Clevo + eGPU results

I recently added a eGPU to my Clevo laptop, and thought I'd post some thoughts and results...

My goals

• Good performance -- I wanted to play the latest games and do some tensorflow work. I have a 1080p, 75Hz display, so that's the top end of my perf requirements.
  
• Quiet -- I generally use my laptop on the couch where my wife is sitting and reading/game/watching TV. Loud fans are pretty much a no-go.
• Cool -- it's annoying, especially in the summer, when my laptop turns into a blazing furnace whenever I try to play anything.

Not my goals

• Price -- I didn't want to spend an insane amount, but cost wasn't a big consideration. Particularly if it lets me delay upgrading my laptop, which I tend to do every couple of years.
• Benchmarking -- I want performance that lets me play what I want to, and play it well, but I'm not really interested in chasing benchmark scores.

What I started with
I have a Clevo P775DM3-G that I purchased a couple years ago. It's big 17" laptop that has a 6700k desktop process and mobile 1080. I generally use it on the couch or on a desk, either at home or work. Yes, it's heavy and big. No, it doesn't bother me.

The big problem with using this laptop for gaming (or any other really intensive activity) is that, on the "Normal" fan setting (ie. usably quiet), it ends up thermally throttling both the CPU and the GPU. As far as I can tell, this is just a limitation of the laptop itself -- it's always done this, the heatsinks are clean, thermal paste is good. It's just that running a desktop processor and a 1080 in a small enclosure means you need to blast the fans to have any hope of keeping up. And on top of the annoyance factor, it's not great to be getting your CPU and GPU to their their thermal throttling points regularly.

So, I decided to give an eGPU a try. I'm well aware of the performance hit that you take when you use an eGPU, but in doing some testing with a GTX980 that I had sitting around, it looked like it would improve my perf while also keeping temperatures and noise levels low, so I took the plunge.

What I bought
I ended up picking up the Razer Core X ($300) and a RTX 2080 TI ($1200). The Core X is a simple, clean little box with a few outputs and a good power supply. Basically a no-muss, no-fuss eGPU, that also happens to be on the lower end of the price spectrum of eGPUs. I totally recommend it.

The 2080 TI is almost surely overkill, but a) I knew I was going to be taking a perf hit by using a eGPU, and I wanted to make sure it was going to improve my current performance, and b) money wasn't a huge issue and like I said, especially if this gets me to delay another $3000 laptop purchase for a year, it ends up making more sense. The specific model was the Gigabyte Gaming OC, mostly because it was the one that briefly popped up in stock on Amazon.

How it works
So I got the Core X, put the 2080TI in it (comfortably fits), and bought the 6' powered TB3 from Cable Matters (~$60). When I plugged it into the TB3 port on my laptop, Windows first installed some drivers and then popped up a dialog:



Well, THAT message doesn't look promising. Thankfully, it's a total lie and eGPUs work great on this computer! Don't worry about it.

Then I had to install desktop video drivers from nvidia, since previously I only had the mobile ones installed.

The only thing you really need to do is make sure that you're on the 1803 version of Windows, which allows you to specify which GPU to run on. Annoyingly you have to do this on an execuable-by-executable basis, but it works great, and it remembers all the settings so you don't need to do it more than once for anything that you want to run. Basically just go to Desktop Settings and select Graphics Settings at the bottom:



Then you can Browse to apps and add them to the list:



And then if you select Options, you can pick which GPU to run on (this was when I was testing with the 980):


And that's it. Then you can just run whatever you want, and it'll render it over on the eGPU and use your local GPU for actually rendering it to your display. You can also use an external display, which will get you slightly better performance, but it wasn't what I was interested in.

And now, I can run whatever I want, and I just need to plug in the eGPU cable. The box itself is sitting next to me behind the end table, and is almost completely silent, even under heavy load. And whereas previously I'd end up with my CPU around 95C and my GPU at 90C, both thermally throttling, now my CPU never goes above 80C and neither GPU breaks 75C, all while on the lowest fan setting! It's great.

Of course, I did some benchmarks to see what kind of performance difference there was between the throttling mobile 1080 and the eGPU 2080 TI:

Ashes of the Singularity:




Far Cry 5 (fairly CPU-limited, it seems):




Fire Strike:




Superposition (Extreme):




Timespy:




Shadow of the Tomb Raider:




Overall

I couldn't be happier! It works great, and does everything I was looking for. Happy to answer any other questions or do other benchmarks, if people are interested.

 

Wow. Your FS Graphics Score with the 2080 Ti eGPU is lower than what I get on my internal 1080 at stock...

Edit: Looks like your 2080 Ti is losing about 40% of its performance

Spoiler

Did you try using an external monitor hooked up to the 2080 Ti instead of the laptop display?

Your internal 1080 is also being throttled by almost 30%.

 

Yep, like I said, I'm aware of the perf loss. But thermal throttling is losing a lot too.

Didn't try using an external monitor -- I'd just buy a desktop machine if I wanted that.

The other interesting thing is that it varies a lot by benchmark. Firestrike is a pretty big drop, but something like Superposition shows a much smaller drop (~18%), and Time Spy is smaller too (~20%). And then Far Cry isn't much faster than my 1080 at all, but Shadow of the Tomb Raider is a LOT faster.

 

I suggested an external monitor hooked directly to the 2080 Ti because it would probably not destroy performance as badly.

But idk. $1600 ain't exactly chump change, and seems like a hugely unnecessary expenditure versus fixing what you already had, which would've gotten you better performance and mobility to boot. A set of P775TM1 heatsinks, liquid metal, and a delid tool would've gotten the job done, then you could spend the rest building a kickass desktop with a discounted 1080 Ti.

 

I guess I'm not convinced I could have gotten any better performance out of my existing laptop -- it's already using liquid metal on both the CPU and GPU, and afaik you can't delid the GPU anyways? That's a lot of work, not without some risk, and there's no guarantee I end up with anything better than where I started.

Anyways, this is the eGPU section of the forum is it not? Not much point telling people that they shouldn't use a eGPU. I was pretty clear about my goals, and given that I now have a silent laptop that's faster than any other (single-gpu) laptop that I'm aware of, I'm pretty happy. Mostly I posted in case others were interested in trying it themselves, so they can see the tradeoffs and what to expect.

 

That's completely normal behavior. The higher the FPS, the more the eGPU gets bottlenecked by the TB3 interface. Superposition and Time Spy don't see as much of a drop as Fire Strike because they run at lower frame rates.

Your bottlenecks in FC5 and SotTR are huge at 1080p. In FC5, the desktop 2080 Ti gets 130 FPS to your 83 FPS (-36%). In SotTR, the desktop 2080 Ti gets 129 FPS to your 77 FPS (-40%), and you are GPU bound only 30% of the time, so 70% of the time you are bottlenecked by the eGPU interface or CPU. In both cases, your 2080 Ti is still slower than a non-throttling 1080, in FC5 massively so.

Your poor 1080 is throttling so badly, even a massively bottlenecked 2080 Ti eGPU looks amazing next to it. Compared to a non-throttling 1080, your 1080 is getting 70% of its normal performance in FC5 and FS, and only 50% (!) of its normal performance in Superposition, Time Spy, and SotTR.

Um excuse me? LOL. You have gotten much better performance out of your existing laptop. This is what you scored in Fire Strike when you first got the laptop 2 years ago. Huge difference. I think it's pretty obvious that you need to take off the heatsink, clean out the fans, and repaste. That should solve your thermal problems, assuming the CPU is already delidded with LM under the IHS and undervolted, and the GPU is also undervolted. If cleaning and repasting don't fix it, like I said, get the upgraded heatsinks from the P775TM1, those will drop your CPU and GPU temps under load by another 10C.

Don't get me wrong, I'm not criticizing your little eGPU experiement. I'm all for doing things in the name of science and providing more information (even if it's the same information that's been out there for a long time regarding TB3 bottlenecking any eGPU faster than a 970). Getting 1070 level gaming performance out of a $1600 2080 Ti eGPU setup is horrendous, when most people can simply purchase an entire 1070-based notebook for that price, and just serves to further validate the fact that if your laptop has a GPU faster than a 1060, any TB3 eGPU setup is a waste of money. It's just not financially practical for most people, nor necessarily useful since your Clevo's abysmal GPU performance is an outlier to begin with and easily fixable, not to mention reducing mobility and convenience.

 

Look at the results again, interestingly enough only firestrike is bottlenecked that hard, the rest is utterly destroying the 1080. Look at timespy for instance.

It was the right decision. Since non nerfed high end RTX cards won't exist for MXM this was the only right thing to do. Modding in your notebook would have been waste of time and you couldn't ever get the performance you have now, not only that but eGPU is still in it's "baby shoes" it will get better and better over the years with drivers etc.

In this community there are people who insists that notebook hardware can keep up with desktop hardware, which is an absolute joke. As you can already read from yrekabakery's post, he is absolutely ignoring facts and claims stuff that doesn't make any sense whatsoever.

Best example here:

This is the highest modded TDP unlocked and clocked notebook GTX 1080
https://www.3dmark.com/spy/3766000

meanwhile you beat it with ease, you probably have some bottlenecks in your RTX 2080TI that aren't only bandwidth bound tho.

 

FC5 and SotTR show similar 36% to 40% bottlenecks with performance well below a non-throttling 1080. TimeSpy and Superposition show little bottlenecking because they are running at lower frame rates. The severity of the bottleneck scales with frame rate, that's why a powerful TB3 eGPU setup is far more suited to 4K 60Hz than 1080p 144Hz, and external monitor plugged directly into the eGPU reduces the bottleneck as well.

Even a 2080 Ti bottlenecked by that much still beats his 1080 (albeit just barely in FC5), because his particular 1080 is throttling so badly that it's only getting 50% to 70% of its normal performance.

 

Mr.fox modded GTX 1080 doesn't get nowhere close to his 2080TI in timespy for instance. So stop with that bogus.

Also he probably has some settings wrong etc. eGPU can be a little fiddling work, I will help the guy out and if he accepts my help, He will probably show u some better numbers.

also yes, external screen does reduce boitteneck even further.

 

@Mr. Fox doesn't have a 2080 Ti. And read what I wrote first before posting bogus. I'm talking about 2080 Ti performance inside a TB3 eGPU enclosure, not inside a desktop. I've already explained to you why the eGPU is less bottlenecked in Time Spy and similar low frame rate synthetic benchmarks than in actual games running at more realistic frame rates.