Can I use 2 TB3 devices on TB3 Add-in-Cards?

[phongle123]

https://www.gigabyte.com/us/Motherboard/GC-TITAN-RIDGE-rev-20#kf

 

Taking a look at the Gigabyte Titan Ridge Thunderbolt 3 Add-in-Card. It has 1x DP, 2x TB3, and 2x MiniDP. What I want to do is use these TB3 for eGPU for rendering and not for gaming. So I do not incur as much of a performance penalty.

I am wondering since there are 2 TB3 ports on here. Would I be able to plug in 2x TB3 eGPU with each eGPU using x2 lanes each or can I only plug in 1x TB3 device?

 

[Pixel5]

you can probably plug in two TB devices but it doesnt make any sense at all especially not for e-GPU´s

Until these cards are available with PCI-E 4.0 to double their bandwidth this is completely useless because technically the 40GB/s is already false because a 4x PCI 3.0 BUS can only deliver 4Gb/s while while order to transfer 40 GigaBit you need 5Gb/s transfer speeds.

 

Why do you want to render via an external GPU at all?

[phongle123]

14 minutes ago, Pixel5 said:

you can probably plug in two TB devices but it doesnt make any sense at all especially not for e-GPU´s

Until these cards are available with PCI-E 4.0 to double their bandwidth this is completely useless because technically the 40GB/s is already false because a 4x PCI 3.0 BUS can only deliver 4Gb/s while while order to transfer 40 GigaBit you need 5Gb/s transfer speeds.

 

Why do you want to render via an external GPU at all?

I understand your reasoning. However, renders using GPUs will stack and from my testing GEN3 x16 internal vs x4 external is only a 6% difference --- for my rendering workload. Whereas for gaming, its 33% performance difference. I'd rather have the performance than not. 

 

Though, I was unable to test it at x2 because my laptop is a TB3 x4 and not an x2 port. Even if the performance of x2 is lower. Using 2 GPUs in x2 will give more performance than 1 GPU in x4.

 

Not any inclination of price to performance, obviously. Time is more important to me. 

[Pixel5]

but this is an internal TB card where would you want to mount it in your laptop?

overall if performance is what matters you are probably better off buying a good PC than using a laptop.

[phongle123]

6 minutes ago, Pixel5 said:

but this is an internal TB card where would you want to mount it in your laptop?

overall if performance is what matters you are probably better off buying a good PC than using a laptop.

I'm going to install it in my desktop. I had to use my laptop to test the eGPU because my desktop does not have a TB3 port. Which was the point of this post. However, when testing, it did not use CPU. So the comparison is valid.

 

It is not, heat is a problem too. I currently have 3 GPUs inside of my PC. Which lower down the clock speeds of the 3 GPUs spewing out heat. I'm using an x370, so my 3rd GPU is already running on PCIE 2.0 x4 (not 3.0). Which is the same as PCIE 3.0 x2. I already have the 3 GPUs and 1 eGPU. So, I would just need to buy 1x eGPU and this TB3 AIC.

 

My cards drop down to 1400Mhz while in the eGPU it stays at 1950Mhz. All in all, TB3 is my better solution. 

 

Like I said time is more important to me. Which is why I asked here first. I don't have to have to return it if it only supports 1x TB3 device. Since if it does work for 2 devices I will switch over to an ITX case to account for the 2 extra eGPUs taking up space.

 

My only question was whether or not the TB3 card will be able to work with 2x TB3 devices. When you answered yes... was that a guess?

[Ankh Tech]

7 minutes ago, phongle123 said:

I'm going to install it in my desktop. I had to use my laptop to test the eGPU because my desktop does not have a TB3 port. Which was the point of this post. However, when testing, it did not use CPU. So the comparison is valid.

 

It is not, heat is a problem too. I currently have 3 GPUs inside of my PC. Which lower down the clock speeds of the 3 GPUs spewing out heat. I'm using an x370, so my 3rd GPU is already running on PCIE 2.0 x4 (not 3.0). Which is the same as PCIE 3.0 x2. I already have the 3 GPUs and 1 eGPU. So, I would just need to buy 1x eGPU and this TB3 AIC.

 

My cards drop down to 1400Mhz while in the eGPU it stays at 1950Mhz. All in all, TB3 is my better solution. 

 

Like I said time is more important to me. Which is why I asked here first. I don't have to have to return it if it only supports 1x TB3 device. Since if it does work for 2 devices I will switch over to an ITX case to account for the 2 extra eGPUs taking up space.

 

My only question was whether or not the TB3 card will be able to work with 2x TB3 devices. When you answered yes... was that a guess?

Does your motherboard have an internal tb3 port, it needs one for the card to work

[phongle123]

5 minutes ago, Ankh tech tips said:

Does your motherboard have an internal tb3 port, it needs one for the card to work

Thanks for chiming in. It needs it, but, that won't be a problem. The internal TB3 header is an Intel verification that is not required. This link shows that it can be fixed by shorting pins 3 and 5.

https://egpu.io/forums/builds/thunderbolt-3-with-threadripper-on-a-asrock-x399-fatality-motherboard-help-needed/#post-59565

[HenrySalayne]

8 minutes ago, Pixel5 said:

you can probably plug in two TB devices but it doesnt make any sense at all especially not for e-GPU´s

Until these cards are available with PCI-E 4.0 to double their bandwidth this is completely useless because technically the 40GB/s is already false because a 4x PCI 3.0 BUS can only deliver 4Gb/s while while order to transfer 40 GigaBit you need 5Gb/s transfer speeds.

 

Why do you want to render via an external GPU at all?

Just to clear this up: upper-case B = Byte, lower-case b = bit. So 40 Gb/s are 5 GB/s.

And 40 Gbit/s is not false. It's the specified speed of the connection. Just like the 480 Mbit/s for USB 2.0 (in theory 60 MB/s, but only <40 MB/s usable), the usable speed is a bit (or sometimes a lot) lower.

 

@phongle123
Besides the bottleneck of the connection, eGPUs are pretty expensive and probably not best practice in this case. You may be better off with PCIe bifurcation. You can split 1 x16 to 4 x4. Or you could get a mainboard with enough PCIe slots or just switch to a platform better suited for these workloads like Threadripper. HEDT platforms have plenty of PCIe lanes.

[Ankh Tech]

2 minutes ago, phongle123 said:

Thanks for chiming in. However, it does not. But that won't be a problem. The internal TB3 header is an Intel verification that is not required. This link shows that it can be fixed by shorting pins 3 and 5.

https://egpu.io/forums/builds/thunderbolt-3-with-threadripper-on-a-asrock-x399-fatality-motherboard-help-needed/#post-59565

Really, thanks you solved all my issues man, and post this as a thread by itself, it's helpful, but where did he get the red wire?

[LogicalDrm]

-> Moved to CPUs, Motherboards and Memory

[Pixel5]

4 hours ago, HenrySalayne said:

It's the specified speed of the connection.

and thats the problem here and why its false, the PCI-E link can not support 40Gb/s on this card so you will never reach the advertised speed.

[AchakBrooks]

On 10/19/2020 at 3:57 AM, phongle123 said:

Thanks for chiming in. It needs it, but, that won't be a problem. The internal TB3 header is an Intel verification that is not required. This link shows that it can be fixed by shorting pins 3 and 5.

https://egpu.io/forums/builds/thunderbolt-3-with-threadripper-on-a-asrock-x399-fatality-motherboard-help-needed/#post-59565

I have a TRX40 motherboard and the Thunderbolt card you mentioned. I am curious about your experience using this setup. Were you able to achieve the results you were looking for? Can both Thunderbolt connections be used with eGPUs, and do they run at x4 or x2 lanes each?

 

I want to try the same setup with my build, but I am not very familiar with non-Apple hardware when using Thunderbolt. I have read about some issues people have had, but I am not sure how much weight to give these reports. I have seen some odd conclusions about "best practices" that seem to focus solely on gaming.

 

I have not experienced these issues with my Apple devices, and I have noticed that many people don't pay attention to how the Thunderbolt controller is being used by the PC laptop, which is sometimes misrepresented by the manufacturer.

 

Typically, I use an iMac Pro and populate two ports with Windows (using 3090s), and x4 with MacOS (using AMD GPUs) for rendering (using Redshift, Octanex, and Cycles). When using eGPUs for rendering, each port provides full x4 PCI lanes (on the iMac Pro), so there is no noticeable difference in render times compared to having the cards hooked up internally