GT5 Ultra professional AI workstation

Specifications

• GPU: RTX Pro 2000/4000 SFF*2 / Full-Height NVIDIA RTX PRO 5000 BlackwellSupplier claim
• AI Compute: 770-3511 TOPS TOPSSupplier claim
• Size: 5.6L 235*117*213mm mmSupplier claim
• CPU: Intel Ultra 5 225Supplier claim
• RAM: DDR5 (U-DIMM*2)Supplier claim
• ROM: PCIe 4.0 + 2*M.2 2280Supplier claim
• I/O: HDMI*2.0*2; DP1.4*2; USB 5G*2; USB 10G*4; USB 2.0*2; RJ45*1; Wi-Fi6; Audio Jack*3Supplier claim
• GPU: NVIDIA Pro 4000 SFF*2 / NVIDIA Pro 2000*2Supplier claim
• Total Compute: 1450 TOPS / 1090 TOPS TOPSSupplier claim
• Series tagline: Portable & MultimodalSupplier claim
• ROM: PCIe 4.0 + Dual M.2 SlotsSupplier claim
• I/O: Video Output:：2×DP1.4 + 1×HDMI®2.0; USB Ports: 8 Total（2×USB 5G + 4×USB 10G + 2×USB 2.0）; Network：1×RJ45 + Wi-Fi6; Audio： 3× Audio JacksSupplier claim
• GPU compatibility: Compatible with Dual NVIDIA RTX Pro 2000 / 4000 SFF (16/24GB VRAM x2, Discrete Graphics)Supplier claim
• Total AI Compute: Up to 1540 TOPS Total AI Compute (Theoretical Peak) TOPSSupplier claim
• TOPS by GPU: NVIDIA Pro 4000 SFF×2: 1,450 TOPS; NVIDIA Pro 2000×2: 1,090 TOPS TOPSSupplier claim
• Chassis Volume: 5.6L 235*117*213mm mmSupplier claim
• Test platform: Dual NVIDIA RTX PRO 4000 Blackwell SFF Edition 24GB; LLM Qwen3-4B; real-time and offline transcription Qwen3-ASR-0.6BSupplier claim
• AI Inference Concurrency: Up to ~60 concurrent LLM inference threadsSupplier claim
• System Aggregate Throughput: Peak 1,617 tok/s (multi-stream concurrent) tok/sSupplier claim
• Offline Speech-to-Text (ASR): 15 min for 1 hour of audio (≈4× real-time)Supplier claim
• Real-time ASR Concurrency: Up to 10 concurrent real-time ASR streamsSupplier claim
• Model benchmark table: Model | Inference Framework | Quantization Format | Single-Stream Peak (tok/s) | Optimal Concurrency | Multi-Stream Peak (tok/s) | Max Context Length | ITL Latency ;; Qwen3.5-4B-AWQ-4bit / vLLM / AWQ-4bit / 87.27 / input=256, output=32 / 1,401 / 262K / 9~15ms ; Qwen3.6-35B-A3B-UD-Q4_K_M / Llama.cpp / Q4_K_M / 70.66 / input=1024, output=32 / — / 65K / 10~13ms ; Qwen3.5-9B-GPTQ-4bit / vLLM / GPTQ-4bit / 53.32 / input=256, output=32 / 958 / 262K / 16~18ms ; Qwen3.6-27B-Q4_K_M / Llama.cpp / Q4_K_M / 16.83 / input=1024, output=32 / — / 131K / 48~62msSupplier claim