Eight 600W GPUs Without Throttling? TFNEW ZH-6201/SW
TL;DR: TFNEW launched the ZH-6201/SW, a 6U dual-socket GPU compute server, on September 21, 2026. Built for eight 600 W full-height dual-width GPU cards — Switch-topology configs reach ten — it pairs 2x 4th/5th Gen Intel Xeon Scalable with 32 RDIMM slots, 12/13 PCIe x16 slots, eight CRPS power supplies in N+N/N+M redundancy, and a DC-SCM 2.0 hot-swappable management module. Get specs and pricing: sales@tfnew.com.
Core Specifications (with Plain-Language Terms)
The ZH-6201/SW is a 6U rackmount server (265 mm tall, 850 mm deep) whose layout exists to keep 600 W class accelerators running at their rated TDP (thermal design power — the heat a GPU is engineered to dissipate under full load; running above it means throttling). Full-height, full-length, dual-width cards — the biggest accelerator form factors — need vertical space and airflow, which is why the chassis is 6U instead of squeezed into a denser box. Computing: two 4th/5th Gen Intel Xeon Scalable processors on an Intel C741 chipset, up to 32 RDIMM or RDIMM-3DS modules at 5600 MT/s. GPU fabric: a passthrough backplane supports 12 PCIe x16 standard slots and 8 GPUs; the PCIe 5.0 Switch option supports 13 x16 slots and up to 10 GPUs. Storage: up to 12 hot-swap 3.5-inch SAS/SATA/NVMe bays. Cooling: 8x 8056 plus 5x 8080 fan modules. Power: 8x CRPS power supplies (2000 W or 2700 W) in N+N or N+M redundancy — dual-feed capable, so a rack keeps the node alive even if one feed drops. Management rides on a DC-SCM 2.0 module — a removable, standards-based management card with an AST2600 BMC, IPMI 2.0 and Redfish — so the management plane can be swapped without touching the compute board.
Specs at a Glance
| Item | ZH-6201/SW Value |
|---|---|
| Form factor | 6U rackmount, 445 W × 265 H × 850 D mm |
| Processor | 2x 4th/5th Gen Intel Xeon Scalable, Intel C741 |
| Memory | Up to 32x RDIMM / RDIMM-3DS, up to 5600 MT/s |
| GPU | 8x 600 W full-height dual-width cards (up to 10 with Switch) |
| PCIe | 12x x16 passthrough / 13x x16 with PCIe 5.0 Switch |
| Storage | Up to 12x 3.5" hot-swap SAS/SATA/NVMe bays |
| Cooling | 8x 8056 + 5x 8080 fan modules |
| Power | 8x CRPS 2000 W / 2700 W, N+N or N+M redundant |
| Management | DC-SCM 2.0, AST2600 BMC, IPMI 2.0 / Redfish |
| OS | Windows Server, Linux, VMware ESXi |
Why 6U, Eight Fans and Eight PSUs Are the Product
A spec sheet with "8 GPUs" hides the real engineering question: can the chassis cool them and keep them fed? At 600 W per card, eight cards pull ~4.8 kW of GPU power alone — heat and AC are the constraints that turn a board into a production system. The ZH-6201/SW answers with physical headroom (6U), eight large 8056 plus five 8080 fan modules, and eight CRPS supplies that run N+N or N+M so a single feed failure never idles the fleet. The DC-SCM 2.0 management card takes the same repair-first philosophy to the control plane: a failed BMC module is replaced in seconds instead of taking a server down.
Where It Fits
- AI training and deep learning: 600 W accelerators at full TDP, eight or ten per node
- Cloud gaming and VR: GPU-density nodes for real-time rendering fleets
- Smart-city and healthcare analytics: big-data + model workloads sharing one 6U footprint
- HPC and media transmission: memory- and bandwidth-hungry scientific and rendering jobs
Why the Numbers Matter
- 8x 600 W GPUs (10 with Switch): flagship accelerator density with full-TDP operation
- 32x RDIMM at 5600 MT/s: resident datasets for training and inference
- N+N / N+M across 8 PSUs: dual-feed resilience rare in this class
- DC-SCM 2.0 + AST2600: next-generation hot-swappable out-of-band management
- 12x 3.5" SAS/SATA/NVMe bays: model data and checkpoints stored at the node
What TF NEW Says
"GPU servers live or die on heat and power, not the sticker GPU count," said a TF NEW product manager. "The ZH-6201/SW is built around the physics: 6U of airflow room, thirteen large fans, and eight CRPS supplies in N+N redundancy so eight 600 W cards can actually run at 600 W. And with DC-SCM 2.0, the management plane is as serviceable as the PSUs."
Quality and Availability
Every ZH-6201/SW runs 48 hours of burn-in under full GPU load before shipping, with thermals validated across the 0°C–35°C operating range and the DC-SCM management module verified on IPMI/Redfish. OS options — Windows Server, Linux and VMware ESXi — are validated in the factory.
Pricing and Availability
The ZH-6201/SW is available now. Pricing varies with processor, memory, GPU interconnect (passthrough vs Switch) 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-6201/SW hold? Eight 600 W full-height dual-width cards, up to ten with the PCIe 5.0 Switch topology.
- What makes the chassis able to cool them? 6U of height, eight 8056 + five 8080 fan modules, and 850 mm of depth airflow.
- What is DC-SCM 2.0? A removable management module standard — the AST2600 BMC card can be swapped without taking the server down.
- What power redundancy does it offer? Eight CRPS supplies (2000 W / 2700 W) in N+N or N+M configuration with dual-feed capability.
- Which OSes are supported? Windows Server, Linux and VMware ESXi, validated at the factory.