[SOLVED] Video Card in Chipset pcie slot

Jul 8, 2019
2
0
10
Hi all,
I'm shopping for a motherboard and I have one gpu (980ti) but I'm expecting to buy a new one in a year or so (maybe when we have a pcie gen 4 rtx or when the 980ti no longer cuts it) and I would then use both graphics cards ( not in a sli confuguration, just as a main gpu plus a second one that can help with calculations in some programs). Will I need both of my pcie slots to be attached to the cpu, or is it ok that the second one is linked to the chipset? I don't really understand the meaning of expansion slots being linked to cpu vs chipset.

I'm considering for instance this mobo https://www.asus.com/Motherboards/TUF-GAMING-X570-PLUS/specifications/
The second gpu would be on the chipset (and limited to x4, I don't know how much that would slow it down either)

Thanks!
 
Solution
Here is someone who tested x8 vs x16 ... you will see that the difference is minimal.

Also, if you are going to use the 2nd GPU as a hardware assist for certain software, I doubt as much data passed to the card as there would be for graphics. Remember, the bandwidth of the slot (lanes) does not affect the speed of the GPU ... only the rate at which you can move data to the GPU for processing. Yes, with a very fast GPU, you could technically starve the GPU (not get data there fast enough) but I think this is quite rare.
Here is a link to a block diagram that shows how the setup works. Your main PCIe x16 slot (and GPU) will be connected directly to the CPU. The secondary PCIe x16 slot is connected to the CPU via the X570 chip. That data will move fast, but not as fast between the CPU and secondary GPU as it would be limited by the x4 connection between CPU and X570 chip.

How fast? PCIe 4.0 x16 can theoretically move 64 GBytes/sec. The x4 connection is limited to 16 GBytes/sec (which is still a lot).
 
Thanks tor the answer. I think I was mainly worried that a gpu somehow shouldn't be connected to the chipset, that it might not work at all or be super slow for having to connect to the cpu via the chipset or something.

This being said, I assume that since the gpu that will eventually be connected to the chipset pcie isn't pcie gen4, then the connection would default to gen 3 and give me half, so 4 GB/sec.

Still, maybe that's actually better than having both gpus on cpu slots, reducing the number of lanes on my primary gpu to 8, which might be relevant if that eventual new gpu is just pcie gen 3.

Thoughts welcome.
Thanks,
E
 
Here is someone who tested x8 vs x16 ... you will see that the difference is minimal.

Also, if you are going to use the 2nd GPU as a hardware assist for certain software, I doubt as much data passed to the card as there would be for graphics. Remember, the bandwidth of the slot (lanes) does not affect the speed of the GPU ... only the rate at which you can move data to the GPU for processing. Yes, with a very fast GPU, you could technically starve the GPU (not get data there fast enough) but I think this is quite rare.
 
  • Like
Reactions: TheIronEddie
Solution