2015 15" Dell XPS 15 9550 (GTX960M) [6th,4C,H] + GTX 1080 @ 16Gbps-TB3 (Mantiz Venus) + Win10 [XCVG]
I've been putting off posting this for a while, but I'm not sure why because there's not much to say. This setup was anticlimactic in a good way.
The short version:
- Set TB security to SL0 and enable all TB features in the BIOS. I'd already done most of this.
- Install graphics card and plug everything in.
- Boot PC. Wait for Windows to install drivers.
- Install new drivers because Windows installed the old ones. Technically optional.
- Play game.
I've been using a Dell Dock for the past year, and I had quite a bit of trouble with that unit. I was expecting to have even more issues with an unsupported eGPU setup, but to my surprise it worked perfectly right out of the box. I was prepared to reinstall Windows (which I did anyway a few days later) but I didn't have to fiddle with anything at all. I did have my firmware and drivers up to date beforehand because of the aforementioned Dell Dock.
I'll follow up with background, benchmarks, more pictures, and a few notes on cooling shortly.
We'd love to see how you mount the radiator of this GTX 1080 Hybrid. It looks awesome naked! 😀
I wish I could answer that, but if you look closely, it's actually not a hybrid. It's a blower design with a near-identical shroud to the Hybrid and the internals of the Founders' Edition cooler. This card isn't easy to find but it was actually the cheapest option and it works well.
First of all, apologies for the double-post. I had some problems trying to format a reply into the same post body.
As promised, benchmarks! I haven't had the time to do all the formal runs I wanted to do, but I've run a few synthetics and I have quite a bit of qualitative data on the games and apps I use.
Most of the formal benchmarks I ran were synthetics. System configuration is the same for all of them- Dell XPS 15 9550, Mantiz Venus, Thunderbolt 3 at 16gbps, i5-6300HQ and GTX 1080. This laptop has been modified with VRM thermal pads and repasted with Arctic MX-4. I'm running Windows 10 1703, driver version 382.53, and the 960M has not been disabled for this test.
|Fire Strike Ultra||4519||4757|
|Fire Strike Extreme||7619||8975|
All 3DMark tests were run on their default settings, demo disabled. Unigine Heaven was run with the "Extreme" preset and Superposition with the "1080p Extreme" preset.
The only formally benchmarked game is Battlefield 1. I ran tests at 1080p and 2160p (4K). The test was the singleplayer mission "Over The Top", from the beginning to the first capture point. Both tests were run at Ultra, fulscreen, Vsync off, FPS cap 200, memory restriction off.
Unfortunately, I wasn't able to run the battery of comparative tests I wanted to run. One of my friends just built a new PC and we were going to try the 1080 in it, but he got a new job and I got bogged down in projects. I was able to run a few with the 960M and one with the 1080 in my server.
3DMark Time Spy - i5-2500K, GTX 1080, PCIe 2.0 x16 (x8 equivalent) - 5617 overall/7218 graphics
3DMark Time Spy - i5-6300HQ, GTX 960M, PCIe 3.0 x16 - 1330 overall/1219 graphics
3DMark Fire Strike - i5-6300HQ, GTX 960M, PCIe 3.0 x16 - 3900 overall/4345 graphics
I've been using this setup for over a month now in various use cases. It's my primary PC and I use it for everything- in fact I'm writing this up on it right now. I have been playing a few games, just not benchmarking, and I guess that speaks for itself that it works well enough that I don't feel the need to examine and tweak it in depth. I'm able to play the latest games at high quality, and I've been spending more time than I'd like to admit doing just that.
Windows Desktop: I noticed right away that Windows is noticably smoother with the eGPU. The Intel HD 530 iGPU really struggles to composite and render at 4K.
Adobe Photoshop: Photoshop hasn't been running great, but I'm not sure if this is because of the eGPU or because I updated to the 2017 version after reinstalling Windows.
Adobe Premiere Pro: It seems to work okay for the little I use it, but doesn't recognize the 1080 as a CUDA device unless the 960M is disabled. Which is irritating, but it works fine in OpenCL mode and disabling the 960M is trivial anyway.
The Elder Scrolls V Skyrim: It's Skyrim. I could do 1080p60 Ultra on the 960M easily. 4K60 Ultra is no sweat at all for the external 1080.
Battlefield 1: Despite the GTX 1080 theoretically being a fully 4K-capable card in this game, I haven't been able to get more than about 40FPS on either High or Ultra. From what I've read, Battlefield 1 is CPU-bound (several commenters recommend i7s!), and the i5-6300HQ is faster than a ULV but still quite a bit slower than a top desktop part. Performance seems to be worse in multiplayer than singleplayer, and reducing resolution scale to 75% (1440p-ish) gives a subjectively better experience that looks almost as good and runs a lot smoother.
Mass Effect Andromeda: According to most sources, this game runs significantly worse than Battlefield 1 (both use the EA DICE Frostbite engine). I can get about 40FPS at 4K after a bit of tweaking, putting the settings between High and Ultra. To me this suggest again that BF1 is CPU-bound. Unlike BF1, Andromeda has no ability to limit framerate, and doing it with Nvidia Inspector resulted in massive framerate drops. With framerate uncapped, they still happen occasionally but are nowhere near as bad, and it's a generally enjoyable (and quite beautiful!) experience.
The Witcher 3 Wild Hunt: I actually didn't benchmark this game until today, but it runs very well at 4K30 Ultra (HairWorks off). There are occasional frame dips and a bit of pop-in, but you hardly notice it and it's a very good experience overall. I uncapped the framerate and ran from the White Orchard village to the Nilfgaardian Garrison, getting an average of 34FPS. 4K60 Ultra isn't really doable on a GTX 1080, period, so I normally play this game capped at 30FPS.
Fallout 4: Take everything I'm about to say with a huge grain of salt, because I run Fallout 4 with a ton of mods, some of which may dramatically affect performance. 4K60 Ultra is out of the question. 4K30 Ultra almost I backed down on the view distance, and now I can get a mostly-steady 30FPS that sometimes drops through the floor. Creation loves CPU and memory bandwidth, neither of which my laptop setup has a lot of, and it's known to be glitchy, especially with a ton of mods piled on top. Still, I love this game. Hopping around the wasteland with a jetpack is a ton of fun.
The only issue I've experienced is actually not a video issue at all. Occasionally, the USB ports will all drop out, generally immediately after I plug in a device or while transferring a lot of data. This has happened with powered and unpowered hard drives plugged into the Venus, and once when I plugged in my mouse to my hub to charge. I'm starting to suspect that my Anker USB hub might be faulty or my hard drive might be puling too much power. I've had similar issues with my TB15 that were eventually (mostly) resolved with a firmware update and better cooling.
I promised I'd talk about cooling mods, so let's talk cooling mods.
The Mantiz Venus comes with a single slim, high-speed fan, which is rather loud. I'm not the only one to swap it out, because even on the other fan output as suggested it's still loud. The Venus also seems to be set up for an open-air cooler, which surprises me as I figured it would be much better for a blower. With an open-air cooler, air would be drawn in by the cooler's fans, then the hot air coming off it would be sucked out of the case by the single case fan.
It took me a few weeks of tinkering, but this is what I eventually came up with. I replaced the fan with a Noctua NF-R8 Redux mounted in the opposite direction. It doesn't blow quite as much air but it's much quieter. I added small heatsinks to the Thunderbolt controller and TI83, as well as a small slim fan to draw more air over that critical area. You'll notice it's pointing out, to exhaust out the side. You'll notice overall airflow is reversed, with the case fan blowing in and the PSU and side fan (as well as positive pressure) exhausting. I also added a duct for the blower itself so it wouldn't starve everything else of air, and hopefully it reduced GPU temps a bit as well. To replace the original plastic sheet, I used an overhead transparency, and it's all held on with double-sided tape.
It was honestly a surprise to me how the Venus was set up, and the noise was an unpleasant surprise as well. With this setup, it's a lot quieter- still not quiet, especially under load, but quieter. Hopefully it also provides better cooling with this setup as well.
A minor update:
Performance in Battlefield 1 has been somewhat disappointing. No matter what I can't get a steady 60FPS. After doing some research, I'm pretty sure it's an issue of that particular game being incredibly CPU-hungry- the game pegs my i5-6300HQ at 100%. I've read reports that people are running into similar issues even with much more powerful CPUs. With that being said, I'd still recommend getting the fastest processor you can for eGPU.
On the bright side, I was able to enable Secure Boot and BitLocker drive encryption and have that work properly whether it's docked or not. I've heard that people have had issues with BitLocker and Thunderbolt docks, and one thread suggested a workaround. There's a Group Policy settings called "Configure TPM platform validation profile for native UEFI firmware configurations". Disabling PCR 2 prevents it from checking the option ROMs, which would cause the PCR check to fail if a laptop wad docked or undocked. A picture is worth a thousand words here so I've included one below.
I've also moved to a much nicer desk setup. The previous white table was temporary and this is more permanent. I thought about getting a longer Thunderbolt cable and moving the box under the desk, but ended up leaving it behind the laptop because those cables are not cheap. I'll try to get a decent picture but it's still not very photogenic.
Hi XCVG, can you let me know what how badly the performance you mentioned is impacted by using the internal screen at 1080p
A few minor updates for the curious:
- The cooling mods seem to work fairly well, though the 1080 spits out a lot of very hot air.
- I still haven't found a good use for the hard drive I installed.
- Playing at 4K isn't really worth is over 1440p. In most games I ended up dropping to 1440p and enjoying higher quality settings and framerates. It's no surprise that this is a subjectively better experience; the GTX 1080 is marginal for 4K.
- I can gain an extra 10FPS in Battlefield 1 or so by disabling the 960M. It only seems to have an effect in Battlefield 1. Still, I usually leave the 960M disabled now. I rarely use it and it's very easy to enable and disable on the fly.
- I still haven't been able to fully solve my USB instability after a firmware update, BIOS update, and a few driver updates. I've just installed a new Thunderbolt driver reported to fix issues with the Aorus Gaming Box so we'll see how that goes.
- It works great with an Oculus Rift! The USB isn't really stable or fast enough for the Rift, though, so I ended up connecting everything to a USB hub which is connected directly to the laptop. I have no issues at all with the Rift, one tracker on USB 3.0, and one tracker on USB 2.0. Unfortunately my room scale is more like cupboard-under-the-stairs scale, but it's still a lot of fun and I'm looking forward to developing on VR.
- I'll probably get a new laptop next year, maybe the XPS 15 9570 if the rumours hold true. Hexcore Coffee Lake, GTX 1060 and 4-lane Thunderbolt, sign me up! All my friends are telling me to build a desktop, but I'm planning on sticking with eGPU.
@smgorman79, I haven't had a chance to try the internal display. If I have time I'll see how it works.
Hi there, I'm just curious as to how well the Mantiz Venus can charge the Dell XPS 15, because I read on some sources that you need the 130w adapter to use it to the fullest, but this one guy on reddit said he was able to use it heavily off of only 85w (on the Caldigit ts3), so could you maybe give some data regarding power draw, or possibly run the benchmarks with the 130w charger?