This email address is being protected from spambots. You need JavaScript enabled to view it.
This email address is being protected from spambots. You need JavaScript enabled to view it.
AFTM GPU CLUSTER in-a-box
Powerful enough to do up to 70 trillion calculations per second. Yet as simple as a workstation.
In Detail
AFTM GPU CLUSTER in-a-box is ready-to-run complete solution designed for small and mid-size research groups in University Research Institutes & Centres and commercial compute-intensive application users.
Are you running simulations or computations?
AFTM GPU CLUSTER in-a-box is immediately intuitive to use even for people without Linux experience. View, drill down, point, click, and move on, even for multiple clusters at once.
With added GPU Software Porting Services (SPS), and a single point of contact for software, hardware and OS support, we make it worry-free to migrate to the GPU platform and run your applications.
Based on the latest "Fermi" NVIDA Tesla 20-series GPUs, the AFTM GPU Cluster offers a ground-breaking 67 TERAFLOPS of double precision performance.
AFTM GPU CLUSTER in-a-box supports NVIDIA GPU Direct T. GPUDirect enables significant application speed-up (up to 30%) by allowing network adapters, solid-state drives (SSDs) and other devices to directly access CUDA memory, eliminating unnecessary system memory copies and CPU overhead.
Are you administrating the Cluster?
The name says it all! The Admin Friendly GPU Cluster was designed to make cluster installation, management and monitoring easy. We have taken the complexity out of the cluster right from the start, allowing you to configure and use it like a simple workstation.
Exclusive benefits:
- Turnkey pre-installed complete solution
- Easy to use, intuitive cluster management system
- GPU Software Porting Services (SPS)
- Support across the Asia-Pacific region
- Single point of contact for support for hardware, software, OS and services
- Scalable from Base to Quarter, Half and Full rack configurations
- Lowest (price + TCO)/performance ratio in APAC
| Base | Quarter | Half | Full | |
|---|---|---|---|---|
|
|
|
|
|
|
|
CPU Cores CPUs Nodes |
48 8 4 |
96 16 8 |
192 32 16 |
408 68 34 (max 39U) |
|
GPU Cores: 2075 series 2090 series |
5,376 6,144 |
10,752 12,288 |
21,504 24,576 |
45,696 52,224 |
|
Total GPUs |
12 |
24 |
48 |
102 |
|
Memory per node: |
32 GB |
32 GB |
32 GB |
32 GB |
|
Total Memory: |
128 GB |
256 GB |
512 GB |
1,088 GB |
|
Peak Theoretical System Performance: Intel E5-2620 GPU Double precision (2090 series) GPU Single precision (2090 series) |
0.768 Tflops 7.98 Tflops 15.972 Tflops |
1.536 Tflops 15.96 Tflops 31.944 Tflops |
3.072 Tflops 31.92 Tflops 63.888 Tflops |
6.528 Tflops 67.83 Tflops 135.762 Tflops |
|
1. Peak Power Consumption: 2. Heat dissipation: |
3,800W 3,950W |
7,600W 7,900W |
14,000W 14,600W |
32,200W 33,500W |
Tech Specs
Architecture
- Separate data, monitoring and management networks
- Platinum-level high-efficiency power supplies
- Hot-swap HDDs and power supplies
- Optimized cooling
- Low power consumption
Capacity
- Up to 32 CPU/GPU nodes
- Configurable 1:1 or 1:2 CPU/GPU ratio
- Dedicated PCI-E 3.0 x16 slot for each GPU
- 42U rack space
Compute Field
- GPU: NVIDIA Tesla 2075/2090 series
- CPU: Dual Intel Xeon E5-2620 series processors per compute node (6-cores) or higher
- Memory: 32 GB DDR3-1600 Low-voltage ECC Registered per compute node (Intel) or higher
- OS: Microsoft® Windows HPC Server 2008 or Linux OS
- Storage (optional): 500 GB per node
Master node/Storage/License server
- CPU: Dual Intel Xeon E5-2620 series processors (6-cores) or higher
- Memory: 16 GB DDR3-1600 Low-voltage ECC Registered per compute node (Intel) or higher
- Storage: 3*3TB RAID5 (maximum 12x3.5” HDDs), Optional RAID BBU
- SAS/SATA Controller 6Gbps, RAID 0,1,10,5,50,6,60 support with Battery Backup Unit
- OS: Microsoft® Windows HPC Server 2008 + Windows HPC Pack 2008 or Linux OS +CMS
- Cluster Management: Admin Friendly immediately intuitive to use complete management solution
Networking/Interconnect
- 1GbE switch
- 40Gb/s (QDR) or 56Gb/s (FDR) InfiniBand switch
Other equipment and accessories
- PDU, Ethernet and power cables
- UPS (optional)
- Rack (48U, 42U, 24U)
Size and weight
Power System
- Auto 400V, 47-63 Hz 3-phase input power
- Power consumption: 30,200W
- Heat dissipation: 33,500W
Warranty
- 3 Years
Support – SW Porting Services
- Application migration package: 1 hr; 10 hrs
- HW and OS support package: 1 hr; 10 hrs
- Software porting support package: upon request
System Scalability
Scalable from Base to Quarter to Half to Full Rack and beyond