AMD
GPUDatacenterReleased Mar 31, 2016

AMD FirePro S9300 x2

8 GB HBM · 8192-bit · 8,192 cores

The AMD FirePro S9300 x2 puts two Fiji GPUs on one board, each with 64 GCN 3 compute units at 850 MHz and its own 4 GB of HBM on a 4096-bit bus, for 8 GB and roughly 1 TB/s of bandwidth in total. AMD aimed the passively cooled card at single-precision compute such as seismic processing and deep learning rather than at attached displays, of which it has none, and it added native half-precision (FP16) support. Its memory is not ECC protected.

General

Manufacturer
AMD
Model
FirePro S9300 x2
Segment
Datacenter
GPU chip
2x Capsaicin XT (Fiji)
Architecture
GCN 3
Release date
Mar 31, 2016

Memory

Memory size
8 GB
Memory type
HBM
Bus width
8192-bit
Bandwidth
1,024 GB/s
Memory clock
500 MHz

Clocks & silicon

GPU clock
850 MHz
Boost clock
—
Shader units
8,192 Stream Processors (2x 64 CU)
Tensor / AI cores
—
RT cores
—
Transistors
17.8 billion
Technology (nm)
28 nm
TSMC 28 nm
TDP / TBP
300 W

Interface & APIs

Bus interface
PCIe 3.0 x16
SLI / multi-GPU
Dual GPU (2x)
DirectX
12 (12_0)
OpenGL
4.6
Vulkan
1.1
Other APIs
OpenCL 2.0

AI support

Precisions
FP8FP16BF16INT8INT4TF32FP64
AI accelerator
None (no matrix units)
Peak AI TOPS
—
Notes
Peak throughput across both GPUs is 13.9 TFLOPS FP32 and 868 GFLOPS FP64, a 1:16 double-precision rate. Half-precision (FP16) arithmetic is supported natively at the single-precision rate; there are no matrix units. Work must be split across the two GPUs explicitly, as each addresses only its own 4 GB.
All AMD GPUs →

Datacenter gpus from AMD announced around the same time as the AMD FirePro S9300 x2. Add any of them to your compare list to see the specs side by side.

AMDgpu

AMD FirePro S7150

8 GB GDDR5 · 256-bit · 1,792 cores

Released
Feb 1, 2016
AMDgpu

AMD FirePro S9170

32 GB GDDR5 · 512-bit · 2,816 cores

Released
Jul 8, 2015