GPU comparison
MI250X vs MI210.
The MI250X is ahead on memory (128 GB vs 64 GB) and speed (2.1x the FP16 throughput and 2.0x the memory bandwidth). The MI210 makes sense when your model fits in 64 GB and its hourly price is lower.
Rent either on Runcrate
- MI250XNot in stock right now
- MI210Not in stock right now
Which is better for
Which to pick for each job. Memory fit and throughput, from the datasheets.
Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
MI250X: 1 GPU, 58 GB spare
MI210: 2 GPUs (128 GB)
MI250X. Fits on one GPU.
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
MI250X: 1 GPU, 112 GB spare
MI210: 1 GPU, 48 GB spare
MI250X. Same GPU count, 2.0x the memory bandwidth.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
MI250X: 1 GPU, 82 GB spare
MI210: 1 GPU, 18 GB spare
MI250X. Same GPU count, 2.1x the FP16 throughput.
Train on a full node
FP16 tensor throughput per node
MI250X: 8x, 3,064 TFLOPS
MI210: 4x, 724 TFLOPS
MI250X. 8 GPUs per node, not 4.
Throughput per watt
FP16 TFLOPS per 100 W of TDP
MI250X: 68 TFLOPS per 100 W
MI210: 60 TFLOPS per 100 W
Either. Close per watt.
Compare another pair
CDNA 2
AMD MI250X
- Memory
- 128 GB HBM2e
- FP16 tensor
- 383 TFLOPS
- Bandwidth
- 3.3 TB/s
CDNA 2
AMD MI210
- Memory
- 64 GB HBM2e
- FP16 tensor
- 181 TFLOPS
- Bandwidth
- 1.6 TB/s
At a glance
Spec by spec. The longer bar is the better number.
Full specs
The complete spec sheet.
| Spec | AMD MI250X | AMD MI210 |
|---|---|---|
| Architecture | CDNA 2 | CDNA 2 |
| Memory | 128 GB | 64 GB |
| Memory type | HBM2e | HBM2e |
| Memory bandwidth | 3.3 TB/s | 1.6 TB/s |
| FP32 | 47.9 TFLOPS | 23.9 TFLOPS |
| FP16 tensor | 383 TFLOPS | 181 TFLOPS |
| INT8 | 383 TOPS | 192 TOPS |
| TDP | 560 W | 300 W |
| Form factor | OAM | PCIe |
| NVLink | No | No |
| GPUs per node | 8 | 4 |
| Hourly price (live) | Not in stock right now | Not in stock right now |
Questions
MI250X or MI210. What buyers ask first.
Other comparisons