PCIe 4 or 5 Switch, 8 or 10 GPUs? TFNEW S74AG5
TL;DR: TFNEW launched the S74AG5, an enterprise GPU server, on September 21, 2026. A 4U/5U dual-socket AI server with 2x 4th/5th Gen Intel Xeon Scalable (350 W), 32 DDR5 DIMM slots (up to 8 TB), and GPU schemes from 8-card passthrough to 10-card PCIe Switch on your choice of PCIe 4.0 or PCIe 5.0 fabric. Get specs and pricing: sales@tfnew.com.
Core Specifications (with Plain-Language Terms)
The S74AG5 is a 4U/5U dual-socket AI server whose GPU scheme is a menu, not a fixed configuration: 8-card passthrough (one GPU per PCIe path for frameworks that prefer direct topology), or 8/10-card PCIe Switch — and the switch fabric itself is selectable at PCIe 4.0 or PCIe 5.0. A PCIe Switch is a fabric chip that fans one root complex's lanes out to many GPUs, trading some per-GPU lanes for more accelerators per node; choosing 4.0 vs 5.0 sets the per-card bandwidth for that fabric. Compute: 2x 4th/5th Gen Intel Xeon Scalable at 350 W TDP on an Intel C741 chipset, with 32 DDR5 DIMM slots (ECC RDIMM/LRDIMM/3DS RDIMM) up to 5600 MT/s and 8 TB per system. Storage: standard 12x 3.5/2.5-inch SATA/SAS drives; passthrough scheme adds 2x U.2 NVMe (up to 4). One chassis, three topologies, two fabrics.
Specs at a Glance
| Item | S74AG5 Value |
|---|---|
| Chassis | 4U/5U dual-socket AI server |
| GPU scheme | 8-card passthrough / 8- or 10-card PCIe Switch (PCIe 4.0 or 5.0) |
| Processor | 2x 4th/5th Gen Intel Xeon Scalable, 350 W, Intel C741 |
| Memory | 32 DDR5, ECC RDIMM/LRDIMM/3DS RDIMM, up to 5600 MT/s, up to 8 TB |
| Storage | 12x 3.5/2.5" SATA/SAS; passthrough: 2x U.2 NVMe (up to 4) |
Why a Selectable Fabric Is the Flexible AI Platform
AI sites rarely know in year one whether their workload wants 8 direct GPUs or 10 on a fabric — and PCIe 5.0 didn't exist when many existing fleets were speced. The S74AG5 removes the guesswork: choose passthrough for framework-friendly direct topology, or Switch for density, and pick the fabric generation at order time. Ten GPUs on a PCIe 4.0 switch keeps mature, well-known ecosystems; eight or ten on PCIe 5.0 doubles the per-card bandwidth budget. Both sit behind the same 350 W dual Xeon, 8 TB DDR5 and 12-bay storage plan — so the platform outlives the topology decision.
Where It Fits
- Flexible AI platforms: topology and fabric chosen per site
- Dense GPU nodes: 10-card PCIe Switch for accelerator-heavy racks
- Framework-optimized training: 8-card passthrough for direct topology
- Fleet modernization: PCIe 4.0 for mature stacks, 5.0 for bandwidth
Why the Numbers Matter
- 8 or 10 GPUs: passthrough or Switch density from one chassis
- PCIe 4.0 or 5.0 fabric: per-card bandwidth chosen at order time
- 2x Xeon 350 W + C741: dual-socket compute behind the GPU scheme
- 32 DDR5 up to 8 TB: ECC RDIMM/LRDIMM/3DS memory depth
- 12x drives + U.2 NVMe: local data beside the accelerators
What TF NEW Says
"An AI platform shouldn't force a topology decision you can't take back," said a TF NEW product manager. "The S74AG5 ships eight-GPU passthrough or eight- or ten-GPU Switch, on PCIe 4.0 or 5.0 fabric, from the same dual-Xeon, 8 TB DDR5, 12-bay chassis. Pick the topology and fabric your workload actually wants — the platform doesn't change."
Quality and Availability
Every S74AG5 is burn-in tested under full GPU load — passthrough and switch paths, memory channels and drive bays validated for 48 hours before shipping. It ships deployment-ready with topology and fabric configuration documentation.
Pricing and Availability
The S74AG5 is available now. Pricing varies with processor, memory, GPU scheme and fabric generation. 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 S74AG5? A 4U/5U dual-socket AI server with 2x 4th/5th Gen Xeon Scalable and GPU schemes from 8-card passthrough to 10-card Switch.
- What GPU schemes are available? 8-card passthrough, or 8/10-card PCIe Switch on PCIe 4.0 or PCIe 5.0 fabric.
- How much memory does it support? 32 DDR5 slots up to 5600 MT/s (ECC RDIMM/LRDIMM/3DS RDIMM), up to 8 TB.
- What storage does it have? Standard 12x 3.5/2.5-inch SATA/SAS bays plus U.2 NVMe in the passthrough scheme.
- Why choose PCIe 4.0 or 5.0? The fabric generation sets per-card bandwidth — 4.0 for mature ecosystems, 5.0 for maximum bandwidth.