AMD Instinct MI300A vs AMD Instinct MI250

AMD Instinct MI300A
AMD Instinct MI250
14592 Shaders
128GB HBM3
2100MHz
13312 Shaders
128GB HBM2e
1700MHz
Peak AI Performance
3.92 POPS
INT4 Tensor Sparse
Peak AI Performance
362.09 TOPS
INT4 Tensor
FP32
122.57 TFLOPS
FP32
45.26 TFLOPS
FP16
122.57 TFLOPS
FP16
90.52 TFLOPS
Form Factor
APU SH5 Socket
-
Form Factor
OAM Module
-
TDP
560W
TDP
500W
-
-
-
-
-
-
-
-
-
-
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
3.92 POPS
INT4 Tensor Sparse
Peak AI Performance
362.09 TOPS
INT4 Tensor
-
-
-
-
-
-
FP8
-
1.96 PFLOPS Tensor (FP16 Accumulate)
3.92 PFLOPS Tensor (FP16 Accumulate) Sparse
1.96 PFLOPS Tensor (FP32 Accumulate)
3.92 PFLOPS Tensor (FP32 Accumulate) Sparse
-
-
-
-
-
-
FP16
122.57 TFLOPS
980.58 TFLOPS Tensor (FP16 Accumulate)
1.96 PFLOPS Tensor (FP16 Accumulate) Sparse
980.58 TFLOPS Tensor (FP32 Accumulate)
1.96 PFLOPS Tensor (FP32 Accumulate) Sparse
FP16
90.52 TFLOPS
362.09 TFLOPS Tensor (FP16 Accumulate)
-
362.09 TFLOPS Tensor (FP32 Accumulate)
-
FP32
122.57 TFLOPS
122.57 TFLOPS Tensor
245.15 TFLOPS Tensor Sparse
FP32
45.26 TFLOPS
90.52 TFLOPS Tensor
-
FP64
61.29 TFLOPS
122.57 TFLOPS Tensor
FP64
45.26 TFLOPS
90.52 TFLOPS Tensor
BF16
-
980.58 TFLOPS Tensor
1.96 PFLOPS Tensor Sparse
BF16
-
362.09 TFLOPS Tensor
-
TF32
490.29 TFLOPS Tensor
980.58 TFLOPS Tensor Sparse
-
-
-
INT4
1.96 POPS Tensor
3.92 POPS Tensor Sparse
INT4
362.09 TOPS Tensor
-
INT8
-
1.96 POPS Tensor
3.92 POPS Tensor Sparse
INT8
-
362.09 TOPS Tensor
-
-
-
-
-
-
-
-
-
Pixel Fillrate
-
Pixel Fillrate
-
-
-
-
-
Texture Fillrate
1915.2 GTexel/s
Texture Fillrate
1414.4 GTexel/s
Manufacturer
AMD
Manufacturer
AMD
Chip Designer
AMD
Chip Designer
AMD
Architecture
CDNA 3
Architecture
CDNA 2
Family
Instinct
Family
Instinct
Codename
Aqua Vanjaram
Aqua Vanjaram XT
Variant
Aqua Vanjaram XT
Codename
Aldebaran
Aldebaran XT
Variant
Aldebaran XT
Market Segment
Server
Market Segment
Server
Release Date
12/6/2023
Release Date
11/8/2021
Foundry
TSMC
TSMC Active Interposer Die
Foundry
TSMC
-
Fabrication Node
N5
N6 Active Interposer Die
Fabrication Node
N6
-
Die Size
3x 370 mm²
-
Die Size
2x 724 mm²
-
Transistor Count
3x 36.5 Billion
-
Transistor Count
2x 28 Billion
-
Transistor Density
98.65M/mm²
-
Transistor Density
38.67M/mm²
-
Form
APU SH5 Socket
Form
OAM Module
Shading Units
14592 Shaders
-
Shading Units
13312 Shaders
-
Texture Mapping Units
912 TMUs
Texture Mapping Units
832 TMUs
Render Output Units
-
Render Output Units
-
Tensor Cores
912 T-Cores
Tensor Cores
832 T-Cores
-
-
-
-
-
-
-
-
Compute Units
228 CUs
Compute Units
208 CUs
-
-
-
-
-
-
-
-
-
-
1000MHz Base
2100MHz
-
-
1000MHz Base
1700MHz
-
-
-
-
L1
-
-
16KB/CU
-
L1
-
-
16KB/CU
-
L2
16MB Shared
L2
16MB Shared
L3
192MB Shared
-
-
-
-
128GB
HBM3
ECC
128GB
HBM2e
ECC
Bus Width
8192Bit
Bus Width
8192Bit
Clock
2600MHz
Transfer Rate
5.2GT/s
Bandwidth
5324.8GB/s
Clock
1600MHz
Transfer Rate
3.2GT/s
Bandwidth
3276.8GB/s
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
TDP
560W
TDP
500W
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
No Ports
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
No Ports
Max Resolution
Unknown
Max Resolution
Unknown
Max Resolution Refresh Rate
-
Max Resolution Refresh Rate
-
Variable Refresh Rate
-
-
-
Variable Refresh Rate
-
-
-
Display Stream Compression (DSC)
Not Supported
Display Stream Compression (DSC)
Not Supported
Multi Monitor Support
Unknown
Multi Monitor Support
Unknown
-
-
-
-
Model
2x VCN 2.6
Model
2x VCN 2.6
Codec
-
-
-
-
-
-
-
-
AVC (H.264)
HEVC (H.265)
-
-
-
-
Codec
-
-
-
-
-
-
-
-
AVC (H.264)
HEVC (H.265)
-
-
-
-
No Decoders
No Decoders
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
OpenCL
3.0
-
-
-
-
OpenCL
3.0
-
-
-
-
-
-
GFX
9.4
-
-
-
-
-
-
-
-
GFX
9.4
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
PCIe Version
5.0
PCIe Lanes
16
-
-
PCIe Version
4.0
PCIe Lanes
16
Multi GPU Support
Supported
Type
Infinity Fabric
Multi GPU Support
Supported
Type
Infinity Fabric
-
-
-
-
-
-
-
-
-
-
-
-
Change Comparison