Team red is taking the Radeon Pro Duo to the data center with its FirePro S9300 X2. This card combines two fully-enabled Fiji packages, each with 4GB of non-ECC HBM RAM.
The card is produced in in a server-friendly form factor. Each GPU runs at 850 MHz good for 13.9 TFLOPS of theoretical FP32 compute performance in a 300W power envelope. The new AMD FirePro™ S9300 x2 Server GPU is the world’s first professional GPU accelerator to be equipped with high bandwidth memory (HBM) and the first accelerator compatible with all AMD’s GPUOpen Professional Compute tools and libraries.
HBM allows the AMD FirePro S9300 x2 Server GPU to exceed the competition with 3.5x the memory bandwidth of NVIDIA’s Tesla M40 and 2.1x the memory bandwidth of NVIDIA’s Tesla K803.
Cooling/Power/Form Factor
- Max Power: 300W
- Bus Interface: PCIe® Gen 3 x16
- Form Factor: Dual Slot, Full Length, Full Height
- Cooling: Passive
Memory
- Size/Type: 8GB HBM
- Bandwidth: 1TB/s (2x 512GB/s)
API and OS Support
- OpenCL™ 1.2
- HC (Heterogeneous Compute)
- C++ AMP
- Linux® 64-bit
Enabled AMD Technologies
- AMD PowerTune technology5
System Requirements
- PCI Express® based server with one available x16 lane slot. AMD recommends PCI Express® v3.0 for optimal performance
- Power supply with two PCIe 8-pin aux power connectors
- Airflow through GPU of at least 25CFM, max inlet temperature 45C
- Minimum 16GB DDR3/DDR4 system memory recommended
Based on third generation AMD Graphics Core Next (GCN) architecture, the AMD FirePro S9300 x2 server GPU delivers up to 13.9 TFLOPS of peak single-precision floating point performance – more than any other GPU accelerator available on the market today for single-precision compute4. Compared to Intel’s flagship Xeon E5 CPU, the raw performance advantage of the FirePro™ S9300 x2 GPU is even more dramatic – over 15X the memory bandwidth and over 12X the peak single precision performance6.
AMD FirePro S Series | ||||||
FirePro S9300 X2 | FirePro S9170 | FirePro S9150 | FirePro S9000 | |||
Stream Processors | 2 x 4096 | 2816 | 2816 | 1792 | ||
Boost Clock | 850MHz | 930MHz | 900MHz | 900MHz | ||
Memory Clock | 1Gbps HBM | 5Gbps GDDR5 | 5Gbps GDDR5 | 5.5Gbps GDDR5 | ||
Memory Bus Width | 2 x 4096-bit | 512-bit | 512-bit | 384-bit | ||
VRAM | 2 x 4GB | 32GB | 16GB | 6GB | ||
FP32 | 13.9 TFLOPs | 5.2 TFLOPs | 5.1 TFLOPs | 3.2 TFLOPs | ||
FP64 | 0.8 TFLOPs (1/16) |
2.6 TFLOPs (1/2) |
2.5 TFLOPs (1/2) |
0.8 TFLOPs (1/4) |
||
Transistor Count | 2 x 8.9B | 6.2B | 6.2B | 4.31B | ||
TDP | 300W | 275W | 235W | 225W | ||
Cooling | Passive | Passive | Passive | Passive | ||
Target Market | HPC | HPC | HPC | HPC + VDI | ||
Manufacturing Process | TSMC 28nm | TSMC 28nm | TSMC 28nm | TSMC 28nm | ||
Architecture | GCN 1.2 | GCN 1.1 | GCN 1.1 | GCN 1.0 | ||
GPU | Fiji | Hawaii | Hawaii | Tahiti | ||
Launch Date | Q2 2016 | 07/2015 | 08/2014 | 08/2012 | ||
Launch Price | $5999 | $3999 | - | - |
A great accelerator is not complete without having a great developer ecosystem. With AMD’s GPUOpen Professional Compute software stack, the AMD FirePro S9300 x2 Server GPU utilizes AMD’s first open source Linux® driver built specifically for compute, as well as support for acceleration using C++ in addition to OpenCL™. Another benefit for those who have code in CUDA, is the ability to easily port the majority of their code over to C++, giving companies the freedom to choose between vendors.