I’m looking to upgrade my motherboard to add a third GPU. The gigabyte GAX79UP4 lists 4 PCI 3.0 slots (2 16x + 2 8X). Does running a GPU on an 8X yield lower performance? I can’t seem to find any boards with 4 16X slots. Thanks in advance.
Look in the section; Improving Rendering Performance.
Specifically “What’s this about PCIe speeds?”
You won’t notice any difference rendering with Cycles between the two, the only info that will travel over the bus isn’t nearly enough to even touch the bandwidth possible with 8x
Great - thanks!
To be absolutely specific:
The theoretical bandwidth per lane of:
- PCIe 2.0: 500 MB/s
- PCIe 3.0: 985 MB/s
Multiply that with 8, respective 16 lanes you get:
PCIe 2.0, x8: 4 GB/s
PCIe 2.0, x16: 8 GB/s
PCIe 3.0, x8: 7,875 GB/s
PCIe 3.0, x16: 15,75 GB/s
Now let’s take the average card with 3 GB of VRAM. To transfer 3 GB of data and completely fill up the VRAM you’d need:
It will obviously take twice as long with half the bandwidth, but that’s really relative, the absolute time difference is marginal for a rendertime of ~10 mins per frame.
PCIe 2.0, x8: 750 ms
PCIe 2.0, x16: 375 ms
delta time: 375 ms
PCIe 3.0, x8: 380 ms
PCIe 3.0, x16: 190 ms
delta time: 190 ms
That said, the typical human reaction time is around 250ms. So like already said, doesn’t matter for Cycles. The data is transfered before you can click “cancel” or hit <esc>
Thanks for the added explanation - some of you guys on here probably forget more about hardware in one day, than I could learn in a lifetime!