H170, H270 for GPU rendering

[Miklos_1]

Hey! I am planning to build a budget GPU render machine for starting out in fstorm. I'll be straight.

My question is can i use a H170 or H270 MOBO to my build with dual GPU? I've read it can only use PCI-e in 1x16, and doesnt support SLI, but as i know for rendering I dont even need SLI.

 

[Herc08]

Well for rendering, motherboard doesn't matter. It depends on CPU, RAM and sometimes GPU. I hope you are getting at least an i5, with 16GB, or else your rendering is going to be frustrating.

 

[Miklos_1]

Well for rendering, motherboard doesn't matter. It depends on CPU, RAM and sometimes GPU. I hope you are getting at least an i5, with 16GB, or else your rendering is going to be frustrating.

Yes I am , i go for the i7 7700 but i want to choose GPU based renderes, like FStorm, or Octane, or Vray RT. So i am wondeering if my MOBO is able to handle 2 Graphicscards

 

[Eximo]

I don't believe you need SLI or Crossfire to install multiple GPUs for rendering. That is more of a gaming thing.

Still, if the plan is to always be doing rendering, bump yourself up to the X99 platform and get something like a 6800k and quad channel memory.

If you are doing this professionally, High performance Xeon and ECC memory.

 

[Herc08]

Well for rendering, motherboard doesn't matter. It depends on CPU, RAM and sometimes GPU. I hope you are getting at least an i5, with 16GB, or else your rendering is going to be frustrating.

Yes I am , i go for the i7 7700 but i want to choose GPU based renderes, like FStorm, or Octane, or Vray RT. So i am wondeering if my MOBO is able to handle 2 Graphicscards

It seems your prioritieis are out of whack. You want high end features, but want to go cheap on the moterboard. If that is the case, go with the H270, but why not step up to the Z270, and get a K processor?

I understand you are on a budget, but what is your [strike]email[/strike] budget?

 

[Tomas73]

I'm not a hardware expert, but I do use GPU rendering, and have built my own multi-GPU system. Here is my conjecture:

Firstly, your are correct, SLI is NOT used for GPU rendering, it will confuse things; so don't use SLI at all. Secondly, I'm not familiar with that board but...you should research the specs of your components, and understand what difference they make, if your doing a self-build. So maybe do a bit of reading on GPU rendering systems, and look again at the motherboards, to see if they will meet your requirements.

You CAN use almost any basic PC to start off with, but if you're looking to upgrade it over time this can require some considerable forethought to save you having to replace components or rebuild entirely.

You need to decide on an overall approach, then you will know what is important to your build. You may be using an external GPU enclosure, or splitters/risers to accommodate the GPUs so this would affect your mobo choice too. The mobo is the heart of the machine.

If you can, it is easier, cheaper and generally more reliable, to fit all your GPU's on the mobo directly. This necessitates an appropriate PCI-e slot layout; multiple dual-width cards take up space. As long the cards physically fit, the next consideration is performance. The PCI-e lanes are more important than the PCI-e speeds, although x16 is ideal for responsive realtime feedback.

Because GPUs share the total number of PCI-e lanes with other components, to relay info back and forth, a limited number of PCI-e lanes can cripple your general performance, even if final rendertimes are okay. (This is why having more PCI-e lanes provides a more stable system, and allows you to operate more GPUs without crippling the system). Once data is in the GPU, it is processed at the GPU's speed, irrespective of the rest of the system, then piped back via PCI-e. (x16 can speed up the amount of time it takes to refresh the scene when you make changes, but won't change the render time.) Don't forget 'x16 3.0' is twice as fast as 'x16 2.0'.

PCI-e lanes are dependant on the CPU and the Mobo. You want a CPU with at least forty lanes. Also if the mobo has an onboard PLX chip this will add a further eight or so, (that can help but is not essential). For example: in order to run eight GPUs reliably, I have dual xeons on a high-end mobo providing a total over 80 lanes. (Btw, I don't use an M.2 NVMe for the OS, although I could, because these use up more PCI lanes, so using a sata SSD saves lanes.)

The BIOS can be a real pain, so an up-to-date mobo and bios configuration is more likely to play nice with recent GPUs.

If you are interested, I have a group called 'GPU Rendering Hardware Solutions' on Facebook, where people with various experience share info:
https://www.facebook.com/groups/1797500417129685/

' Hope this reply helps.

 

[Miklos_1]

I'm not a hardware expert, but I do use GPU rendering, and have built my own multi-GPU system. Here is my conjecture:

Firstly, your are correct, SLI is NOT used for GPU rendering, it will confuse things; so don't use SLI at all. Secondly, I'm not familiar with that board but...you should research the specs of your components, and understand what difference they make, if your doing a self-build. So maybe do a bit of reading on GPU rendering systems, and look again at the motherboards, to see if they will meet your requirements.

You CAN use almost any basic PC to start off with, but if you're looking to upgrade it over time this can require some considerable forethought to save you having to replace components or rebuild entirely.

You need to decide on an overall approach, then you will know what is important to your build. You may be using an external GPU enclosure, or splitters/risers to accommodate the GPUs so this would affect your mobo choice too. The mobo is the heart of the machine.

If you can, it is easier, cheaper and generally more reliable, to fit all your GPU's on the mobo directly. This necessitates an appropriate PCI-e slot layout; multiple dual-width cards take up space. As long the cards physically fit, the next consideration is performance. The PCI-e lanes are more important than the PCI-e speeds, although x16 is ideal for responsive realtime feedback.

Because GPUs share the total number of PCI-e lanes with other components, to relay info back and forth, a limited number of PCI-e lanes can cripple your general performance, even if final rendertimes are okay. (This is why having more PCI-e lanes provides a more stable system, and allows you to operate more GPUs without crippling the system). Once data is in the GPU, it is processed at the GPU's speed, irrespective of the rest of the system, then piped back via PCI-e. (x16 can speed up the amount of time it takes to refresh the scene when you make changes, but won't change the render time.) Don't forget 'x16 3.0' is twice as fast as 'x16 2.0'.

PCI-e lanes are dependant on the CPU and the Mobo. You want a CPU with at least forty lanes. Also if the mobo has an onboard PLX chip this will add a further eight or so, (that can help but is not essential). For example: in order to run eight GPUs reliably, I have dual xeons on a high-end mobo providing a total over 80 lanes. (Btw, I don't use an M.2 NVMe for the OS, although I could, because these use up more PCI lanes, so using a sata SSD saves lanes.)

The BIOS can be a real pain, so an up-to-date mobo and bios configuration is more likely to play nice with recent GPUs.

If you are interested, I have a group called 'GPU Rendering Hardware Solutions' on Facebook, where people with various experience share info:
https://www.facebook.com/groups/1797500417129685/

' Hope this reply helps.

Thanks! Sucha detailed answer, i will definetly join the FB group.