AMD Instinct MI440X OAM vs AMD Instinct MI455X
Side-by-side specifications, performance, and rental pricing for datacenter AI workloads.
Instinct MI440X: CDNA 5, 2026 Instinct MI455X: CDNA 5, 2026
AMD Instinct MI440X OAM
Full specs →Architecture
CDNA 5
Year
2026
AMD Instinct MI455X
Full specs →Architecture
CDNA 5
Year
2026
Memory
432 GB
Specifications
Instinct MI440X Instinct MI455X
Architecture
CDNA 5CDNA 5
Launch Year
20262026
Form Factor
OAMEAM
Memory
—432 GB
Memory Bandwidth
—23.3 TB/s
TDP
——
Process Node
—TSMC 2nm/3nm
Spec Instinct MI440X Instinct MI455X
Architecture CDNA 5CDNA 5
Launch Year 20262026
Form Factor OAMEAM
Memory —432 GB
Memory Bandwidth —23.3 TB/s
TDP ——
Process Node —TSMC 2nm/3nm
Performance (TFLOPS)
Instinct MI440X Instinct MI455X
FP64
No verified data available5 TFLOPS
FP32
No verified data available315 TFLOPS
TF32
No verified data available No verified data available
BF16
No verified data available5,033 TFLOPS
10,066 TFLOPS sparse
FP16
No verified data available5,033 TFLOPS
10,066 TFLOPS sparse
FP8
No verified data available20,133 TFLOPS
sparse not published
FP6
No verified data available20,133 TFLOPS
sparse not published
FP4
No verified data available40,265 TFLOPS
sparse not published
INT8
No verified data available5,033 TOPS
10,066 TOPS sparse
Precision Instinct MI440X Instinct MI455X
FP64 No verified data available5 TFLOPS
FP32 No verified data available315 TFLOPS
TF32 No verified data available No verified data available
BF16 No verified data available5,033 TFLOPS
10,066 TFLOPS sparse
FP16 No verified data available5,033 TFLOPS
10,066 TFLOPS sparse
FP8 No verified data available20,133 TFLOPS
sparse not published
FP6 No verified data available20,133 TFLOPS
sparse not published
FP4 No verified data available40,265 TFLOPS
sparse not published
INT8 No verified data available5,033 TOPS
10,066 TOPS sparse
FLOPS by Precision
What actually differs
Both GPUs launched in 2026: the Instinct MI440X on CDNA 5 and the Instinct MI455X on CDNA 5.
For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.
Get Comparison Updates
New GPUs added weekly. Be the first to see how they compare.