AM5 board compatible with three GPUs
AM5 board compatible with three GPUs
Hello, I'm researching the setup of a gaming PC for virtual machines and computation with three GPUs. The CPU should be a 79xx/X model based on requirements. The integrated GPU is needed for the Windows OS, which means using AM5 and Intel E/P cores won't work for VMs. Among the available AM5 boards, I couldn't find one that can accommodate three 2-slot GPUs without overheating or requiring water cooling. I initially considered the MSI PRO B650-P but discovered it only has two x1 PCIe x16 slots. The spacing between ports 2 and 3 is a concern. Are there overlooked boards that fit three air-cooled RTX 3080 or similar GPUs? A 4080 or 4090 might be ideal, though they seem too large. Alternatively, would using older Xeons be necessary due to cost? Let me know your thoughts.
You can use it with just PCIe 4.0 x4 and a larger case to accommodate a GPU in the lowest slot. The ASRock B650 LiveMixer fits this setup: https://www.asrock.com/mb/AMD/B650 LiveMixer/Specification.us.asp#Specification. It offers three slots—fourteen, sixteen, and four—connecting to the CPU, with the top for the CPU, middle for the CPU, and the bottom for the GPU.
I examined several options and found it difficult to determine the designer since most models leave a significant gap between slots 1 and 2 and then slot 3, limiting compatibility to just two GPUs unless you have a single-slot card. Right now I’m considering ASRock X670E PG LIGHTNING or ROG STRIX X670E-E GAMING. Both appear to have only two slots between slots 2 and 3, which means finding a suitable GPU that fits without overheating might be necessary. I checked one of them, but it’s currently unavailable in the central EU where I live (maybe I should have mentioned that earlier). Additionally, I didn’t mention that a built-in 2.5Gbit LAN would be beneficial.
Analysts review the present market trends. Regular users typically possess just one GPU, and PCIe ports are often bulky now. Placing them near the top would block access, so they position them lower to accommodate extra single slot cards. Utilizing M.2 drives in unused areas helps prevent wasted space. Disappointing. ... if you're open to higher-end options, ASUS likely produced some Turbo models around that time. I'm certain they released the 2080 Tis. They tend to overheat, but fans handle pressure better than a solid build. An axial design works if enough air is forced through intake fans; however, running my 780 Classifieds side by side caused the top card to overheat without performance loss. Alternatively, you can mount cards on waterblocks, which adds around $100 per card plus loop costs. If you choose older Xeons as suggested, you'll probably face similar challenges since two-slot cards were common and most setups were designed for that configuration. My boards supporting three GPUs all feature two-slot spacing for the primary GPU slots.
It seems this setup is pretty specialized. The cards likely fit between 3060 to 3080, excluding the 4xxx models which are too demanding. The goal is a bigger enclosure with many intake fans on the sides and front, exhaust at the top, and positive pressure throughout. I just double-checked—ASRock X670E PG LIGHTNING only has its final PCIe x16 slot in x1 mode, while the ProArt X670E-CREATOR only offers it in x2 mode. From AM5 boards, the ROG STRIX X670E-E GAMING is the best fit with two x16 slots and one x4 slot, though it’s quite pricey. The main drawback is that it costs as much as the CPU you’d install.
"Too hot" refers to situations where components generate excessive heat. In many setups the 4090 performs about twice as well as a 3090 but uses a bit more power. When using several lower-end cards together, total heat output might exceed what a single high-speed card can handle. For tasks involving virtual machines, ensure your motherboard supports strong IOMMU configurations to route traffic only to the intended device. Reflecting on the original question... you're wondering why multiple machines are needed and why liquid cooling isn't feasible—though it's costly, if GPU stacks are viable, a basic loop setup could work too. Sharing more context would help offer more targeted advice.
They’re mainly intended for gaming Windows virtual machines or computation setups, not for a single dedicated machine. They’re spread out and rarely monitored closely for issues like leaks. Adding water cooling would increase both expense and risk if a leak occurs. Unfortunately, some aspects are fixed by the operating system itself. Based on the specs—like an RTX 3080 in x4 mode—it seems insufficient, so it might not function properly. I’m unsure whether switching to Thread-Ripper, EPYC, or Xeon 2 cards would help much, especially if you plan to run six VMs instead of two.
This scenario is quite unusual. Designing a game VM machine differs significantly from a standard compute box, usually you choose one or the other. If you require a GPU per VM, you'd need 2-3 machines regardless. To accommodate six GPUs in a single system, they must be in a single slot—either as slow cards that act like displays or as a water-based setup.