Does anyone have any experience using an egpu?

[andyboy]

I have a Dell Precision 5550 laptop with a Quadro T2000 display adapter. It doesn't seem to have enough power to do interactive renders in a really usable way. It's fine for products, but not good for architectural. I'm considering getting an egpu and sticking a really good RTX adapter in there. Does anyone have any experience using Vray with this kind of setup? I'm using Vray for Rhino. The alternative is to buy a desktop with a good RTX card, but I would really only use it for this one task.

[glorybound]

Following... I have a laptop with an RTX 3080 and it is ok, but I am limited to smaller projects. I would love a quadro rtx 6000 in an external box.

[OzBoz]

Following... I have a laptop with an RTX 3080 and it is ok, but I am limited to smaller projects. I would love a quadro rtx 6000 in an external box.

Why would you spent that amount of money on a RTX6000 while a RTX 4090 is faster and cheaper? Well if you need 48GB it might be a solution but than you can save a lot by buying a simple desktop and add 2x 3090 inside with NVLink (which Nvidia doesn't support on 40xx series)

[Nicinus]

Why would you spent that amount of money on a RTX6000 while a RTX 4090 is faster and cheaper? Well if you need 48GB it might be a solution but than you can save a lot by buying a simple desktop and add 2x 3090 inside with NVLink (which Nvidia doesn't support on 40xx series)

Well for one it won't be a laptop, and I'm guessing Bobby also considers Vantage which from what I understand doesn't support NVLink.

[glorybound]

Well for one it won't be a laptop, and I'm guessing Bobby also considers Vantage which from what I understand doesn't support NVLink.

You are correct!

[OzBoz]

Well for one it won't be a laptop, and I'm guessing Bobby also considers Vantage which from what I understand doesn't support NVLink.

Don't really get your answer, a eGPU is not exactly a laptop as well is it? Laptops don't really have the same performance as their desktop counterparts with the same name. Still, a 4090 as eGPU would be faster than a RTX6000 as eGPU. If performance is what you're looking for a laptop always is a bad decision but can be needed to be mobile. Carrying a laptop + eGPU around is not exactly being mobile. So it could be smarter to just have a small PC build with 4090 (AIO liquid cooled if you want a compact case) which is still cheaper and makes your render machine independent from your laptop which is another plus if you want more speed.

The NVLink is a strange shout out, I mentioned it because of the possibility to get 48GB of VRAM. That Vantage doesn't support NVLink (*yet) doesn't matter because when you use an eGPU there is no NVLink anyway. There mighht be some chassis-like frames for multiple cards but good luck carrying them around with your mobile solution.

[glorybound]

Another machine would be more nodes, no? The RTX6000 has 48Gb of RAM, so that is the main reason. You don't take the eGPU on the go; you use it when you are at your desk. The mobility allows you to model and texture map, rendering out test still, and then you plug into your eGpu at the office. The last thing I want is to be a part time IT guy, which I was when I had five render machines.

[OzBoz]

Another machine would be more nodes, no? The RTX6000 has 48Gb of RAM, so that is the main reason. You don't take the eGPU on the go; you use it when you are at your desk. The mobility allows you to model and texture map, rendering out test still, and then you plug into your eGpu at the office. The last thing I want is to be a part time IT guy, which I was when I had five render machines.

I'm not sure how it currently works with the licenses long time ago you could just run the DR client on extra machines in your network and you didn't need a full V-Ray installed on other machines.

Most RTX6000's have 48 GB, not all, some 24GB. But the performance of a RT6000 is like half the performance of the 4090. A mobile 3080 has 8GB I think so depending on what you render I think 24GB is already 3x the amount of VRAM. Depends of course on the things you mostly render if you need more than 24GB.

NVidia tries to push you towards the Axxxx or RTX6000 series by removing NVLink from the 40xx series. Good for them but as long as you don't do deep learning AI stuff you pay twice for half the performance. For me it would make more sense to put a 4090 in an eGPU casing but since you have to disable the 3080 for rendering (else you have a 8GB limit) and the price of a eGPU casing is oddly high I would drop it in a small desktop. It will take a bit longer for the render to start depending on your network speed compared to a eGPU I think but it also makes you more flexible as you can also use that desktop maybe for some extra backup or handling batch tasks with images etc since it's there.

[Nejc Kilar]

...
Most RTX6000's have 48 GB, not all, some 24GB. But the performance of a RT6000 is like half the performance of the 4090. A mobile 3080 has 8GB I think so depending on what you render I think 24GB is already 3x the amount of VRAM. Depends of course on the things you mostly render if you need more than 24GB.
...

Just so folks aren't confused, I reckon you are referring to the Ampere A6000 graphics card? That one is basically ~ a 3090 with more VRAM in terms of performance. A little bit slower but still same "tier". Then there is the A6000 Ada card that is actually the current gen version of the A6000 and is basically a ~4090 for when it comes to performance, with more VRAM of course. Again, a little bit slower but same "tier" regardless. The naming scheme for these is... Confusing, yes

Laptop GPUs are typically not to be compared to their desktop counterparts because they are most often than not in a completely different performance tier with laptop parts being slower. So that's definitely something to keep an eye out on.

[OzBoz]

The naming scheme for these is... Confusing, yes

But instead of one RTX A6000 you can buy 3x RTX4090. That won't help you VRAM wise but the RTX Axxx series are basically cards with super expensive VRAM created by angles with golden dust. Their extraordinary pricing is basically only worth it if you want to a lot with AI-models.

I'm not sure how good the out-of-memory currently performs in V-Ray but if you really need 48GB of VRAM I would just look for two RTX3090Ti's and use NVLink. Saves you like $6000 and it's faster than a single RTX A6000. Only thing is that Nvidia will hate you for it.

[Nejc Kilar]

But instead of one RTX A6000 you can buy 3x RTX4090. That won't help you VRAM wise but the RTX Axxx series are basically cards with super expensive VRAM created by angles with golden dust. Their extraordinary pricing is basically only worth it if you want to a lot with AI-models.

I'm not sure how good the out-of-memory currently performs in V-Ray but if you really need 48GB of VRAM I would just look for two RTX3090Ti's and use NVLink. Saves you like $6000 and it's faster than a single RTX A6000. Only thing is that Nvidia will hate you for it.

Oh yeah I do agree and I use 4090s myself but things can get tricky if you are trying to scale up. NVLinking makes it problematic to fit multiple GPUs in one case and if we are talking scaling then power concerns become an issue too. I've heard people complain about the heat as well, 2x 3090s net you about 600W-700W at full rendering load whereas a single 4090 consumes about half of that

[andyboy]

None of these responses really answer my question. Let me ask it another way.

I have a Dell Precision 5550 with an Nvidia T2000 graphics card. My Vray render performance sucks.

Should I:

Set up an EGPU with an external box and some kind of super fast Nvidia RTX card. Will an EGPU alone increase my render performance?

Or,

Just buy a new PC with a super fast RTX card.

[glorybound]

If money isn't an object, and there are no other obsticles like space, then a new PC would be best

[OzBoz]

Oh yeah I do agree and I use 4090s myself but things can get tricky if you are trying to scale up. NVLinking makes it problematic to fit multiple GPUs in one case and if we are talking scaling then power concerns become an issue too. I've heard people complain about the heat as well, 2x 3090s net you about 600W-700W at full rendering load whereas a single 4090 consumes about half of that

I've quite a big desktop case and had a 3090. Now I got a 4090 which has an AIO cooler which performs like crazy even with a 90% power limit. But I don't want my 3090 collecting dust. Even though the 4090 doesn't take much space I'm still thinking about best way to also get the 3090 inside. Wrote the people at SeaSonic if my current 1000W will be enough but they advice me a stronger power supply. That can't be a SeaSonic since they are suddenly 30mm deeper which get me a conflict with some HDD's sitting there at the bottom. Like I said, quite a puzzle even though the Phantek Evolve X is quite a huge and wide case.

I don't have the budget but you can order some really nice builds with multiple cards and custom water cooling which is basically the best way. I wouldn't dare to build my own custom loop kind of thing. I've build many PC's but that's a step too far

Just buy a new PC with a super fast RTX card.

I'm with Bobby, but if you are on a budget, basically it can be an old PC with a new fast RTX card. But a second hand gaming desktop with a 3090(Ti) would also be a big upgrade.

[Muhammed_Hamed]

None of these responses really answer my question. Let me ask it another way.

I have a Dell Precision 5550 with an Nvidia T2000 graphics card. My Vray render performance sucks.

Should I:

Set up an EGPU with an external box and some kind of super fast Nvidia RTX card. Will an EGPU alone increase my render performance?

Or,

Just buy a new PC with a super fast RTX card.

Hello,

We are testing this now, it is quite easy to setup an external box(plug and play). The only issue is bandwidth with high-end GPUs, you will be getting 80 to 90% of the performance of a 3090 because of the bandwidth of Thunderbolt 3 which is not bad at all(talking about V-Ray GPU performance at least).. and there are no boxes that support Thunderbolt 4 at the moment

I haven't tested with 4080s or 4090s yet as they don't fit physically in an external box unless you mount the GPU outside and use a riser cable

I will make a post about using an eGPU in the announcements section soon, with some performance numbers

Best,
Muhammed