I'm in Australia and have a total budget of AU$6,000 for a workstation that will run Linux 24/7. I already own a Cooler Master HAF X case and a Super Flower Leadex 7 Pro 1.2kW Platinum PSU, and I have roughly 4TB of NVMe and SATA storage. The case still needs suitable fans and dust filters.
The system will be used for containers, virtual machines, video/photo/audio production, CAD, 3D modeling, gaming, and local AI training, tuning, and inference. I'm aiming for at least 64GB of total VRAM, potentially using two to four GPUs. I also need a motherboard with built-in IPMI or BMC management, plenty of PCIe lanes and physical slots, ECC memory support, and a platform that can be seriously undervolted for lower temperatures and continuous operation.
I'm considering used server hardware, EPYC, or Threadripper because ordinary consumer platforms generally lack the PCIe lanes and remote-management features I need. A Ryzen 7 9800X3D costs around AU$900, while a RTX 5090 can cost roughly AU$5,000–8,000 locally, so US pricing and simple currency conversions are not useful here. What platform, motherboard, CPU, memory, and GPU combination would realistically meet these requirements within the budget?
5 Answers
This combination is closer to a used workstation or server than a normal gaming PC. Current consumer platforms generally won’t provide both IPMI and enough PCIe lanes for two to four GPUs. I’d focus on second-hand EPYC or older Threadripper Pro systems, and check Australian listings from specialist workstation retailers and used enterprise-hardware sellers. Expect the motherboard and ECC memory to consume a significant part of the budget.
The power and cooling requirements need more attention. A 1.2kW PSU may be enough for two efficient GPUs, but four cards plus a high-core-count CPU can leave little headroom, especially under sustained workloads. Also, don’t run a powerful 24/7 workstation directly on carpet without proper airflow and dust control. Use a solid platform, add intake filtration, undervolt the GPUs and CPU, and verify the Australian circuit’s continuous-load limits rather than relying only on the breaker rating.
A parts-list website aimed at ordinary desktops won’t cover many of these components. The sensible approach is to start with the required PCIe layout, IPMI/BMC, and ECC type, then find a compatible used server or workstation bundle. Once the platform is selected, allocate the remaining money to RAM and GPUs. A consumer CPU such as a Ryzen 7 or Core Ultra chip is unlikely to satisfy the core-count and lane requirements.
Before choosing hardware, define which workload actually needs 64GB of VRAM. Multiple GPUs don’t always combine their memory into one usable pool, especially for gaming, many CAD applications, and some AI software. You may need a specific multi-GPU or model-parallel workflow for separate cards to help. If the workload can be split across jobs, two cards may be more practical than four.
The VRAM is mainly for local AI work alongside large CAD, 3D, and media projects. Cloud compute isn’t attractive because the machine would be used continuously, so I’d rather make the large upfront purchase.
A used server platform may be the most realistic option. Older EPYC systems can offer many cores, ECC support, IPMI, and lots of PCIe connectivity, while older workstation platforms such as Xeon or Xeon W can sometimes be found cheaply with suitable boards. Check the exact slot wiring carefully: four physical x16 slots may actually run at x8 or x4, and some slots can be disabled depending on the installed drives or CPUs.
That’s the direction I’m leaning toward, but local prices for used EPYC and Threadripper hardware are higher than I expected. I’m trying to avoid spending the entire budget on the platform before buying GPUs.