Minisforum is pushing local AI computing into much larger model territory with its new MS-S1 MAX-P495, a workstation designed to run large language models without sending workloads to the cloud. The Mini PC combines AMD’s Ryzen AI Max+ PRO 495 with up to 192GB of unified memory and allows up to 160GB of that memory to be allocated as VRAM, giving it enough capacity for AI models that would normally require much larger systems.
The Ryzen AI Max+ PRO 495 is based on AMD’s Zen 5 architecture and features 16 cores and 32 threads, with boost speeds of up to 5.2 GHz. It also includes Radeon 8065S graphics and delivers up to 55 TOPS of NPU performance, while total AI compute reaches up to 131 TOPS.
Memory is one of the biggest upgrades over the earlier MS-S1 MAX with the Ryzen AI Max+ 395. The previous system supported up to 128GB of unified memory and 96GB of VRAM, while the MS-S1 MAX-P495 raises those limits to 192GB and 160GB, respectively. This additional capacity allows the workstation to handle larger models locally.
Minisforum says the system can run 300B-class models, with DeepSeek V4 Flash 284B in Q4 configuration reaching around 12 tokens per second during its testing. Two 192GB systems can also be connected to run Qwen3.5-397B at approximately 16 tokens per second. For larger deployments, the platform supports 2U rack configurations, and Minisforum reports that a four-system setup successfully ran DeepSeek-R1 671B Q4.
The MS-S1 MAX-P495 is available for pre-order with 192GB of RAM and a 2TB SSD for $9,249. A 20% promotional discount lowers the price to $7,399, with shipping expected to begin in mid-October.
Minisforum MS-S1 MAX-P495 – Specifications
| Specification | Minisforum MS-S1 MAX-P495 |
|---|---|
| Processor | AMD Ryzen AI Max+ PRO 495 |
| CPU Architecture | Zen 5 |
| CPU Cores / Threads | 16 / 32 |
| Maximum Clock Speed | Up to 5.2 GHz |
| GPU | Radeon 8065S Graphics |
| NPU Performance | Up to 55 TOPS |
| Total AI Compute | Up to 131 TOPS |
| Maximum Unified Memory | 192GB |
| Maximum Allocatable VRAM | 160GB |
| Local AI Models | Up to 300B-class |
| Tested Model | DeepSeek V4 Flash 284B Q4 |
| Single-System Performance | ~12 tok/s |
| Dual-System Model | Qwen3.5-397B |
| Dual-System Performance | ~16 tok/s |
| Rack Support | 2U deployment |
| Storage Configuration | 2TB SSD |
| Pre-order Price | $9,249 |
| Discounted Price | $7,399 |
| Shipping | Expected mid-October |
