GPU server
4U4G-TURIN/HPR
A 4U single-socket SP5 platform for up to four dual-slot PCIe GPUs, designed for AI inference, rendering, simulation and mixed accelerator workloads that value PCIe flexibility.
Observed barebone market range
$5,000–$8,000 USD
The wider band reflects regional channel, tax and availability differences.
This is third-party market context for the barebone manufacturer part number, not a TORmem selling price or binding quote. It normally excludes CPUs, memory, GPUs, storage, add-in networking, freight, taxes, integration and support. TORmem systems are configured and quoted to order.
Availability & lead time
- Production phase
- Complete DVT (design validation test)
- Standard lead time
- 16-20 wks
- Forecast lead time
- 8–12 week or faster
- Minimum order
- Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.
Specifications
- 4U rackmount chassis with 2+1 redundant 2700W 80 PLUS Titanium CRPS
- Single Socket SP5 (LGA 6096P), supporting AMD EPYC 9005/9004 and 97x4 series processors up to the validated platform TDP
- 8 DDR5 DIMM slots (1DPC), supporting RDIMM and RDIMM-3DS up to the validated capacity and speed
- Up to 4 dual-slot or 7 single-slot FHFL PCIe 5.0 x16 accelerator positions
- 1 additional FHFL single-slot PCIe 5.0 x8 position
- 2 M.2 PCIe 5.0 x4 positions with support for 22110/2280 form factors
- 2 RJ45 10GbE ports via Broadcom BCM57416
- Dedicated management with ASPEED AST2600 BMC and IPMI
- 4 fixed 80 mm middle fans plus 2 fixed 80 mm rear GPU fans
- Chassis dimensions: 710 × 430.4 × 174.8 mm with GPU fans
- TPM header and remote platform monitoring
- GPU model, power limit, cable set and ambient-temperature envelope are validated per configuration
Specification reference: ASRock Rack product specifications
Ready for testing; availability is verified in real time.
Configure & request a quote
Compose quote requestOpens your mail client with the configuration filled in. Nothing is sent or stored by this site.
Also available on Pantheon for GPUs, turbines, networking and memory.
How this is priced
TORmem quotes 4U4G-TURIN/HPR at three integration levels. What changes between them is how much of the build TORmem does before it ships.
What changes the configured price
- CPU model and socket count
- DDR5 capacity, speed and population
- GPU or accelerator model, quantity and power envelope
- NVMe/SAS/SATA media and controllers
- NIC, DPU, optics and cabling
- Burn-in, firmware baseline, warranty and deployment support
- Quantity, freight, destination and required delivery date
What the market range means
TORmem researches comparable barebone listings to give buyers a current procurement baseline. The final quote is for the actual CPU, memory, accelerator, storage, network, validation, support and delivery scope—not a generic chassis listing.
| Level | What's included | How it's quoted |
|---|---|---|
| L6Barebone | Chassis, motherboard, backplane, power supplies, fans. No CPU, memory, GPU or drives. | Quoted per unit, independent of configuration. |
| L10Configured node | Everything in L6, plus CPU, DDR5, GPUs, NVMe, network adapters — assembled, firmware-matched and burn-in tested. | Quoted from your configuration. Available now. |
| L11Rack-integrated | Everything in L10, plus rack, PDUs, switching, optics, cabling, cluster-level validation and on-site delivery. | Quoted per rack. |
Barebone cost Available soon; full L10 pricing available now with customer configuration.
Minimum order Standard Model: No minimum order quantity applies. Availability will be confirmed based on real-time stock status.
Best fit
- Four-GPU inference replicas, rendering, simulation and visualization
- Mixed PCIe accelerators where flexibility matters more than a fixed scale-up fabric
- Rack deployments needing integrated 10GbE management/data connectivity
Specify before ordering
- Not an HGX/NVSwitch platform; tightly coupled training workloads should be evaluated separately
- GPU TDP, cabling, airflow and facility ambient conditions must be validated as a system
- The observed market band is for the barebone platform, not a configured four-GPU server
Compare the range
Other gpu server platforms TORmem builds on the same generation.
| Model | Configuration | Production phase | Std lead time | Market reference |
|---|---|---|---|---|
| 1U1G2E-TURIN | 1-GPU Barebone | Complete DVT (design validation test) | 16-20 wks | $2,400–$2,700 |
| 2U2G-TURIN/HPR | 2-GPU Barebone | Mass production | 16-20 wks | $2,800–$3,300 |
| 4U4G-TURIN/HPRthis page | 4-GPU Barebone | Complete DVT (design validation test) | 16-20 wks | $5,000–$8,000 |
| 4U8G-TURIN2/RF+ | 8-GPU Barebone | Complete DVT (design validation test) | 16-20 wks | $7,000–$7,600 |