ABS AI Workstation - 144GB Aggregate VRAM (2x 72GB RTX PRO 5000 Blackwell), AMD Ryzen Threadripper PRO 9975WX, 128GB ECC DDR5, 2TB+4TB NVMe M.2, 2000W 80+ Gold, Ubuntu (Zaurion ZRP9965WX-RP500072)
- Item #: 9B-89-850-013
- Mfr. Part #: ZRP9975WX-RP500072X2
- Compliance Documents Required for Purchase
- Assembled in USA
- TAA Compliant
- Operating System: Ubuntu
- CPU: AMD Ryzen Threadripper PRO 9975WX
- GPU: 2x RTX Pro 5000 Blackwell (72GB GDDR7)
- Motherboard: Gigabyte MW83-RP0
- Memory: 128GB DDR5 Server RAM - ECC, Registered
- OS SSD: 2TB M.2 NVMe (Installed)
- Data SSD: 4TB M.2 NVMe (Installed)
- LAN: 2 x 10GB/s LAN, 1 x 10/100/1000 Mbps Management LAN
- PSU: Single 2000W ATX 80 PLUS Gold power supply
- Barebone: Gigabyte W773-W80
- AI Agent Ready: Optional pre-installed OpenClaw (enterprise AI agent framework)
- Local LLM Support: Optional Qwen 3.5 27B or higher models for on-prem AI inference
Built to Be Shared.
Two independent 72 GB GPUs and 32 CPU cores mean this one tower doesn't have to sit in a queue. 144 GB of aggregate VRAM, split across two real people or two real workloads — not a single card everyone waits their turn for.
What This Pairing Actually Buys You
An AMD server-pedigree CPU platform next to two independent 72GB GPUs, built for concurrent use:
144 GB Aggregate, Two Independent Cards
- Two independent 72 GB GPUs — 144 GB aggregate, not a single unified pool. Two people, or two workloads, each get dedicated GPU access.
- Blackwell 5th-gen Tensor Cores with FP41 accelerate agentic and generative AI on each card.
- 4th-gen RT Cores1 drive photoreal rendering previews on either GPU.
- Universal MIG1 can partition either card further, for even more concurrent users.
32 Cores, 64 Threads, Split as Needed
- 32 cores / 64 threads on the sTR5/WRX90 platform — enough parallelism to keep more than one workflow moving without either starving the other.
- Fewer cores and threads than this lineup's Intel-based dual-GPU alternative, on the same Compare Options list — a genuine platform difference, not a universal rule about either vendor.
- 8-channel DDR5 memory keeps both GPUs supplied with data without the CPU becoming the shared bottleneck.
- ECC memory support brings server-class reliability to long shared-use sessions.
Who's Sharing This Workstation
Two independent GPUs mean this isn't just a faster single-user machine - it's a machine built to serve more than one person at once. Some illustrative roles it can cover simultaneously:
The Model Trainer
Runs a long fine-tuning job against one of the two 72GB GPUs, using a share of the 32 CPU threads for data loading.
The Inference User
Queries a served model on the second GPU in real time, without waiting for the trainer's job to finish.
The Builder
Compiles code, runs tests, and orchestrates agents on the remaining CPU threads, barely touching either GPU.
The Renderer
Picks up whichever GPU is free for a batch render job, once the trainer's or the inference user's session wraps up.
Illustrative role framing, not a literal guaranteed resource-partitioning scheme for any specific software or OS scheduler.
Six Jobs, Shared Across a Team
The listing's own AI Features, translated into what a dual-GPU, multi-user machine actually enables:
AI Development
One GPU trains while the other serves inference, so a finished result doesn't have to wait for a free slot on a single shared card.
Data Science
32 threads split across two people's preprocessing pipelines, each feeding their own GPU.
HPC
Two independent simulations can run side by side instead of taking turns on one GPU.
AI-Driven Rendering & Graphics
One GPU renders a production scene while the other previews the next set of changes.
Video Content & Streaming
Export a finished cut on one GPU while continuing to edit the next one on the other.
Game Development
One teammate runs the live engine build while another tests changes on the second GPU, on the same machine.
Scene photos are AI-generated illustrations of typical deployments, not photographs of this exact product.
As Configured
| ZRP9975WX-RP500072X2 | |
|---|---|
| CPU | AMD Threadripper PRO 9975WX — 32-core / 64 threads, sTR51 |
| GPU | 2x NVIDIA RTX PRO 5000 Blackwell — 72 GB GDDR7 each, 144 GB aggregate |
| Memory | 128 GB DDR5, ECC Registered |
| Storage | 2 TB M.2 NVMe (OS) + 4 TB M.2 NVMe (Data) |
| Networking | 2× 10 Gb/s LAN + 1× 10/100/1000 Mbps management LAN |
| 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 (2 populated), 4× M.2 (2 populated, 2 open) |
| OS | Ubuntu |
| Optional | OpenClaw agent framework (pre-installed); Qwen 3.5 27B+ local LLM package |
Quick Answers Before You Buy
Can this workstation really be shared by more than one person?
Two independent 72GB GPUs mean two people (or two workloads) can each have dedicated GPU access at the same time. 32 CPU cores / 64 threads support both without either starving the CPU-side work.
How does this compare to the Intel-based dual-GPU option in this lineup?
Both carry two independent 72GB RTX PRO 5000 Blackwell GPUs and 128GB DDR5. This configuration uses an AMD Threadripper PRO CPU platform instead; the Intel-based option in this lineup carries more CPU cores and threads.
Is the 144GB of VRAM one big pool or two separate GPUs?
Two independent 72GB GPUs — 144GB aggregate, not a single unified pool. Two people, or two workloads, each get dedicated GPU access.
Does it come with an AI agent framework or local LLM pre-installed?
The AI agent framework (OpenClaw) is pre-installed; a Qwen 3.5 27B or higher local LLM package is an optional add-on configured at order time.
Is it eligible for government / public-sector purchasing?
It is TAA compliant and assembled in the USA, with compliance documents available for purchase review.
How is it supported?
1-year parts and labor limited warranty backed by ABS manufacturer support. As a Build to Order system, it is custom built and stress-tested after purchase.
Warranty & Returns
Warranty, Returns, And Additional Information
Warranty
- Limited Warranty period (parts): 1 year
- Limited Warranty period (labor): 1 year
Return Policies
- Return for refund within: non-refundable
- Return for replacement within: non-replaceable
- This item is covered by NeweggBusiness.com's Manufacturer Only Return Policy.
- Read full return policy for details.
Manufacturer Contact Info
- Manufacturer Product Page
- Manufacturer Website
- Support Phone: 1-800-685-3471
- Support Email: [email protected]
- Support Website
- View other products from ABS















