Hardware
Local AI hardware starts with memory capacity.
Vendor-advertised accelerator memory, paired with explicit qualifications and open-model weight estimates.
Representative catalog
Advertised memory by hardware
Dedicated VRAM/HBM and unified memory are labeled separately. Bars use a logarithmic scale.
| Hardware | Class | Memory | Type | Qualification | Source |
|---|---|---|---|---|---|
| NVIDIA GeForce RTX 4090Consumer GPU; software support and usable free VRAM depend on the system configuration. | consumer gpu | 24 GB | GDDR6X | Dedicated | Vendor ↗ |
| AMD Radeon RX 7900 XTXBackend and kernel support should be checked separately from capacity. | consumer gpu | 24 GB | GDDR6 | Dedicated | Vendor ↗ |
| NVIDIA GeForce RTX 5090Consumer GPU; advertised capacity does not guarantee runtime compatibility. | consumer gpu | 32 GB | GDDR7 | Dedicated | Vendor ↗ |
| AMD Radeon PRO W7900Workstation GPU with ECC support. | workstation gpu | 48 GB | GDDR6 ECC | Dedicated | Vendor ↗ |
| NVIDIA H100 SXMH100 NVL is a separate 94GB configuration; this row represents the 80GB SXM capacity. | datacenter accelerator | 80 GB | HBM | Dedicated | Vendor ↗ |
| NVIDIA H200Datacenter accelerator capacity. | datacenter accelerator | 141 GB | HBM3e | Dedicated | Vendor ↗ |
| AMD Instinct MI300XDatacenter accelerator capacity. | datacenter accelerator | 192 GB | HBM3 | Dedicated | Vendor ↗ |
| AMD Instinct MI325XDatacenter accelerator capacity. | datacenter accelerator | 256 GB | HBM3E | Dedicated | Vendor ↗ |
| Apple Mac Studio M3 Ultra (maximum configuration)Maximum configurable unified memory shared by CPU, GPU, and the operating system; not equivalent to dedicated VRAM. | unified memory system | 512 GB | Unified memory | Unified / shared | Vendor ↗ |
Connect models to memory
See which open weights fit.
The open-model catalog estimates Q4 and Q8 weight memory with conservative format and loading overhead, then compares Q4 estimates against capacity tiers from 8GB through 512GB.
Open the model hardware-fit matrix →