thunderbolt 6 be faster than Oculink PCIE 4.0. But not today with TB5.
@@icaroleal7351 The current version of OCulink (OCulink-2) is the equivalent of PCIe 4.0 x8, which has a max speed of 16 GB/s. Thundebolt 5 bi-directional has a max speed of 10 GB/s and 15 GB/s in uni-directional mode. But OCulink has the potential of becoming even faster because it was already used in the server world as an interconnect between devices before it became available to us end users.
Where is the link for this eGPU case video he mentioned during this video at 9:05? I can't find it.
9:18 what do you use to connect the GPU to your pc???
I see you have resizable BAR enabled. I cant find the relevant bios setting in mine. can you help?
Do you have any recommendations on what DDR5 memory to buy for this?
This looked great but having the oculink port on the front just makes things messy.
ETA PRIME, Could you please, please, please let me know what base did you use to put the GPU and PSU together? I have a similar setup but can't find a mounting base to put them together.
You should try ADT-Link K43SG, it s a cheap egpu with PCIE 4 x4 with Oculink port, it s perfect for this Mini PC.
I use these mini pc to replace my Nvidia Shields that nvidia refuses to upgrade the hardware from 2015.
What i wanna see is AMD's finest GPU the RX 7900 XTX with the Oculink setup on this mini PC, i am thinking of getting ASRock Radeon RX 7900 XTX Taichi OC for this setup and if the 4070 or 7800 XT is the same performance then i might go with the 2 lower options, gonna go with AMD as i am a Linux user and AMD is better at Wayland on Linux, if i was more into x11 i would go with an Nvidia card as they are not opensource the drivers and Nvidia runs really well on Linux with the newest drivers(40 series only through)
I wanna see one of those portable e-gpu's with a 4060-ti or 4070 in it and likewise have a spare NVME drive slot.
It would be cool to see this mini-pc's performance on the GPD portable GPU. I likely won't be going out to buy a 4070 anytime soon and the portable GPU has the advantage of well ... being PORTABLE. LOL
Why don't you include comparative benchmark graphs?
Another bestest evah miniPC... exactly the same as the 50 bestest evah before it...
*@ETA PRIME*
If you have a RTX dual fan GPU. Like a 4070 TI Super. It'd be nice to see a tiny build (like the single fan 4060 tiny builds you've done). Whether it be in a super small sff case or a mini workstation style case. Backpack size 4070 build.
Or even a setup like what you have here in this video but in a small enclosure or sff case to house all these components.
Hello ETA Prime, hope you are well. I have a question regarding the GEM12 & GPD G1 egpu. I bought GEM12 6900HX, I connected with the GPD G1 Oculink. My problem is, when I shutdown the GEM12, it reboots automatically. Is there a way that I can shutdown the GEM12 normally with the Oculink connected? Thank you.
...does it support USB4/TB eGPU...? My GMKTec K6 doesn't :( ...no PCIe Tunneling function... maybe this test can be added to your future Mini PC reviews :)
Great question. Is there a easy way to figure out if a system supports USB PCIe Tunnelling, or have to plug in ans try? Especially as most of these minipc lack a proper manual, not that all the usual Mobo manu properly list it
Are these suitable for music production?
Stay away. These “best ever mini pcs” come with shit tier thermal management/cooling, not ideal for music production work imo.
@@verfasor Too bad, thanks.
I'd be interested in Ai LLM tests. Stable diffusion and the new Meta 8b LLM model.
When you add a graphics card to a mini PC you realize that you're better off just making a mini ITX build
Can You try with RTX 4000 ADA Gen or RTX A2000?
8945HX said: Are you Ok？
Link to create the oculink dock videos….please
Thanks
only for US people ? I can only see the USB4 interface on the mini-PC in my country. There is no OCuLink here.
Is the 880M everncoming?
Can't find the video of how to make low cost Oculink egpu? Can you post the link? TIA
This is the review I am waiting😊
is it same power as a egpu from onexplayer?
What is optimal Gpu for oculink use?
Benchmark request: could you run “llama.cpp” with Llama3-8B (Q4 or Q8 , maybe both) eg get it to generate a paragraph and report tokens per second . You might be able to do this through the ollama UI. This handy little LLM at 4bit runs on an entry level MacMini M1 8gb at 11toks/sec , fast enough for conversation. I bet these latest Ryder’s could do pretty well..
Even running on CPU on an old AMD 3100 cpu, I’m getting 8.86 told/second. Even if they don’t have GPU support it’ll be great to know how the CPU copes .. should be at least double given the DDR5 and core count
can i use this for virtualization?
Man , you can test oculink fps and pci x16 fps????
so the oculink port is limited to x4 and cannot run at x16？
How do you properly power on the egpu setup? I know you need to power down the pc and egpu then plug in the oculink cable. Do you need to turn on the pc then the egpu or the egpu then the pc or does it matter which you do first? I just don't want to fry my unit by doing the process incorrectly 😅
You need to provide external power to the egpu, as the OCulink cable is just a PCIe lane on a cable. Ideally, you have to turn the GPU first, then the PC. Or, if you use USB C instead, the setup is hot swappable.
Many of the pcie to oculink adapter boards will have a jumper or a switch thus giving a choice to turn the gpu on manually or automatically with the host as one would any other desktop.
Excellent video, I'd like to see how powerful a console the iGPU will do.
No checking if you can boost the CPU to higher power in the BIOS ? No overclocking options ? Your attention to detail has gone down in the recent months ... these reviews are no longer usefull ...
He's never been useful. Doesn't respond to questions, and his position on tech has changed for the worst.
@@KRASHPROJECT some of the earlier ones where he actually experiments with power modes an shows bios settings are nice.
It's more like a showcase channel. These are probably all paid content. And he's also shilling shady key resellers, so not exactly high scruples there. I watch because it shows me what's coming down the pike, but I would never trust it as my only review if I ever considered buying one of these.
Yeah, i would also have liked to hear the fan, i wouldn t like to own a noisy mini Pc
Another excellent video !
yeah man!
This is faster than the um780 correct?
Pls don't stop testing FORZA I pretty much only play Forza xD
Love the content
Welp, i guess i gotta get rid of my Minisforum UM780 XTX for AOOSTAR Gem12 now because of 2 ssd & oculink..
I been looking at minisforum um780 xtx. Is this gem better then that?
If you need to use OCULINK
minisforum um780 xtx will be occupied by an M.2 slot
The GEM will not take up a slot, i.e. a real OcuLink mini pc with two M.2 slots.
And according to reliable information, the GEM12 series is selling very well, that's why the GEM12 series is always out of stock on their website
how many pins on the ram to upgrade it?
Add Alan Wake 2 for gaming testing.
Your amazon link is broken / incorrect.
Dragon's Dogma and FF7 Remake, on the external GPU's, please, and thank you.
Do we have to install oculink driver?
Another week, another of several mini PC's getting released lol
eGPUs are perfect for AIOs and Laptops.
helldivers 2 supports dlss?
hello prime I like te see some emulators running on this gem pc
Chimera OS?
Can you go into more detail about connection the psu
The link goes to a crazy "UK" Amazon listing.. cost = £2,238.25 lmao
could you test Apex Legends on these mini pcs?
ETA; I always dig your videos. Question: I'm looking for an Oculink based GPU enclosure (RTX 3000 or 4000 series based). Not something DIY but a good clean solution that could sit in the living room next to something like this Mini PC. Do you know of any good options? I've been looking but it appears that only USB4 options are available and they're usually pretty expensive considering their lower bandwidth potential as compared to oculink. Keep up the great work sir. 🫡
I was looking for the answer to that question several months back and have found that unfortunately there is not. If you got any diy in you, could make something... cab etc. The only cheap, practical way is to use the pcie/oculink adapter $40-60, a $500-600 gpu and a $500-600 ryzen mini pc and you will get similar price/performance to a desktop - I personally like the exposed look with proper cable management, but to each their own.
@@ahilker1 Hey thanks for the feedback. I don't mind the DIY look but I have little ones. I can imagine my son losing a finger or cheese ending up in the fans 🤣🤣
You shouldn t mind that much about those beautifull overpriced egpu and buy those DIY because they are the most effective ones, like ADT -Link K43SG with PCIE 4 x4 and Oculink port, as you said all beautiful expensive egpu case are only PCIE 3 with thunderbolt port and sometime it s not even thunderbolt 4. With the DIY egpu you ll lose 0 to 10% max in performance unlike your beautiful egpu with worst spec where you ll lose lot more performances
Some mod a Silent Passive heatsink for it.