Minisforum EliteMini UM780 XTX (AMD Ryzen 7 7840HS)

[Zareek]

@phxrider, nice job, and thanks for all the pics of the process. This is exactly why I expected it would be like and the reason I'm probably no going to do it myself. I used to mod things all the time, but I gave up a lot of my tools and work space for doing that sort of stuff a few years back.

I'm kind of surprised there are not any turn key oculink enclosures by now. As mini PCs gain traction, I really expect this market to take off. To me, it's kind of ideal. I'd like to see Minisforum do something like this with oculink and at least two m.2 drives at the same time. The lack of a turnkey oculink enclosure or reasonably priced oculink GPU and the fact you have to give up an m.2 slot to use it are the two things holding me back from jumping on this. Aoostar has a mini that has oculink and like three m.2 slots at the same, but it took me a year to get comfortable trying out Minisforum. Interestingly enough, it appears Aoostar is also making an oculink enclosure.

 

[phxrider]

Yeah I get it 100%... It was a lot of research and time, although I am nowhere near super-equipped tool-wise for this, I really had to improvise with very basic tools and did it on my kitchen counters (as you can tell from the pics haha).

I hope someone starts making a Razer Core X style Oculink enclosure, that would be nice, but I don't think that's going to happen. Oculink isn't suitable for the masses - remember we're talking about people who put their iPhones in an armored Otter Box and STILL manage to break the screen twice a year. Oculink is a server connector, it's not meant to be plugged and unplugged 50000 times and be handled by people who give zero thought to these things. It will have to be T-Bolt 5 or something that really makes this take off. Corporations consider things like this, return rates, likelihood of having to warranty it, etc and IMO they'd be scared of that with Oculink. For now, IMO it's going to be a hobbyist/nerd thing.

For some reason, it appended this to my last post.... but I think this is it for updates - I received the 1 meter Oculink cable and it works just fine. FYI, Oculink is supposed to be good for 1.5m without any signal loss, at least that's what I read somewhere while doing the legwork for this. I think it's past the point where I can edit the original post to say it's tested, as there is no option to do that.

Quick benchmark - this is FC6 @ 1440p on ultra everything with all the DXR (ray tracing) options turned ON. This is actually pretty representative of real world FPS, I've seen highs in the 90s and lows about 60 in real world play.

 

[HeadRusch1]

I played Far Cry 6 and Starfield, both played at very high FPS on Ultra with all the DXR options on in FC6, 70-80 FPS most of the time. Starfield may have dipped a little lower but I don't think it was much lower, it played like a real gaming PC, subjectively, performance didn't seem far off from the same GPU in a full size PC with x16 slots.

Had the same experience over USB 4.0 as well, the dock I paid $100 for off Alixpress was delivering nearly the same throughput as I was getting with my card running native in an x16 slot.....granted it's not an RTX 40 series or anything so maybe that's where you hit a wall, but I think if you are shooting for 60fps you won't have a problem using either technology with an eGPU, although it probably varies game ot game...not all of them are starved in the same ways.

I personally no need for an eGPU enclosure though....I mean you have the Power Supply, the octopus cables off the Power Supply, the GPU itself and your chosen interface (let's assume mini-motherboard style).....subtlty is out the window at that point Maybe you could throw Irony a bone and just put the mini PC inside of a PC case and just install everyting in the mounting brackets in the case
Inception that rig!

 

[phxrider]

I personally no need for an eGPU enclosure though....I mean you have the Power Supply, the octopus cables off the Power Supply, the GPU itself and your chosen interface (let's assume mini-motherboard style).....subtlty is out the window at that point Maybe you could throw Irony a bone and just put the mini PC inside of a PC case and just install everyting in the mounting brackets in the case
Inception that rig!

Let me guess... you're single?

The enclosure makes it easier to bring out for gaming and put away when not gaming - or just tuck the enclosure away somewhere, like now I have it on the chair where the power supply was in the pic with everything set up pre-enclosure, but tucked under the table where it's not visible. The whole point of the mini is to be able to have a neat-ish station on half of the dining room table that can be stashed away in 5 minutes if we have company and need the table as an actual dining room table. Before that, we each had our laptops and it was really a kludge and took up a lot of surface space right where you need it.

 

[Zareek]

So I had a moment of weakness, they had the 64GB/1TB model on sale, I grabbed one. I'm impressed with the performance per watt on these. I'm seeing 10 Watts at the wall while web browsing or writing this post. 15 Watts, watching a YouTube video. 20 Watts playing web based solitaire in another window while watching a YouTube video.

My end goal with this is to find a Linux distro that I can daily drive and leave the massive gaming PC for just gaming.

 

[phxrider]

So I had a moment of weakness, they had the 64GB/1TB model on sale, I grabbed one. I'm impressed with the performance per watt on these. I'm seeing 10 Watts at the wall while web browsing or writing this post. 15 Watts, watching a YouTube video. 20 Watts playing web based solitaire in another window while watching a YouTube video.

My end goal with this is to find a Linux distro that I can daily drive and leave the massive gaming PC for just gaming.

Nice, I need to take the wall power meter off of my PC and try it out on mine.

With the eGPU, I could legit get by with one of these as my main gaming PC, and my "real" gaming PC is a 7800x3d/7900xtx, so that's saying something. Any game I'm interested in is going to be almost entirely GPU limited anyway, I don't play eSports type low res / low graphics settings / super high FPS stuff that needs a lot of CPU. I keep the eGPU turned off and the monitor plugged directly into the mini when I'm not gaming - even on my primary PC, it wouldn't be much of a hardship to shut it down, flip the power switch and swap the monitor cable to the box when I want to game. Honestly, I can see this being the wave of future gaming PCs in an increasingly power-conscious world, when fast eGPUs become commercially available.... with the right marketing to the right demographics. If the big computer vendors don't do it, some smaller, relatively unknown Chinese company will (maybe even Minisforum).

On another note, this mini boots (well technically, POSTs) WAY faster than my AM5 desktop computer. It's about as fast as AM4 at POST - in other words, you flip in on and in a few seconds you see the manufacturer logo and Windows donut. It's this way regardless of whether it's on internal or external graphics.

Have you seen any of the supposed leaks for the Strix Point APUs? If those are to be believed, they're talking about a 4070 level iGPU, that would sure make an epic mini PC (although I don't know if the mobile ones that would go in a mini will be that powerful, IIRC that one was a desktop chip - still, AMD's APUs tend to make it to mobile form at a lower TDP and clock speed). I still like the idea of being able to shut down the high-powered GPU entirely, though.

 

[phxrider]

Benchmarking for the UM780XTX... Results columns are with eGPU on the left and without on the right.

Noteables:
* This is a barebones model with Crucial memory (same type as stock) and Silicon Power US75 SSD (much faster than stock).
* I looked at their database and this PC getting into the 31,000s range for CPU Mark is significantly faster than the typical/average 7840HS.
* This PC as-is (without an eGPU), scores above 80% of all scores in their database.
* Adding the 6800xt pushed the Passmark score up over 4000 points, or 50%.
* With the 6800xt, it scored above 90% of all scores in their database.

Not bad for a mini-PC!

Due to the fairly low number of submissions using this CPU, the average was in the 28K range, and my submissions pushed it over 29K.

I submitted 2 additional ones that I deleted from my submissions, due to not having known beforehand that you have to manually tell Passmark to use the C: drive in preferences when running from a USB (which you can legitimately do per your license), or else it will benchmark the USB drive instead of the SSD, leading to an abysmal Diskmark score, which drags the overall Passmark score down to like 1500 (which is like 10th or 15th percentile). I am not sure if deleting them from my submissions removes them from the average of all subs, if they are still in there, there will be a few more >31K CPU Marks considered in the average for 7840HS. Ironically, the Samsung USB thumb drive still benched better than 15% of all drives, but a score in the 900s vs over 51K for the US75 SSD killed the overall performance score.

Baseline # ending in 55 is with eGPU with 6800xt, ending in 70 is the standard internal Radeon 780M.

 

[phxrider]

Experimenting with making an RGB template using a laser printer and standard paper..... The bleed-through in the black area actually gives it a cartoonish look, and it shows stronger in photos than to the eye. (We have a bit of a poop emoji fetish in our house. )

 

[Zareek]

Experimenting with making an RGB template using a laser printer and standard paper..... The bleed-through in the black area actually gives it a cartoonish look, and it shows stronger in photos than to the eye. (We have a bit of a poop emoji fetish in our house. )

Lol, I thought about doing this, but I just decided to turn the damn LED off. I find it incredibly annoying that it just cycles through colors. No real control, so it is just dumb, IMO. I'm using this as a chance to try legit running Linux as my daily driver, and it is going great for the most part.

 

[Lew Zealand]

Lol, I thought about doing this, but I just decided to turn the damn LED off. I find it incredibly annoying that it just cycles through colors. No real control, so it is just dumb, IMO. I'm using this as a chance to try legit running Linux as my daily driver, and it is going great for the most part.

Same here, if I could choose a static color or even which colors to cycle and how quickly, then I'd consider designing a nice top plate. Off is a great backup choice at least.

 

[phxrider]

Lol, I thought about doing this, but I just decided to turn the damn LED off. I find it incredibly annoying that it just cycles through colors. No real control, so it is just dumb, IMO. I'm using this as a chance to try legit running Linux as my daily driver, and it is going great for the most part.

Haha, this is less obtrusive than the tiger, due to the fact that it's a white piece of paper and not completely clear, so it cuts down on the light that can pass through it. I turned it off originally until I decided to try this, now it's just noticeable enough to catch our eyes once in a while and crack us up (my gf and I). I do wish it had more control over the lighting available, with any luck it's software addressable and someone will write a utility for it at some point.

 

[phxrider]

Update: I spent $700 so you don't have to....

I decided to upgrade to a 7900XT in the eGPU. My advice would be not to, with some reservations. The uplift is not what it should be over a 6800XT. With the 6800XT, game performance was nearly or exactly the same in this eGPU with the 7840HS as when I had the GPU in a full-fledged 16x slot with a 7800X3D. With the 7900XT, I see major CPU limitation and rarely full GPU utilization.

I test with 2 games, Far Cry 6 and Starfield. It had exactly ZERO uplift in FPS in Far Cry 6, and GPU util doesn't get near 100%. FC6 is notorious for not multithreading well, so this makes perfect sense. Starfield had some uplift, but not what you'd expect. I am able to play at 1440 with no FSR now, but in the most demanding, lower FPS areas such as Akila City and New Atlantis, FPS can drop into the 50s and you can see it's only using ~80% of the GPU, whereas my 7800X3D system with 7900XTX is using 100% of a faster GPU in them. You can get near 100% GPU in simpler areas that aren't as CPU intensive.

So I think the RTX3080/6800XT level is about as high as you can go with these Mini PCs with mobile processors without taking a performance hit from the CPU not being able to keep up. This means with AMD, the 7900GRE would be about as high as you'd want to spend the money for (seeing that it's only a few dollars higher than the 7800XT). I guess that would be the 4070 Super for Nvidia? This may shift upward depending on how GPU vs CPU limited the games you play are.

Also note, with this enclosure, you need to stick with thinner card models that only screw into 2 slots, i.e. reference or Sapphire Pulse.

 

[HOkay]

Update: I spent $700 so you don't have to....

I decided to upgrade to a 7900XT in the eGPU. My advice would be not to, with some reservations. The uplift is not what it should be over a 6800XT. With the 6800XT, game performance was nearly or exactly the same in this eGPU with the 7840HS as when I had the GPU in a full-fledged 16x slot with a 7800X3D. With the 7900XT, I see major CPU limitation and rarely full GPU utilization.

I test with 2 games, Far Cry 6 and Starfield. It had exactly ZERO uplift in FPS in Far Cry 6, and GPU util doesn't get near 100%. FC6 is notorious for not multithreading well, so this makes perfect sense. Starfield had some uplift, but not what you'd expect. I am able to play at 1440 with no FSR now, but in the most demanding, lower FPS areas such as Akila City and New Atlantis, FPS can drop into the 50s and you can see it's only using ~80% of the GPU, whereas my 7800X3D system with 7900XTX is using 100% of a faster GPU in them. You can get near 100% GPU in simpler areas that aren't as CPU intensive.

So I think the RTX3080/6800XT level is about as high as you can go with these Mini PCs with mobile processors without taking a performance hit from the CPU not being able to keep up. This means with AMD, the 7900GRE would be about as high as you'd want to spend the money for (seeing that it's only a few dollars higher than the 7800XT). I guess that would be the 4070 Super for Nvidia? This may shift upward depending on how GPU vs CPU limited the games you play are.

Also note, with this enclosure, you need to stick with thinner card models that only screw into 2 slots, i.e. reference or Sapphire Pulse.

Thanks for the insight, it's interesting to hear some real world results. I had a similar experience plugging my 4090 into a TB3 enclosure connected to my Legion Go. The tldr is that you get CPU &/or connector bandwidth limited at fairly modest (by today's standards) frame rates. A high refresh rate gaming experience it is not! I'd agree with your 6800XT/3080 rough guideline as the best GPU that's worth it. I've got a 4060ti 16GB in there now & that gives me a chunky uplift over the iGPU without hitting the limits too hard & giving too much in the way of inconsistent frame times.

 

[crazyeyesreaper]

Update: I spent $700 so you don't have to....

I decided to upgrade to a 7900XT in the eGPU. My advice would be not to, with some reservations. The uplift is not what it should be over a 6800XT. With the 6800XT, game performance was nearly or exactly the same in this eGPU with the 7840HS as when I had the GPU in a full-fledged 16x slot with a 7800X3D. With the 7900XT, I see major CPU limitation and rarely full GPU utilization.

I test with 2 games, Far Cry 6 and Starfield. It had exactly ZERO uplift in FPS in Far Cry 6, and GPU util doesn't get near 100%. FC6 is notorious for not multithreading well, so this makes perfect sense. Starfield had some uplift, but not what you'd expect. I am able to play at 1440 with no FSR now, but in the most demanding, lower FPS areas such as Akila City and New Atlantis, FPS can drop into the 50s and you can see it's only using ~80% of the GPU, whereas my 7800X3D system with 7900XTX is using 100% of a faster GPU in them. You can get near 100% GPU in simpler areas that aren't as CPU intensive.

So I think the RTX3080/6800XT level is about as high as you can go with these Mini PCs with mobile processors without taking a performance hit from the CPU not being able to keep up. This means with AMD, the 7900GRE would be about as high as you'd want to spend the money for (seeing that it's only a few dollars higher than the 7800XT). I guess that would be the 4070 Super for Nvidia? This may shift upward depending on how GPU vs CPU limited the games you play are.

Also note, with this enclosure, you need to stick with thinner card models that only screw into 2 slots, i.e. reference or Sapphire Pulse.

It has less to do with CPU and everything to do with bandwidth / latency etc and power usage. CPU can be a limiting factor due to its max TDP in certain situations but this typically only applies to cards at the Ultra high end. That said, what few people consider is the latency and overhead that comes with Thunderbolt 3 / USB4 implementation. This can cause further performance drop in various titles. Buying a $700 mini-PC then a $300 enclosure and adding a $2000 GPU isn't really a great way to get good performance or value. eGPUs basically shine in the midrange as you yourself have now found out. Going beyond a certain performance threshold nets almost no difference except in certain outlier scenarios. OCulink from what I have seen is a bit more stable. The performance drop is less noticeable. eGPUs are a great mid range option where you can take 1 GPU and get multiple uses from it but the technology still has some teething issues. Put another way if you take a 4090 and use it as an eGPU you could see as much as a 50% drop in performance compared to use in a desktop. When using Thunderbolt 3 / USB4. Swapping to OCulink that performance difference shrinks to about 20% granted this is at 4K resolution. Lower resolutions yeah you will likely run into issues due to CPU but even high end desktop CPUs are a limiting factor at lower res. Just something to think about overall.

TLDR: Performance drop is normal going beyond mid range GPUs offers little value or sense depending on the setup. USB4 / Thunderbolt 3 with overhead complicate matters OCulink is proving to be better overall in terms of raw performance stability but adoption is limited for now. So expect the technology to advance but it will continue to have teething issues.

 

[phxrider]

It has less to do with CPU and everything to do with bandwidth / latency etc and power usage. CPU can be a limiting factor due to its max TDP in certain situations but this typically only applies to cards at the Ultra high end. That said, what few people consider is the latency and overhead that comes with Thunderbolt 3 / USB4 implementation. This can cause further performance drop in various titles. Buying a $700 mini-PC then a $300 enclosure and adding a $2000 GPU isn't really a great way to get good performance or value. eGPUs basically shine in the midrange as you yourself have now found out. Going beyond a certain performance threshold nets almost no difference except in certain outlier scenarios. OCulink from what I have seen is a bit more stable. The performance drop is less noticeable. eGPUs are a great mid range option where you can take 1 GPU and get multiple uses from it but the technology still has some teething issues. Put another way if you take a 4090 and use it as an eGPU you could see as much as a 50% drop in performance compared to use in a desktop. When using Thunderbolt 3 / USB4. Swapping to OCulink that performance difference shrinks to about 20% granted this is at 4K resolution. Lower resolutions yeah you will likely run into issues due to CPU but even high end desktop CPUs are a limiting factor at lower res. Just something to think about overall.

TLDR: Performance drop is normal going beyond mid range GPUs offers little value or sense depending on the setup. USB4 / Thunderbolt 3 with overhead complicate matters OCulink is proving to be better overall in terms of raw performance stability but adoption is limited for now. So expect the technology to advance but it will continue to have teething issues.

When I can see 100% CPU and only 80% GPU being utilized (Starfield), I'm pretty certain it's CPU.

Far Cry 6 never uses more than ~40% CPU and 80% GPU, but it's the same ~40% CPU use on my 7800X3D even when GPU is at 100%. But as I said, FC6 is notorious for being single-core CPU limited.

TDP would still be CPU limiting. (There is certainly no TDP limit on the 7900XT with a high end 750w PSU powering only the GPU.) The CPU in the mini is able to go up to 70W so it's on par with a desktop CPU - the major limit on AMD's APUs is the smaller cache than the desktop CPUs, it has about half the cache of what a 7700 has. As we know from comparing 8700G vs 7700 vs 7800X3d, cache is king with gaming.

 

[Lew Zealand]

It has less to do with CPU and everything to do with bandwidth / latency etc and power usage. CPU can be a limiting factor due to its max TDP in certain situations but this typically only applies to cards at the Ultra high end. That said, what few people consider is the latency and overhead that comes with Thunderbolt 3 / USB4 implementation. This can cause further performance drop in various titles. Buying a $700 mini-PC then a $300 enclosure and adding a $2000 GPU isn't really a great way to get good performance or value. eGPUs basically shine in the midrange as you yourself have now found out. Going beyond a certain performance threshold nets almost no difference except in certain outlier scenarios. OCulink from what I have seen is a bit more stable. The performance drop is less noticeable. eGPUs are a great mid range option where you can take 1 GPU and get multiple uses from it but the technology still has some teething issues. Put another way if you take a 4090 and use it as an eGPU you could see as much as a 50% drop in performance compared to use in a desktop. When using Thunderbolt 3 / USB4. Swapping to OCulink that performance difference shrinks to about 20% granted this is at 4K resolution. Lower resolutions yeah you will likely run into issues due to CPU but even high end desktop CPUs are a limiting factor at lower res. Just something to think about overall.

TLDR: Performance drop is normal going beyond mid range GPUs offers little value or sense depending on the setup. USB4 / Thunderbolt 3 with overhead complicate matters OCulink is proving to be better overall in terms of raw performance stability but adoption is limited for now. So expect the technology to advance but it will continue to have teething issues.

It may not even be that going past a certain level of GPU performance is the problem as, at least as I've tested with Thunderbolt 3, some specific games don't affected much about the latency introduced by an eGPU, just dropping framerate and increasing frametimes by a little, while others are completely crippled into uselessness by it with excessive frametime instability. Oculink is not supposed to be plagued by the latency induced by the TB3 translation layer but I really want to test this out for myself.

I started dGPU PC gaming with an eGPU and the games I was playing in 2018 simply were a little slower on the external setup when compared to building a real gaming PC less than a year later. Ark: SE, Rocket League, Dirt Rally, Minecraft, Tomb Raider '13, others. Testing those same games on my UM780 XTX with a faster eGPU sees similar performance reductions today. Lower but still plenty playable.

But testing CP2077 and Hogwarts Legacy and some others makes the eGPU experience useless as I'm guessing DX12 and modern game design makes the GPU and CPU exchange a lot more information, the achilles heel of eGPUs. Again I'd like to see if it's more latency or bandwidth-limited and Oculink could provide that clarity. I suspect latency which means Oculink could be a good deal better (up to 6800 XT or so as described by @phxrider ), but I need to get a parts list together for Oculink to know for sure.

 

[phxrider]

I got around to doing some messing around with my Razer Core X with a 6800xt, and the result was a resounding "don't bother". Oculink is SO much better performing, it's basically ridiculous to compare the two. T-Bolt 3 was slow enough that there was no difference between running the game on the eGPU and routing it back through the iGPU to the monitor and connecting the monitor directly to the eGPU (like having a mux switch vs not having one on a gaming laptop). I could get ~30 FPS in games I can get 80 FPS in with Oculink. For full transparency, I hooked it up to my Minisforum V3 tablet and not the UM780XTX, but they shouldn't be THAT much different (~150%), both are USB4 and similar CPUs other than TDP is higher on the mini, I would say the TDP difference would be good for MAYBE 25% difference.

BTW with the latest Starfield updates I'm getting in the high 100s FPS on Ultra everything with FSR3 at 100% scaling (basically same as turning FSR off) and in-game frame generation on, with perfect responsiveness on Oculink with 7900xt. Sometimes it pushed the refresh rate of my monitor @ 165 Hz, I turned it up to 180Hz to avoid that.

Update: re V3 tablet vs UM780XTX performance - I was curious what the actual CPU difference is, so using CPU-Z's built in benchmark (yes I know this isn't the greatest benchmark, but it's fine for apples-to-apples comparisons like nearly identical AMD CPUs), The mini (7840HS) is 27% faster in multicore, that's very in line with expectations, but the big surprise was the V3 (8840U) was actually 1.3% faster in single core!