Two 600W GPUs per Rack Unit? TFNEW ZH-5201/SW
TL;DR: TFNEW launched the ZH-5201/SW, a 5U GPU compute server, on September 21, 2026. It packs up to ten 600 W full-height dual-width accelerators into five rack units — the highest GPU-per-U density in TF NEW's line — on 2x 4th/5th Gen Intel Xeon Scalable, 32 DDR5 slots and a PCIe 5.0 Switch with 13 x16 slots, so a 42U rack can sit ~80 GPUs while keeping room for networking. Get specs and pricing: sales@tfnew.com.
Core Specifications (with Plain-Language Terms)
The ZH-5201/SW is a 5U rackmount server (219 mm tall, 436 mm wide, 870 mm deep) sized for one purpose: maximum GPU count per rack unit. "Rack unit" (U) is the 44.45 mm height increment of a server rack; at ten GPUs in five U, this chassis delivers 2 GPU per U — the density that lets a full 42U rack mount eight such nodes (~80 GPUs) and still leave a couple of U for top-of-rack switches and cabling troughs. Inside: two 4th/5th Gen Intel Xeon Scalable processors on an Intel C741 chipset, 32 DDR5 slots at 4400/4800/5600 MHz, 12 hot-swap 3.5-inch bays (2.5-inch compatible), a PCIe 5.0 Switch fabric exposing 13 x16 standard slots so up to ten full-height full-length dual-width accelerators can all talk at full fabric speed, ten high-performance fans, and 4x CRPS hot-swap supplies in 3+1 redundancy (2000 W or 2700 W). Management is a DC-SCM 2.0 module with an AST2600 BMC (IPMI 2.0 / Redfish).
Specs at a Glance
| Item | ZH-5201/SW Value |
|---|---|
| Form factor | 5U rackmount, 219 H × 436 W × 870 D mm |
| GPU density | Up to 10x 600 W FHF L dual-width cards — 2 GPU / U |
| Processor | 2x 4th/5th Gen Intel Xeon Scalable, Intel C741 |
| Memory | 32x DDR5 slots, 4400/4800/5600 MHz |
| PCIe | PCIe 5.0 Switch, 13x x16 standard slots |
| Storage | 12x 3.5" bays (2.5" compatible) |
| Power | 4x CRPS hot-swap, 3+1 redundant, 2000 W / 2700 W |
| Cooling | 10 high-performance fan modules |
| Management | DC-SCM 2.0, AST2600 BMC, IPMI 2.0 / Redfish |
| Environment | 0°C–35°C operating range |
Why GPU-per-U Density Is the Rack-Scale Metric
Datacenter GPU fleets don't rent servers, they rent racks. The question buyers actually ask is GPUs per rack: how many accelerators can a 42U column hold once power, cooling and switching are accounted for. At 2 GPU/U, the ZH-5201/SW turns a 42U rack into an ~80-GPU cluster with space left for networking — versus deeper chassis designs that spend the same cooling budget on fewer cards per U. The PCIe 5.0 Switch fabric is what makes the density usable: ten cards at full x16 each, no per-slot lane-sharing surprises, which is the difference between a density figure and a density that trains.
Where It Fits
- GPU-dense private clouds: serve more vGPUs or GPU instances from the same rack footprint
- Large-model training clusters: ten accelerator fabric nodes for LLM fine-tuning and MoE workloads
- Cloud gaming and VR fleets: real-time rendering density that scales by the rack
- HPC and rendering farms: memory-heavy jobs on 32 DDR5 slots inside a tight footprint
Why the Numbers Matter
- 10 GPUs in 5U (2 GPU/U): the density line that makes ~80 GPU racks realistic
- 13x x16 PCIe 5.0 via Switch: full-bandwidth fabric for ten-card training jobs
- 32x DDR5 up to 5600 MHz: model weights and activations resident per node
- 3+1 CRPS redundancy: hot-swap power without giving up a slot to a spare unit
- DC-SCM 2.0 + AST2600: serviceable management plane, IPMI 2.0 and Redfish automation
What TF NEW Says
"Fleet buyers price racks, not boxes," said a TF NEW product manager. "The ZH-5201/SW is engineered to that math: ten 600 W GPUs in five U, full x16 each over a PCIe 5.0 Switch, on two Xeon Scalable with 32 DDR5 slots. A 42U rack becomes an eighty-GPU cluster with room for network — density that stays usable at the top of the rack."
Quality and Availability
Every ZH-5201/SW runs 48 hours of burn-in under full GPU load before shipping, with the Switch fabric validated across all thirteen x16 lanes and thermals verified in the 0°C–35°C envelope. It ships deployment-ready with DC-SCM 2.0 baselines and a documented GPU compatibility matrix.
Pricing and Availability
The ZH-5201/SW is available now. Pricing varies with processor, memory, GPU count and power configuration. Contact sales@tfnew.com or use the request-a-quote form at www.tfnew.com/contact for configuration guidance and lead time.
About TF NEW
TF NEW (TF New Technology) is an enterprise AI infrastructure provider based at www.tfnew.com. It designs and delivers private AI, compute, networking, storage, cooling and edge systems as deployment-ready solutions — AI servers, AI workstations, edge AI, liquid cooling and enterprise rack servers — for AI factories, datacenters and edge projects worldwide.
Media Contact
TF NEW | sales@tfnew.com | www.tfnew.com/contact
FAQ
- How many GPUs can the ZH-5201/SW hold? Up to ten 600 W full-height full-length dual-width cards — 2 GPU per rack unit.
- What is the GPU-per-U advantage? A 42U rack can host ~80 GPUs with room left for top-of-rack networking.
- How are the GPUs connected? Through a PCIe 5.0 Switch fabric exposing 13 x16 standard slots at full bandwidth.
- What processors and memory does it use? 2x 4th/5th Gen Intel Xeon Scalable with 32 DDR5 slots up to 5600 MHz.
- Is the management plane serviceable? Yes — a DC-SCM 2.0 module with AST2600 BMC (IPMI 2.0 / Redfish) is hot-swappable.