eGPU on Windows Laptop. First tests

Rhino Windows Hardware
,
1

hi all,

I’m setting up new hardware for the team, and while shopping for new GPUs for them I noticed how much of rendering performance bump we are going to get with the latest gen of RTX cards, considering that they are doubling the numbers of Cuda cores compared to the previous gen. Almost everyone in the team is on RTX 2080TIs (4352 Cuda cores), and we are upgrading them to either 3090 (10496 Cuda cores) or 3080ti (similarly spec’d as the 3090s but a lot cheaper, rumored to launch next month).

I personally don’t work on a desktop anymore, and my laptop has a Quadro RTX5000 that works very well for realtime work, but it’s a bit slow in GPU rendering (Octane), VR (Unreal) and using PBR shaders in Rhino. So I did some research to see if I could have an eGPU for when I’m working at the office (or home now during lockdown). I could not find anything at all, besides gaming benchmarks. Also most eGPU info is on Mac stuff and using really shitty video cards anyway.

So I decided to buy one and try myself…

I got the Razer Chroma X (Razer Core X Chroma review: colorful, capable and expensive - The Verge), and keep in mind this is just a box, with cables, fans and a power supply. And LEDs, lots of LEDs. So it’s not cheap for an empty box.

I have not bought the new GPUs yet, but I do have an RTX 2080ti in a customer’s demo machine, so I borrowed it and tried it with that for now to see if this has any potential.

…It works!

The process is simple: install the GPU in the box (PCI slot and power connectors), plug the box to your PC using the USB-C cable. Keep in mind this does not work with any laptop with USB-C ports, your laptop has to have Thunderbolt 3. I did not install any software, the driver for the 2080ti auto-loaded once this boxed was plugged it. I should explore and see if I should install better/newer drivers. I guess Rhino’s System info should be able to tell me what’s loaded on these correct? (EDIT: I see that Holomark tells me that too, but @Holo, see how both of my cards ar listed as 4GB memory, when the Quadro has 16, and the 2080ti has 11GB, do you know why? )

Some small problems: The internal Quadro RTX 5000 stops working as soon as the eGPU is connected, so I do not have dual GPus. At least I have not figured out how yet. After unplugging the eGPU I had to uninstall the internal GPU from my Device Manager and restart the machine to get it to work again.

The Results:

Viewport PBR Shaders, with this model:

capture_gf_012352AM_210106
capture_gf_012352AM_2101061333×750 230 KB

Laptop GPU (Quadro RTX 5000 mobile):

capture_gf_123502AM_210106
capture_gf_123502AM_2101061477×80 3.13 KB

eGPU (RTX2080ti):

capture_gf_113708PM_210105
capture_gf_113708PM_2101051711×70 3.39 KB

It’s a lot smoother, on a stress-test model that I know it’s very demanding for Rhino.

HoloMark V2 R6:
Laptop GPU (Quadro RTX 5000 mobile):

capture_gf_125426AM_210106
capture_gf_125426AM_2101061309×763 113 KB

eGPU (RTX2080ti):

capture_gf_121155AM_210106
capture_gf_121155AM_2101061309×763 116 KB

I’ll do some Octane tests tomorrow too and see what the difference is, but in general this is pretty promising, especially for those considering getting a thinner laptop with less power (and less cost/ heat/thickness/weight) than a Precision 7740.

From a financial standpoint this would be similar to getting a good but mid-range laptop GPU, rather than upgrading to a top of the line laptop video card, and with a lot more performance when you are not on the go.
Here are some upgrade costs for the Dell Precision’s GPUs:

capture_gf_015924AM_210106
capture_gf_015924AM_2101061396×473 32 KB

We also have here one of those thin and cute Precision 15" with 4GB Quadro cards (they are basically like a Dell XPS), I’ll check if that has an Thunderbolt 3 port to test it too.

I’ll report back more finding…

Gustavo

4 Likes
2

Hi Gustavo, great stuff! I tried an egpu years ago, but that machine didn’t have thunderbolt3 and would only work on older nvidias and amd cards, so good to hear that it works smoothly now!
The vram detection 4GB limit is caused by a 32bit registry thingy, and @nathanletwory has added some stuff to the latest v7sr that will tell me the full ram. I’ll add it to Holomark 3. Talking about HM3, care to do some alpha-testing on your egpu?

3

Hi Jørgen, sure let’s test it. I want to compared also with one of the desktops with the same exact GPU connected directly in the motherboard and see if the Thunderbolt is affecting performance.

Thx,

G

1 Like
4

Some small problems: The internal Quadro RTX 5000 stops working as soon as the eGPU is connected, so I do not have dual GPus. At least I have not figured out how yet. After unplugging the eGPU I had to uninstall the internal GPU from my Device Manager and restart the machine to get it to work again.

So after fixing this hiccup, does your laptop display run off of the Quadro 5000, and the eGPU runs your eternal display?

I have an integrated Intel 630 GPU on my Razer Blade 15", together with and internal RTX 2080, and I’m wondering where the 630 going to sit in the mix and more importantly if I’m going to be able to work around it.

5

I had no problem doing whatever I wanted with the eGPU and internal GPU on my Razer…something…13 when I went nuts buying 1080ti’s in 2018 to render an animation and was scrounging for places to put them.

1 Like
6

I researched this more. In my case, I can never have the internal GPU (Quadro RTX 5000) and external GPU connected to the eGPU box ((RTX 2080) being detected/usable at the same time. When the eGPU is connected, the internal GPU is just gone.

The reason: for both GPUs to work together they have to use compatible drivers. In other words: be both GeForce, or (I assume) be both Quadro. You cannot mix them up.

I think that maybe the reason Jim can run both internal and eGPU is that in the case of the Razer Something 13 the internal GPU is also a GeForce card (same drivers) as his eGPU (1080ti).

G

1 Like
7

yeah the laptop has a geforce but it also has Intel integrated…funny it certainly seems like you shouldn’t be having the issues you are, “mixing and matching” totally different gpus is a thing.

1 Like
8

Thanks good info here, I’ll make sure to use another GeForce GPU.

9

yeah I researched this a bit and the conflict of eGPU with internal was the most common cause. I’m waiting for a new external GPU and once that’s more final I will try some more things. Currently I also have Optimus disabled in my BOIS< so that might be part of the problem. I could always try to Softquadro the external GPU (is that still a thing), but I’m afraid that’s already pushing my luck too much.

G

10

Thanks for the help here, I ordered an eGPU enclosure and I guess I’ll just cross my fingers and hope that RTX 3080s are available at some point.

1 Like
11

hi @Holo, I just got an RTX 3090 in a dark alley of a sketchy city near me. I want to try Holomark on it. Is the Holomark 3 anywhere near ready to test on the eGPU?

For now I’ll test with Holomark2_R6_2

thx,

G

12

…here are the results on the mini-precision 5540 (XPS style thin laptop, with a Lil’ Quadrito T2000) + RTX 3090 on eGPU:

capture_gf_081637PM_210607
capture_gf_081637PM_2106071222×818 131 KB

so now you know, no need to have a badass giant laptop, when you can do a lot better with a thin/affordable laptop (this was $2,258.88 shipped, with taxes and 3-years support on Dell Outlet) + an eGPU at your desk. In this case, I was also running 2 4K monitors side-by-side

I think this is going to be our next Pro-setup for people that need some mobility.

G

3 Likes
13

BTW, here’s the same mini-laptop WITHOUT the external eGPU.

capture_gf_093408PM_210607
capture_gf_093408PM_2106071258×794 130 KB

Kind of unacceptable IMO.

14

The question of course is whether the laptop in its non-eGPU config is running off the Intel or the Nvidia GPU…

Either way the T2000 is much less powerful than the RTX 3090…: base core clock lower than RTX 3090, 1/6th L2 cache of RTX 3090, 1/10th of compute cores, fill rates 1/3rd-1/4th of RTX 3090, etc, etc.

15

I’ve noticed that the eGPU was running constrained, in fact, the Varjo XR-3 would not even run from it without crashing. So I updated the PUS to give more power to the RTX 3090:

ePSU_on_eGPU
ePSU_on_eGPU1154×866 169 KB

This fixed the VR headset crashes, and also gives a bit better Holomark score:

capture_gf_111549PM_210614
capture_gf_111549PM_2106141255×796 128 KB

G

16

Hi Gustavo,

Sorry, late to the party here, but FWIW I have a desktop with a Geforce RTX2080 running alongside a Quadro P4000. They both use the Geforce Studio Driver . Delving into the Nvidia documentation, this officially supports Quadro.

Only problem is that Windows Update occasionally sees the P4000, installs the Quadro driver instead and kills the 2080 and I have to reinstall the correct driver. So mixed hardware pairing works with Geforce driver, but not with Quadro.

Regards
Jeremy

17

I think you can disable Windows Update’s downloading of 3rd party drivers like video cards… don’t remember where the setting is though.

18

Yes, you can set a device-specific group policy - only downside is that it doesn’t just stop automatic updates, it stops manual ones too, so when you do want to update a driver you have to disable the policy.

It really just needs Microsoft to tweak Update to recognise the mixed card use case and load the correct driver, but establishing a way to communicate that to Microsoft effectively looks to be even more time-consuming than reinstalling…

19

Hmm, interesting. I didn’t know this was possible - It would have been useful when I had just the A6000 and RTX 2070. Good thing to keep in mind for the future.

20

Dunno, I’ll have to check, but my computers are set to not download anything other than Microsoft updates, and I have no problem updating my video drivers. I’m not on any kind of company network though.