- Artificial Intelligence
- Big Data Analytics
- High-performance Computing
- Research Lab/National Lab
- Astrophysics, Business Intelligence
- Dual Socket SP3 support AMD EPYC 7002 series Processors
- Up to 8TB Registered ECC DDR4 3200MHz SDRAM in 32 DIMMs
- 9 PCI-E 4.0 x16 (Option: 10 PCI-E 4.0 x16 slots without NVMe devices)
- 2 GbE LAN ports
- Up to 24 Hot-swap 2.5" drive bays; 4x 2.5" SATA drives and 4x 2.5" NVMe drives supported with included H/W
- 8 Hot-swap 11.5K RPM cooling fans
- 2000W Redundant (2+2) Power Supplies; Titanium Level (96%+)
- 2 x RJ45
- 1 x RJ45 IPMI_dedicated
- 2 x USB 3.0
- 1 x VGA
Dual CPU server board.
- AMD EPYC 7002 series Processors (Socket SP3).
- System on Chip (SoC).
Dual AMD EPYC 7002-Series processors with up to 64 Cores (per CPU)
- Up to 1536 GB DDR4 2666 MHz ECC Buffered
- 24 RAM slots.
Note: actual RAM clock speed depends on the CPU specifications and may vary.
- 9 PCI-E 4.0 x16 (FHFL) slots
- or 10 PCI-E 4.0 x16 (FHFL) slots without NVMe devices
- 4 x SATA3 6Gb/s port(s)
- Support Raid 0, 1, 10
- 4 x NVMe (PCI-E 4.0 x4) ports via SlimSAS
Up to 10 Nvidia Quadro/RTX/Titan graphics cards with NVLink support.
- Nvidia RTX 2080Ti 11Gb (3 DP, HDMI, USB Type-C)
- Nvidia RTX Titan 24Gb (3 DP, HDMI, USB Type-C)
- Nvidia RTX 3080 10 Gb (3 DP, HDMI)
- Nvidia RTX 3090 24 Gb (3 DP, HDMI)
- Nvidia Quadro RTX 5000 16Gb (4 DP, VirtualLink)
- Nvidia Quadro RTX 6000 24Gb (4 DP, VirtualLink)
- Nvidia Quadro RTX 8000 48Gb (4 DP, VirtualLink)
Maximum Digital Resolution: 7680x4320 @60Hz
Multi Monitor Support: 4
- Dual 1 Gigabit Ethernet
- Dedicated IPMI (ATEN)
- 100 Gbps Dual Port InfiniBand EDR (Mellanox ConnectX VPI Adapter; QSFP28) (optional)
Case Drive Bays:
- Up to 24 x 2.5" SAS/SATA drive bays
- 4 x 2.5" SATA supported natively
- 4 x 2.5" NVMe supported natively
Custom liquid cooling system (including loop for 2 x CPUs and 8-10 x GPUs). System consist from server itself and cooling unit for better stability and cooling performance.
- 2 x CPU water block
- 8-10 x GPU water blocks
- External Cooling unit (radiator, pump, reservoir)
- Liquid cooling loop design may vary from the pictures shown on the product page, but will be of equal or better specs than what's specified above.
2000W Redundant (2+2) Power Supplies; Titanium Level (96%+)
Note: ZX9000 comes with 2 x 2000W Redundant (2+2) power supplies. Total 4 power supplies (2 main + 2 backups).
That means that maximum output power is 4000 W (2 x 2000W) with 4000W (2 x 2000W) for backup power supplies.
If one or both power supplies will fail, the other backup power supply starts running to keep the computer powered up.
The switch between the power supplies is seamless, so as not to interrupt the normal use of the computer.
Power consumption examples:
1. ZX9000 with 8 x NVIDIA RTX 2080 Ti + 2 x AMD EPYC 7252 CPUs = 2400W-2600W (100% load for all the GPUs and the CPUs at the same time).
BTUs (per hour): 8188.8
2. ZX9000 with 8 x NVIDIA TITAN RTX + 2 x AMD EPYC 7402 CPUs = 2600W-2800W (100% load for all the GPUs and the CPUs at the same time).
BTUs (per hour): 9212.4
100-120V (standard US outlet):
If you have a standard 20 A / 120V (2400Watts max) US outlet, you can connect two power supplies to one dedicated power circuit (each outlet should have its dedicated circuit). In total, you need two (2) 20 A /120V outlets with a dedicated circuit. This is true for the configurations 1 and 2 mentioned above.
Actual power consumption is not the same as maximum power supply power. Power supply power means how much watts you can get from the power supply and your actual power consumption should not be higher than max power supply power.
Actual power consumption depends on your configuration (graphics card and CPU).
Power supplies efficiency:
1000W @ 110-120V
2000W @ 220-240V
Information for reference only. Consult your electrician for more details.
Noise level and temperature
- Idle noise level: up to 54 db
- Idle temperature (GPUs): Average 33 C / 91.4 F (min 31C - max 36C)
- Max load noise level (stress test): up to 59 db
- Max load temperature (GPUs): Average 59 C / 138.2 F (min 54C - max 65C
8 x NVIDIA TITAN RTX. Noise level measured from 6 ft / 1.8 m distance. Environment: 24 C / 75 F, 38 db. Noise meter directed to the front panel of the server. Temps measured after 30 min test. 120V power source. Luxmark 3.0. Ubuntu 18.04. Drivers 418. CUDA 10.0
Dimensions & Weight
- Rackmountable; 4U; horizontal placement
- Height: 7.0 inches (17.8 cm)
- Width: 17.2 inches (43.7 cm)
- Depth: 29.0 inches (73.7 cm)
- Weight (8 GPU; with coolant): 86 lbs (39.1 kg)
- Rackmountable; 3U; horizontal or vertical placement
- Height: 4.1 inches (10.3 cm)
- Width: 17.3 inches (44.9 cm)
- Depth: 22.4 inches (57.7 cm)
- Weight (with coolant): 27 lbs (12.24 kg)
Shipping Package 1 (server)
- Height: 14 inches (35.56 cm)
- Width: 28 inches (71.12 cm)
- Length: 42 inches (101.6 cm)
- Weight: 120.0 lbs (54.43 kg)
Shipping Package 2 (cooling unit)
- Height: 11 inches (27.94 cm)
- Width: 22 inches (55.88 cm)
- Length: 27 inches (68.58 cm)
- Weight: 30 lbs (13.6 kg)
Shipped in two boxes on the pallet. Shipped in two boxes.
Electrical and Operating Requirements
- Line voltage: 100–240V AC
- Frequency: 50Hz to 60Hz, single phase
- Operating temperature: 50° to 95° F (10° to 35° C)
- Operating Relative Humidity: 8% to 90% (non-condensing)
In the Box
- Bizon ZX9000.
- 4 x power cords (USA).
- User manual.
- Benchmark report.
- BIZON Z–Stack (Ubuntu 18.04 with preinstalled deep learning frameworks).
- BIZON Z–Stack (Ubuntu 20.04 with preinstalled deep learning frameworks).
- CentOS (with preinstalled deep learning frameworks) (optional).
- Windows 10 Pro (optional).
- Microsoft Windows Server 2019 Standard (optional).
All specifications are subject to change without notice. The entire materials provided herein are for reference only. Weight varies by configuration and manufacturing process. Advertised performance is based on maximum theoretical interface values from respective Chipset vendors or organization who defined the interface specification. Actual performance may vary by system configuration. Pictures shown for the options on the configuration page are generic and for reference only. In case of using NVLINKs there is no top lid.