Radeon Instinct

AMD Radeon Instinct
Design firm Advanced Micro Devices
Type Servers

Radeon Instinct is AMD's brand of deep learning oriented GPUs.[1][2] It replaced AMD's FirePro S brand in 2016. Compared to the Radeon brand of mainstream consumer/gamer products, the Radeon Instinct branded products are intended to accelerate deep learning, artificial neural network, and high-performance computing/GPGPU applications.

The Radeon Instinct product line directly competes with Nvidia's Tesla lines of deep learning and GPGPU cards.

Products

The three initial Radeon Instinct products were announced in December 2016, with each based on a different architecture.

MI6

The MI6 is a passively cooled, Polaris 10 based card with 16 GB of GDDR5 memory and with a <150 W TDP.[1][2] At 5.7 TFLOPS (FP16 and FP32), the MI6 is expected to be used primarily for inference, rather than neural network training. The MI6 has a peak double precision (FP64) compute performance of 358 GFLOPS.[3]

MI8

The MI8 is a Fiji based card, analogous to the R9 Nano, and expected to have a <175W TDP.[1] The MI8 has 4 GB of High Bandwidth Memory. At 8.2 TFLOPS (FP16 and FP32), the MI8 is marked toward inference. The MI8 has a peak (FP64) double precision compute performance 512 GFLOPS.[4]

MI25

The MI25 is a Vega based card, utilizing HBM2 memory. The MI25 performance is expected to be 12.3 TFLOPS using FP32 numbers. In contrast to the MI6 and MI8, the MI25 is able to increase performance when using lower precision numbers, and accordingly is expected to reach 24.6 TFLOPS when using FP16 numbers. The MI25 is rated at <300W TDP with passive cooling. The MI25 also provides 768 GFLOPS peak double precision (FP64) at 1/16th rate.[5]

Software

MxGPU

The MI6, MI8, and MI25 products all support AMD's MxGPU virtualization technology, enabling sharing of GPU resources across multiple users.[1][6]

MIOpen

MIOpen is AMD's deep learning library to enable GPU acceleration of deep learning.[1] Much of this extends the GPUOpen's Boltzmann Initiative software.[6] This is intended to compete with the deep learning portions of Nvidia's CUDA library. It supports the deep learning frameworks: Theano, Caffe, TensorFlow, MXNet, The Microsoft Cognitive Toolkit, Torch, and Chainer. Programming is supported in OpenCL and Python, in addition to supporting the compilation of CUDA through AMD's Heterogeneous-compute Interface for Portability and Heterogeneous Compute Compiler.

Chipset Table

Model Launch Code Name Archi-
tecture		 Fab (nm) Transistors (Billion) Die Size (mm2) Bus interface Clock rate Core config[lower-alpha 1] Fillrate Memory Processing Power
(GFLOPS)		 TDP (Watts) API support (version)
Core (MHz) Boost (MHz) Memory (MT/s) Pixel (GP/s)
(Boost)[lower-alpha 2]		 Texture (GT/s)
(Boost)[lower-alpha 3]		 Size (GiB) Bus width (bit) Bus type Bandwidth (GB/s) Half Precision
(Boost)		 Single Precision
(Boost)[lower-alpha 4]		 Double Precision OpenCL
Radeon Instinct MI6 [1][7][6][8][9] 2016-12-12 Polaris 10 GCN 4th gen 14 5.7 232 PCIe 3.0 x16 1120 1233 1750 7000 eff 2304:144:32:36 39.46 177.6 16 256 GDDR5 224 5800 5800 358 150 2.0
Radeon Instinct MI8 [1][7][6][10][11] Fiji XT GCN 3rd gen 28 8.9 596 1000 1000 1000 4096:256:64:64 64.0 256.0 4 4096 HBM 512 8200 8200 512 175 2.0
Radeon Instinct MI25 [1][7][6][12][13][14] Vega 10 XT GCN 5th gen 14 12.5 510 1400 1500 1704 4096:256:64:64 96.0 384 16 2048 HBM2 436.2 24600 12300 768 300 2.0
Radeon Instinct MI25 mxgpu (Prototype, near equal Radeon Pro V340 mxgpu)[15] 2017-06-17 Vega 10 XT GCN 5th gen 14 2x 12.5 2x 510 1400 ? 1500 ? 1704 2x 4096:256:64:64 2x 96.0 ? 2x 384 ? 16 2048 HBM2 2x 436.2 2x 24600 ? 2x 12300 ? 2x 768 ? 300 2.0
  1. ↑ Single-precision shader processors : Texture Mapping Units : Render Output Units (Compute Units)
  2. ↑ Pixel fillrate is calculated as the number of ROPs multiplied by the base (or boost) core clock speed.
  3. ↑ Texture fillrate is calculated as the number of TMUs multiplied by the base (or boost) core clock speed.
  4. ↑ Single precision performance is calculated from the base (or boost) core clock speed based on a FMA operation.

See also

References

  1. 1 2 3 4 5 6 7 8 Smith, Ryan (12 December 2016). "AMD Announces Radeon Instinct: GPU Accelerators for Deep Learning, Coming in 2017". Anandtech. Retrieved 12 December 2016.
  2. 1 2 Shrout, Ryan (12 December 2016). "Radeon Instinct Machine Learning GPUs include Vega, Preview Performance". PC Per. Retrieved 12 December 2016.
  3. ↑ "Radeon Instinct MI6". Radeon Instinct. AMD. Retrieved 22 June 2017.
  4. ↑ "Radeon Instinct MI8". Radeon Instinct. AMD. Retrieved 22 June 2017.
  5. ↑ "Radeon Instinct MI25". Radeon Instinct. AMD. Retrieved 22 June 2017.
  6. 1 2 3 4 5 Kampman, Jeff (12 December 2016). "AMD opens up machine learning with Radeon Instinct". TechReport. Retrieved 12 December 2016.
  7. 1 2 3 Shrout, Ryan (12 December 2016). "Radeon Instinct Machine Learning GPUs include Vega, Preview Performance". PC Per. Retrieved 12 December 2016.
  8. ↑ "Radeon Instinct MI6". Radeon Instinct. AMD. Retrieved 22 June 2017.
  9. ↑ https://www.techpowerup.com/gpudb/2927/radeon-instinct-mi6
  10. ↑ "Radeon Instinct MI8". Radeon Instinct. AMD. Retrieved 22 June 2017.
  11. ↑ https://www.techpowerup.com/gpudb/2928/radeon-instinct-mi8
  12. ↑ Smith, Ryan (5 January 2017). "The AMD Vega Architecture Teaser: Higher IPC, Tiling, & More, coming in H1'2017". Anandtech.com. Retrieved 10 January 2017.
  13. ↑ "Radeon Instinct MI25". Radeon Instinct. AMD. Retrieved 22 June 2017.
  14. ↑ https://www.techpowerup.com/gpudb/2983/radeon-instinct-mi25
  15. ↑ https://www.techpowerup.com/gpudb/3269/radeon-instinct-mi25-mxgpu
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.