AMD Radeon Instinct MI6 Professional Graphics Card

Product status: Released | Report Error
Overview
Manufacturer
AMD
Original Series
Radeon Instinct
Release Date
June 20th, 2017
PCB Code
109-D12267
Model
AMD D122 SKU 000
Graphics Processing Unit
GPU Model
Polaris 10
Architecture
Polaris GCN 4.0
Fabrication Process
14 nm FinFET
Die Size
232 mm2
Transistors Count
5.7B
Transistors Density
24.6M TRAN/mm2
Stream Processors
2304
TMUs
144
ROPs
32
Clocks
Base Clock
1237 MHz
Boost Clock
1237 MHz
Memory Clock
1750 MHz
Effective Memory Clock
7000 Mbps
Memory Configuration
Memory Size
16384 MB
Memory Type
GDDR5
Memory Bus Width
256-bit
Memory Bandwidth
224 GB/s
VideoCardz.com

Physical
Interface
PCI-Express 3.0 x16
Height
2-slot
Power Connectors
1 × 6-pin
TDP/TBP
150 W
Recommended PSU
450 W
API Support
DirectX
12.0
Vulkan
1.0
OpenGL
4.5
OpenCL
2.1
Shader Model
5.0

Performance
Pixel Fillrate
39.6 GPixels/s
Texture Fillrate
178.1 GTexel/s
SPFP Performance
5.7 TFLOPS
Performance per W
38 GFLOPS/W
Performance per mm2
24.6 GFLOPS/mm2





 ModelClocksMemory
AMD Radeon Instinct MI25
4096 Unified Cores
1501 / 472 MHz 16 GB HBM2
AMD Radeon Instinct MI8
4096 Unified Cores
1000 / 500 MHz 4 GB HBM
AMD Radeon Instinct MI6
2304 Unified Cores
1237 / 1750 MHz 16 GB GDDR5
 ModelClocksMemory
AMD Radeon Instinct MI6
2304 Cores
1237 / 1750 MHz 16 GB GDDR5
AMD Radeon Pro 580
2304 Cores
1193 / 1700 MHz 8 GB GDDR5
AMD Radeon Pro Duo (Polaris)
4608 Cores
1243 / 1750 MHz 32 GB GDDR5
AMD Radeon RX 480
2304 Cores
1120 / 2000 MHz 8 GB GDDR5
AMD Radeon RX 580
2304 Cores
1256 / 2000 MHz 8 GB GDDR5
AMD Radeon RX 580M
2304 Cores
1120 / 2000 MHz 8 GB GDDR5
AMD Radeon Pro 575
2048 Cores
1098 / 1700 MHz 4 GB GDDR5
AMD Radeon Pro WX 7100
2048 Cores
1243 / 1750 MHz 8 GB GDDR5
AMD Radeon RX 470
2048 Cores
926 / 1650 MHz 4 GB GDDR5
AMD Radeon RX 470M
2048 Cores
926 / 1650 MHz 4 GB GDDR5
AMD Radeon RX 570
2048 Cores
1168 / 1750 MHz 4 GB GDDR5
AMD Radeon RX 570M
2048 Cores
926 / 1650 MHz 4 GB GDDR5
AMD Radeon Pro 570
1792 Cores
1004 / 1700 MHz 4 GB GDDR5
AMD Radeon Pro WX 5100
1792 Cores
1000 / 2000 MHz 8 GB GDDR5
AMD Radeon RX 470D
1792 Cores
926 / 1650 MHz 4 GB GDDR5
AMD Radeon RX 570 8GB (V)
2048 Cores
1168 / 1750 MHz 8 GB GDDR5
AMD Radeon RX 580 4GB (V)
2304 Cores
1256 / 1750 MHz 4 GB GDDR5
AMD Radeon RX 580 XTR (V)
2304 Cores
1256 / 2000 MHz 8 GB GDDR5

PERFORMANCE

5.7 TFLOPS of Peak Half or Single Precision Performance in Single-Slot Card Using Under 150 Watts TDP 1

  • 5.7 TFLOPS peak FP16 | FP32 GPU compute performance.

    With 5.7 TFLOPS peak FP16 or FP32 compute performance on a single board with 16GB GDDR5 memory, the Radeon Instinct MI6 server accelerator provides unmatched single-precision performance with large memory on a single-slot card for machine and deep learning inference and edge training applications, along with providing a cost-effective solution for HPC development systems needing more memory. 1

  • 16GB ultra-fast GDDR5 GPU Memory on 256-bit memory interface.

    With 16GB of GDDR5 GPU memory and up to 224GB/s of memory bandwidth, the Radeon Instinct MI6 server accelerator provides a well-balanced, versatile single-precision compute solution for handling demanding machine intelligence and deep learning inference applications, along with providing a cost-effective solution for edge-training applications due to its large memory and low power requirements.

  • Up to 38 GFLOPS per watt of peak FP16 and FP32 GPU compute performance.

    With up to 38 GFLOPS/watt peak FP16 or FP32 GPU compute performance, the Radeon Instinct MI6 server accelerator provides a versatile, efficient, solution for machine intelligence and deep learning inference and edge-training applications. 2

  • 36 Compute Units (2,304 Stream Processors).

    The Radeon Instinct MI6 server accelerator has 36 Compute Units each containing 64 stream processors, for a total of 2,304 stream processors that are available for running many smaller batches of data simultaneously against trained deep learning neural networks to get quick results. Single-precision performance in a low-cost, efficient, solution is crucial to these types of system installations, and MI6 accelerator provides the outstanding single-precision performance in a single-slot GPU card.

FEATURES

Passively Cooled Accelerator for Scalable Server Deployments

  • Passively cooled server accelerator based on “Polaris” architecture.
    The Radeon Instinct MI6 server accelerator, based on the “Polaris” architecture with a 14nm FinFET process and is designed for highly-efficient, scalable server deployments for single-precision inference and edge-training applications in machine intelligence and deep learning, along with HPC general purpose and development systems. This GPU server accelerator provides customers with a cost-effective, versatile, compute solution while consuming only 150W TDP board power.
  • 150W TDP board power, single-slot, 9.5” GPU server card.
    The Radeon Instinct MI6 server GPU card is a full-height, single-slot card and works with PCIe® Gen 3 compliant motherboards. The MI6 GPU card is designed to fit in most standard server designs providing a low-cost, highly-efficient, server solution for heterogeneous machine intelligence and deep learning inference and edge-training, along with HPC-class system deployments.
  • Ultra-fast GDDR5 with up to 224GB/s memory bandwidth.
    The Radeon Instinct MI6 server accelerator is designed with 16GB of ultra-fast GDDR5 memory allowing numerous batches of large data to be quickly handled simultaneously for demanding machine intelligence and deep learning inference and edge-training applications, along with HPC workloads.
  • MxGPU SRIOV HW Virtualization.
    The Radeon Instinct™ MI6 server accelerator is designed with support of AMD’s MxGPU SRIOV hardware virtualization technology to drive greater utilization and capacity in the data center.

USE CASES

Inference for Deep Learning

Today’s exponential data growth and dynamic nature of that data has reshaped the requirements of data center system configurations. Data center designers need to build systems capable of running workloads more complex and parallel in nature, while continuing to improve system efficiencies. Improvements in the capabilities of discrete GPUs and other accelerators over the last decade are providing data center designers with new options to build heterogeneous computing systems that help them meet these new challenges.

Datacenter deployments running inference applications, where lots of new smaller data set inputs are being run at half-precision (FP16) or single-precision (FP32) against trained neural networks to discover new knowledge, require parallel compute capable systems that can quickly run data inputs across lots of smaller cores in a power-efficient manner.

The Radeon Instinct MI6 accelerator is a powerful, cost-sensitive solution for machine intelligent and deep learning inference deployments in the datacenter delivering 5.7 TFLOPS for each of half or single precision floating point performance in a single-slot 150 watt TDP card. 1 The MI6 accelerator, based on AMD’s “Polaris” architecture with 16GB ultra-fast GDDR5 memory and up to 224 GB/s bandwidth, combined with the Radeon Instinct’s open ROCm software platform, provides datacenter designers with a versatile, highly-efficient, solution for inference deployments.

Key Benefits for Inference:

  • 5.7 TFLOPS half or single precision compute performance 1
  • 38 GFLOPS/watt peak FP16|FP32 performance for efficient inference and edge-training deployments 2
  • 358 GFLOPS peak double precision (FP64) compute performance
  • 2.4 GFLOPS/watt peak FP64 performance
  • 16GB GDDR5 on 256-bit memory interface provides ultra-fast memory performance
  • Passively cooled, Single-slot, GPU card for scalable server deployments
  • ROCm software platform provides open source Hyperscale platform
  • Open source Linux® drivers, HCC compiler, tools and libraries for full control from the metal forward
  • Optimized MIOpen deep learning framework libraries
  • Large BAR Support for mGPU peer to peer
  • MxGPU SR-IOV hardware virtualization for optimized system utilizations

 

Edge-training for Deep Learning

Today’s exponential data growth and dynamic nature of that data has reshaped the requirements of data center system configurations. Data center designers need to build systems capable of running workloads more complex and parallel in nature, while continuing to improve system efficiencies. Improvements in the capabilities of discrete GPUs and other accelerators over the last decade are providing data center designers with new options to build heterogeneous computing systems that help them meet these new challenges.

Data centers running machine intelligence and deep learning applications using edge-training deployments, where the goals are to focus on more cost-effective, efficient compute systems for training tasks using a very large number of lower-cost, edge-servers  to process less compute-intensive training tasks helping lower overall data center costs by achieving higher efficiencies. These systems require accelerators providing good single-precision performance with larger memory in a dense, low-power package.

The Radeon Instinct MI6 accelerator is a versatile, low power server accelerator that is a perfect fit for the requirements of low-cost edge-training deployments for machine intelligence and deep learning applications in the data center delivering 38 GFLOPS/watt peak half precision (FP16) or single precision (FP32) floating point performance in a single-slot 150 watt TDP GPU card. 2 The Radeon Instinct MI6 accelerator, based on AMD’s “Polaris” architecture with 16GB ultra-fast GDDR5 memory and up to 224 GB/s bandwidth, combined with the Radeon Instinct’s open ecosystem approach with the ROCm software platform, provides data center designers with a versatile, highly-efficient, solution for edge-training deployments.

Key benefits for Edge-training:

  • 5.7 TFLOPS peak half or single precision compute performance 1
  • 38 GFLOPS/watt peak FP16|FP32 performance in single-slot card 2
  • 358 GFLOPS peak double precision (FP64) compute performance
  • 2.4 GFLOPS/watt peak FP64 performance
  • 16GB GDDR5 on 256-bit memory interface provides ultra-fast memory performance
  • Passively cooled for scalable server deployments
  • ROCm software platform provides open source Hyperscale platform
  • Open source Linux® drivers, HCC compiler, tools and libraries for full control from the metal forward
  • Optimized MIOpen deep learning framework libraries
  • Large BAR Support for mGPU peer to peer
  • MxGPU SR-IOV hardware virtualization for optimized system utilization
  • Open industry standard support of multiple architectures and industry standard interconnect technologies 3

 

Heterogeneous Compute for HPC General Purpose and Development

The HPC industry is creating immense amounts of unstructured data each year and a portion of HPC system configurations are being reshaped to enable the community in extracting useful information from that data. Traditionally, these systems were predominantly CPU based, but with the explosive growth in the amount and different types of data being created, along with the evolution of more complex codes, these traditional systems don’t meet all the requirements of today’s data intensive HPC workloads. As these types of codes have become more complex and parallel, there has been a growing use of heterogeneous computing systems with different mixes of accelerators including discrete GPUs and FPGAs. The advancements of GPU capabilities over the last decade has allowed them to be used for a growing number of these mix precision parallel codes like the ones being used for deep learning applications. Scientists and researchers across the globe are now using accelerators to more efficiently process HPC parallel codes across several industries including life sciences, energy, financial, automotive and aerospace, academics, government and defense.

The Radeon Instinct MI6 accelerator, combined with AMD’s revolutionary ROCm open software platform, is a versatile, efficient heterogeneous compute solution delivering 5.7 TFLOPS for each of peak half or single precision performance in a single-slot, 150 watt TDP GPU card with 16GB of ultra-fast GDDR5 memory and up to 224 GB/s of memory bandwidth. 1 The Radeon Instinct MI6 accelerator is an ideal heterogeneous compute solution for cost-sensitive general purpose and development systems being deployed in the Financial Services, Energy, Life Science, Automotive, Academic (Research & Teaching), Government Labs and other HPC industries.