ABS AI Workstation - 72GB VRAM (NVIDIA RTX PRO 5000 Blackwell 72GB), AMD Ryzen Threadripper PRO 9965WX, 128GB ECC DDR5, 2TB+4TB NVMe M.2, 2000W 80+ Gold, Ubuntu (Zaurion ZRP9965WX-RP500072)
- Item #: 9B-89-850-011
- Mfr. Part #: ZRP9965WX-RP500072
- Compliance Documents Required for Purchase
- Assembled in USA
- TAA Compliant
- Operating System: Ubuntu
- CPU: AMD Ryzen Threadripper PRO 9965WX
- GPU: RTX Pro 5000 Blackwell (72 GB GDDR7)
- Motherboard: Gigabyte MH53-G40
- 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-H5D-AA01
- 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
Every Core Has a Job.
The same 72 GB RTX PRO 5000 Blackwell as this chassis lineup's entry tier, but paired with a 24-core, 48-thread Threadripper PRO instead of a smaller Xeon-based CPU. The difference shows up before the GPU ever gets involved — in how fast data gets loaded, tokenized, and handed off, and how many things can happen at once.
What This Pairing Actually Buys You
A server-derived CPU platform with real parallelism, next to a memory-rich professional GPU:
72 GB of GDDR7, Kept Fed
- 72 GB of VRAM — identical to the entry Xeon-based tier in this lineup, so the CPU is the only thing that changes here.
- Blackwell 5th-gen Tensor Cores with FP41 accelerate agentic and generative AI at architecture-native speed.
- 4th-gen RT Cores1 drive photoreal rendering previews.
- PCIe Gen 5 keeps the CPU-to-GPU link from becoming the bottleneck when 48 threads are feeding it.
24 Cores, 48 Threads, All Working
- 24 cores / 48 threads on the sTR5/WRX90 platform give real headroom for parallel data pipelines, not just single-threaded orchestration.
- 8-channel DDR5 memory keeps 48 threads supplied with bandwidth instead of stalling on memory access.
- ECC memory support brings server-class reliability to long preprocessing and training runs.
- Built to keep pace with the GPU — more threads means fewer moments where the GPU is idle waiting on the CPU.
CPU and GPU, Working the Same Job
Neither one waits on the other. Here's how a typical AI workload actually splits across the two:
- Data ingestion and cleaning
- Tokenization and preprocessing
- Agent orchestration and tool-calling
- Build, compile, and test jobs
- Routing requests across multiple users or projects
- Model inference and fine-tuning
- Ray-traced and neural rendering
- Video encode / decode acceleration
- Physics and simulation compute
- Real-time engine previews
On the entry Xeon-based tier, a smaller CPU can still fall behind when several of these run at once. 24 cores and 48 threads give this configuration room to run more of them in parallel.
Six Jobs, Balanced Across CPU and GPU
The listing's own AI Features, translated into what this specific CPU/GPU pairing actually does:
AI Development
Same 72GB ceiling for local models as the entry tier, but 48 threads mean the agent-orchestration and tool-calling layer around the model doesn't wait on the CPU.
Data Science
48 threads handle heavier parallel ETL and preprocessing pipelines than the entry tier, feeding the GPU continuously instead of intermittently.
HPC
The server-derived sTR5/WRX90 platform with ECC memory and 8-channel DDR5 keeps memory bandwidth ahead of a 48-thread CPU on long simulation runs.
AI-Driven Rendering & Graphics
CPU-side scene prep (geometry, physics) keeps pace with GPU-side ray tracing instead of bottlenecking it, on the same 72GB GPU.
Video Content & Streaming
48 threads handle multi-stream encode/transcode queuing while Blackwell-generation video engines handle the GPU-accelerated encode itself.
Game Development
Faster build and compile times from the extra threads, while the same 72GB GPU drives real-time engine previews.
Scene photos are AI-generated illustrations of typical deployments, not photographs of this exact product.
As Configured
| ZRP9965WX-RP500072 | |
|---|---|
| CPU | AMD Threadripper PRO 9965WX — 24-core / 48 threads, sTR51 |
| GPU | NVIDIA RTX PRO 5000 Blackwell — 72 GB GDDR7 |
| 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, 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
What does the extra CPU parallelism actually change versus the entry tier?
Both configurations carry the identical 72GB RTX PRO 5000 Blackwell GPU. This one adds a 24-core/48-thread Threadripper PRO instead of a smaller Xeon-based CPU — more headroom for data preprocessing, agent orchestration, and build jobs alongside GPU work.
How does work actually split between the CPU and the GPU here?
The CPU handles data ingestion, tokenization, agent orchestration and tool-calling, and build/compile jobs. The GPU handles model inference and fine-tuning, ray-traced rendering, and video encode/decode.
How much storage headroom is there?
This configuration populates 2 of the platform's 4 M.2 slots (2TB OS + 4TB data), leaving 2 M.2 slots and all 4 installed SATA hot-swap bays (plus 4 more optional) open.
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















