Discussing PCIe bifurcation and optimizing a 3x GPU setup. Looking for expert tips on Mobo advice.
Discussing PCIe bifurcation and optimizing a 3x GPU setup. Looking for expert tips on Mobo advice.
Hello everyone, I’m gearing up for a PC build centered around machine learning, deep learning, local AI creation, and some AAA 1080p/1440p gaming. I’ve assembled most of the components but need guidance on picking a motherboard that provides enough PCIe lanes for my triple GPU configuration. Here’s an overview of my plan: My starter build is at https://pcpartpicker.com/list/bqjhHG. I’m planning to use an AMD Ryzen 9 7950X3D at 4.2 GHz with 16 cores, paired with an MSI VENTUS 2X BLACK OC GeForce RTX 4060 Ti and a 16 GB video card in a Corsair 6500D case. For power, I’m opting for a be quiet! Straight Power 12 1000 W 80+ Platinum certified fully modular ATX unit. RAM will be TEAMGROUP T-Create Expert 64 GB (two 32 GB modules). Cooling will rely on ARCTIC P14 and P12 fans, plus a liquid cooler. Storage is a Silicon Power US75 4 TB M.2-2280 PCIe 4.0 X4 NVMe drive. My ultimate goal is outlined here: https://pcpartpicker.com/list/37tsGP. Regarding the GPU setup, my main goal is an x8/x8/x8 PCIe split or a similar equivalent. The Ryzen 9 supports up to 24 lanes, which should cover it. But I need a board that can manage all three GPUs in their respective slots without issues. The 4060’s 128-bit bus and PCIe limitations are a concern, especially at that price. Running multiple GPUs might still restrict bandwidth per slot. I’m leaning toward an x8x8x8 configuration, trusting the chipset can handle it. Storage might be a bottleneck due to dedicated lanes, but that’s a trade-off I’m comfortable with. I’m not sure if any board today supports this exact layout, but I’ll keep checking the ASUS compatibility list and community insights. Thanks for any advice or experiences you can share!
The CPU performs at most 8/8 with the main graphics ports, using 4 lanes for the chipset, another 4 for M.2, and a fourth for another M.2 or additional slots. Achieving 8/8/8 would require skipping lanes for M.2, which would make the board less competitive compared to others. For the X670E, 8/8/4 seems to be the practical limit. Keep in mind that bifurcation tables typically describe splitting lanes within a single primary slot (like running multiple M.2 cards on an expansion card), not distributing lanes across different slots, as this is usually fixed in the BIOS and not adjustable.
I'm not familiar with the chipset for storage, but switching to SATA might help. The 24 usable lanes could be allocated only to the x16 slots, but I'm not sure about the compatibility details from ASUS. The support page mentions PCIE bifurcation for Hyper M.2 cards, but I need to verify if that applies here. This is your first build, so it's a bit confusing—please clarify if you need more details.
The cost per VRAM is the key selling factor, particularly for ML/DL tasks. Gaming takes a back seat and Nvidia comes out priced too high. Would support the red team, but it falls flat for AI.
The 4070 Ti Super offers roughly double the processing power of the 4060 Ti while costing less than twice as much. It retains 16GB of VRAM. Two 4070 TiS units are only about three times the cost of three 4060 Ti 16GB models. If budget is tight, opting for a faster GPU variant could still provide a viable solution.