Built for How Business Buys.
Built for How Business Buys.
A server-pedigree sTR5/WRX90 CPU platform, built for massively multi-threaded work, sits next to the family's flagship 96 GB RTX PRO 6000 Blackwell GPU. 8-channel DDR5 memory and 6× PCIe 5.0 x16 slots mean the platform is rated for far more than what ships in the box — room to add memory, storage, and GPUs as workloads grow.
A server-pedigree CPU platform next to the family's flagship GPU is built for work that leans on both at once:
This configuration ships with one GPU and one populated M.2 slot — the platform itself has room to grow:
Front: 2× USB 3.2 Gen1 Type-A + 1× USB 3.2 Gen1 Type-C, power button, audio/mic jack, and 4× 3.5″/2.5″ SATA hot-swap bays installed (4 more optional — 8 total possible). Rear: dual 10Gb/s LAN (Broadcom BCM57416) plus a dedicated MLAN management port (Realtek RTL8211FD), USB and audio connectivity, and the 2000 W power inlet.
8× DIMM slots on 8-channel DDR5, 6× PCIe 5.0 x16 slots (platform-rated for up to 4× dual-slot GPUs), and 4× M.2 SSD slots with only one populated by the 1 TB OS drive.2 Headroom built in, not an afterthought.
The listing's own AI Features, translated into workloads that lean on both the CPU and the GPU:
Accelerate AI development and inference workloads, and build agentic AI applications.
Accelerate your data pipelines.
Accelerate scientific breakthroughs with enhanced accuracy.
Deliver stunningly realistic designs and simulations with speed and precision.
Boost performance for live media and video pipelines.
Drive new possibilities for game development with AI.
Scene photos are AI-generated illustrations of typical deployments, not photographs of this exact product.
An optional add-on that offloads GPU memory to flash, reducing the VRAM full fine-tuning actually requires:
| Target Model Size | VRAM Needed (Without aiDAPTIV+) | VRAM Needed (With aiDAPTIV+) |
|---|---|---|
| ~13B | 120–200 GB | 32 GB |
| ~34B | 250–350 GB | 96 GB — this workstation's GPU |
| ~70B | 500–700 GB | 192 GB |
Phison aiDAPTIV+ SSD is an optional add-on and is not included by default. VRAM figures are vendor-published estimates for full fine-tuning.
The "Zaurion Ruby" trim name is used across more than one ABS chassis — here's how this exact configuration is positioned:
The server-pedigree Threadripper PRO platform, sTR5/WRX90, paired with the family's flagship 96 GB GPU.
A separate, compact mid-tower lineup that shares only the "Ruby" trim name — a different chassis and CPU platform. Confirm the model number before ordering.
This chassis lineup also offers the identical 96 GB RTX PRO 6000 Blackwell on other CPU platform tiers — compare options on this listing.
6× PCIe 5.0 x16 slots rated for up to 4× dual-slot GPUs — headroom beyond this single-GPU configuration.
| ZRP7975WX-RP60001 | |
|---|---|
| CPU | AMD Ryzen Threadripper PRO 7975WX — sTR5 socket, WRX90 chipset1 |
| GPU | NVIDIA RTX PRO 6000 Blackwell — 96 GB GDDR7 |
| Memory | 64 GB DDR5, 8-channel |
| Storage | 1 TB M.2 NVMe (OS) + 1.92 TB SSD (Data) |
| Networking | 2× 10 Gb/s LAN (Broadcom BCM57416) + MLAN management port (Realtek RTL8211FD) |
| Motherboard | Gigabyte MH53-G40 (Gigabyte W773-H5D-AA01 barebone) |
| Power | Single 2000 W ATX 80 PLUS Gold |
| Expansion | 8× DIMM (8-channel DDR5), 6× PCIe 5.0 x16, 4× M.2 SSD, 4× hot-swap bays installed |
| OS | Ubuntu |
| Optional | Phison aiDAPTIV+ SSD; OpenClaw/NemoClaw agent framework; local LLM package |
A server-pedigree sTR5/WRX90 platform built for heavy, massively multi-threaded CPU work — simulation, encoding, and batch rendering — running alongside the same 96 GB RTX PRO 6000 Blackwell GPU used elsewhere in this lineup.
No — this is the tower-class Zaurion Ruby, a multi-GPU-capable platform. A separate compact mid-tower lineup shares only the trim name. Confirm the model number (ZRP7975WX-RP60001) before ordering.
8× DIMM slots (8-channel DDR5), 6× PCIe 5.0 x16 slots rated for up to 4× dual-slot GPUs, and 4× M.2 SSD slots with only one populated. This unit ships with one GPU on a single 2000 W supply.
Not by default. OpenClaw/NemoClaw and a local LLM package are optional add-ons configured at order time.
An optional SSD that offloads GPU memory to flash, reducing the VRAM full fine-tuning requires — useful for fully fine-tuning models larger than 96 GB alone can hold. Not included by default.
It is TAA compliant and assembled in the USA, with compliance documents available for purchase review.