8,000 MT/s Memory in an 8-GPU Server? TFNEW S768G6
TL;DR: TFNEW launched the S768G6, an enterprise GPU server, on September 21, 2026. A 6U AI supercomputing platform with up to 2x Intel Xeon 6 processors (350 W), 32 DDR5 DIMM slots with MCR DIMM support up to 8,000 MT/s, and up to eight 600 W GPU cards in dual-passthrough, single-passthrough or Switch schemes. Get specs and pricing: sales@tfnew.com.
Core Specifications (with Plain-Language Terms)
The S768G6 is a 6U rackmount AI supercomputing server built around a memory-bandwidth story: MCR DIMM, or Multiplexer Combined Ranks, is a DDR5 technology that doubles the effective ranks a DIMM presents to the memory controller, letting a single slot run at up to 8,000 MT/s — the fastest DIMM class in TF NEW's rack line. That matters for AI training, where model weights and optimizer states live in memory and bandwidth decides iteration speed. The platform: up to 2x Intel Xeon 6 processors at 350 W TDP, 32 DDR5 DIMM slots (MCR DIMM to 8000 MT/s; RDIMM to 6400 MT/s at 1DPC or 5200 at 2DPC), up to eight 600 W GPU cards, and three GPU schemes: 8-card dual-passthrough (standard), 8-card single-passthrough (optional) and 8-card Switch expansion — the newest Xeon 6 generation with 12 hot-swap front bays (3.5"/2.5") and U.2 NVMe per scheme (2x dual-passthrough, 4x Switch).
Specs at a Glance
| Item | S768G6 Value |
|---|---|
| Chassis | 6U AI supercomputing server |
| GPU scheme | 8-card dual-passthrough (std) / single-passthrough / Switch |
| Processor | Up to 2x Intel Xeon 6, TDP up to 350 W |
| Memory | 32 DDR5 DIMM; MCR DIMM up to 8000 MT/s; RDIMM 6400 (1DPC) / 5200 (2DPC) |
| Storage | 12x front hot-swap 3.5"/2.5" bays; U.2 NVMe per scheme |
| GPU | Up to 8 cards at 600 W each |
Why Memory Bandwidth Is the Hidden Lever in AI Supercomputing
An 8-GPU node is only as fast as its slowest pipeline stage, and on many training runs that stage is memory. The S768G6 attacks it directly: 32 DIMM slots that accept MCR DDR5 at 8000 MT/s — the highest-speed memory option in the lineup — fed from two 350 W Xeon 6 processors. Add the three GPU topologies (dual-passthrough for per-GPU CPU affinity, single-passthrough for flow simplicity, Switch for GPU-to-GPU fabric) and 600 W accelerators, and the 6U box is a complete AI compute cell: CPUs, memory fabric and eight GPUs on one power rail plan.
Where It Fits
- AI training nodes: MCR DIMM bandwidth feeding 8-GPU training
- Foundation-model work: Xeon 6 + 32 DIMM + 600 W accelerators
- Inference platforms: dual/single-passthrough affinity schemes
- AI supercomputing scale-out: 6U cell with Switch topology option
Why the Numbers Matter
- MCR DIMM up to 8000 MT/s: the fastest DIMM class in the rack line
- 32 DDR5 slots: depth and speed together in one memory map
- 2x Xeon 6 up to 350 W: latest-generation CPU compute
- 3 GPU schemes: dual/single passthrough and Switch topologies
- 8x 600 W GPUs + 12 hot-swap bays: compute and data in one 6U
What TF NEW Says
"AI nodes starve on memory bandwidth before they starve on compute," said a TF NEW product manager. "The S768G6 takes 32 DDR5 slots to 8000 MT/s with MCR DIMMs behind two Xeon 6 processors, then couples the fabric to eight 600 W GPUs in passthrough or Switch schemes. Six units of rack, engineered around the bandwidth the models actually consume."
Quality and Availability
Every S768G6 is burn-in tested under full GPU load — all PCIe paths, MCR/RDIMM memory channels and drive bays validated for 48 hours before shipping. It ships deployment-ready with GPU scheme and memory configuration documentation.
Pricing and Availability
The S768G6 is available now. Pricing varies with processor, memory, GPU scheme and storage 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
- What is the S768G6? A 6U AI supercomputing server with up to 2x Intel Xeon 6 and up to eight 600 W GPUs.
- What is MCR DIMM? Multiplexer Combined Ranks DDR5 that presents more effective ranks per slot, running up to 8,000 MT/s.
- What GPU schemes are available? 8-card dual-passthrough (standard), single-passthrough (optional) or Switch expansion.
- How much memory does it support? 32 DDR5 slots; MCR DIMM to 8000 MT/s, RDIMM to 6400/5200 MT/s.
- What storage does it have? 12x front hot-swap 3.5"/2.5" bays plus per-scheme U.2 NVMe.