Intel Gaudi 3 vs AMD Radeon Instinct MI50

Intel Gaudi 3
AMD Radeon Instinct MI50
16384 Shaders
128GB HBM2e
438MHz
3840 Shaders
16GB HBM2
1725MHz
Peak AI Performance
1.84 PFLOPS
FP8 Tensor (FP32 Accumulate)
Peak AI Performance
26.5 TFLOPS
FP16
FP32
14.35 TFLOPS
FP32
13.25 TFLOPS
FP16
28.7 TFLOPS
FP16
26.5 TFLOPS
Form Factor
OAM Module
-
Form Factor
PCIe Card
2.0-Slots
TDP
900W
TDP
300W
-
-
-
-
-
Power Connectors
-
2x 8-Pin
-
-
GB6 OpenCL N/A
0%
GB6 OpenCL N/A
0%
GB6 Metal N/A
0%
GB6 Metal N/A
0%
GB6 Vulkan N/A
0%
GB6 Vulkan N/A
0%
GB5 OpenCL N/A
0%
GB5 OpenCL N/A
0%
GB5 CUDA N/A
0%
GB5 CUDA N/A
0%
GB5 Metal N/A
0%
GB5 Metal N/A
0%
GB5 Vulkan N/A
0%
GB5 Vulkan N/A
0%
OCT 2020.1 N/A
0%
OCT 2020.1 N/A
0%
OCT Metal N/A
0%
OCT Metal N/A
0%
Peak AI Performance
1.84 PFLOPS
FP8 Tensor (FP32 Accumulate)
Peak AI Performance
26.5 TFLOPS
FP16
-
-
-
-
-
-
FP8
-
-
-
1.84 PFLOPS Tensor (FP32 Accumulate)
-
-
-
-
-
-
-
FP16
28.7 TFLOPS
-
-
458.75 TFLOPS Tensor (FP32 Accumulate)
-
FP16
26.5 TFLOPS
-
-
-
-
FP32
14.35 TFLOPS
229.38 TFLOPS Tensor
-
FP32
13.25 TFLOPS
-
-
-
-
-
FP64
6.62 TFLOPS
-
BF16
28.7 TFLOPS
1.84 PFLOPS Tensor
-
-
-
-
-
TF32
458.75 TFLOPS Tensor
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Pixel Fillrate
-
Pixel Fillrate
110.4 GPixel/s
-
-
-
-
Texture Fillrate
-
Texture Fillrate
414 GTexel/s
Manufacturer
Intel
Manufacturer
AMD
Chip Designer
Intel
Chip Designer
AMD
Architecture
Gaudi 3
Architecture
GCN 5
Family
Gaudi
Family
Instinct
Codename
Gaudi 3
Gaudi 3
-
-
Codename
Moonshot
Vega 20
Variant
Vega 20 XL GL
Market Segment
Server
Market Segment
Server
Release Date
4/9/2024
Release Date
11/18/2018
Foundry
TSMC
-
Foundry
TSMC
-
Fabrication Node
N5
-
Fabrication Node
N7
-
-
-
-
Die Size
331 mm²
-
-
-
Transistor Count
13.2 Billion
-
-
-
-
Transistor Density
39.97M/mm²
-
Form
OAM Module
Form
PCIe Card
Shading Units
16384 Shaders
-
Shading Units
3840 Shaders
-
Texture Mapping Units
-
Texture Mapping Units
240 TMUs
Render Output Units
-
Render Output Units
64 ROPs
Tensor Cores
64 T-Cores
-
-
-
-
-
-
-
-
-
-
Compute Units
8 CUs
Compute Units
60 CUs
-
-
-
-
-
-
-
-
-
-
-
438MHz
-
-
1450MHz Base
1725MHz
-
-
-
-
L1
-
-
-
Unknown
L1
-
-
16KB/CU
-
L2
Unknown
L2
4MB Shared
-
-
-
-
-
-
128GB
HBM2e
ECC
16GB
HBM2
ECC
Bus Width
8192Bit
Bus Width
4096Bit
Clock
1808MHz
Transfer Rate
3.6GT/s
Bandwidth
3702.8GB/s
Clock
1000MHz
Transfer Rate
2GT/s
Bandwidth
1024GB/s
-
-
-
-
eSRAM
96MB
12800GB/s
-
-
-
-
-
-
-
-
-
-
-
TDP
900W
TDP
300W
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
No Ports
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
1x mini-DisplayPort 1.4
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Max Resolution
Unknown
Max Resolution
7680x4320
Max Resolution Refresh Rate
-
Max Resolution Refresh Rate
60Hz
Variable Refresh Rate
-
-
-
Variable Refresh Rate
-
FreeSync
-
Display Stream Compression (DSC)
Not Supported
Display Stream Compression (DSC)
Not Supported
Multi Monitor Support
Unknown
Multi Monitor Support
3
-
-
Content Protection
HDCP 2.2
No Encoders
-
Model
VCE 4.1
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Codec
-
-
-
-
-
-
-
-
AVC (H.264)
HEVC (H.265)
-
-
-
-
No Decoders
Model
UVD 7.2
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Codec
MPEG-1
MPEG-2
MPEG-4
JPEG
VC-1
-
-
-
AVC (H.264)
HEVC (H.265)
-
-
-
-
-
-
-
-
Direct X
12
Direct 3D
12_1
-
-
OpenCL
3.0
-
-
OpenGL
4.6
OpenCL
2.1
Vulkan
1.3
-
-
-
-
-
-
-
-
-
-
Shader Model
6.7
-
-
GFX
9
-
-
-
-
Not a Card
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Power Connectors
-
-
-
2x 8-Pin
-
-
-
-
-
PCIe Version
5.0
PCIe Lanes
16
Slots Required
2.0
PCIe Version
3.0
PCIe Lanes
16
Multi GPU Support
Supported
Type
RoCE
Multi GPU Support
Supported
Type
CrossFire XDMA
-
-
-
-
-
-
Height
111 mm (4.37 in)
Width
267 mm (10.51 in)
Depth
37 mm (1.46 in)
Change Comparison