GPU comparison
RTX 4090 vs MI210.
Pick the MI210 for memory: 64 GB vs 24 GB, with 63% more memory bandwidth. A QLoRA fine-tune of a 70B model (about 46 GB) fits on one MI210, not on one RTX 4090. For models under 24 GB pick by price.
Rent either on Runcrate
- RTX 4090from $0.58/hr
$0.024 per GB/hr
- MI210Not in stock right now
Live hourly price, then per GB of memory.
Which is better for
Which to pick for each job. Memory fit, from the datasheets.
Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
RTX 4090: 4 GPUs, more than one 2x node
MI210: 2 GPUs (128 GB)
MI210. Fits in one node (2 GPUs).
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
RTX 4090: 1 GPU, 8 GB spare
MI210: 1 GPU, 48 GB spare
MI210. Same GPU count, 63% more memory bandwidth.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
RTX 4090: 2 GPUs (48 GB)
MI210: 1 GPU, 18 GB spare
MI210. Fits on one GPU.
Compare another pair
Ada Lovelace
NVIDIA RTX 4090
- Memory
- 24 GB GDDR6X
- Bandwidth
- 1.0 TB/s
At a glance
Spec by spec. The longer bar is the better number.
Full specs
The complete spec sheet.
| Spec | NVIDIA RTX 4090 | AMD MI210 |
|---|---|---|
| Architecture | Ada Lovelace | CDNA 2 |
| Memory | 24 GB | 64 GB |
| Memory type | GDDR6X | HBM2e |
| Memory bandwidth | 1.0 TB/s | 1.6 TB/s |
| FP32 | 82.6 TFLOPS | 23.9 TFLOPS |
| FP16 | 165.2 TFLOPS, non-tensor | 181 TFLOPS, tensor |
| INT8 | 661 TOPS | 192 TOPS |
| TDP | 450 W | 300 W |
| Form factor | Consumer | PCIe |
| NVLink | No | No |
| GPUs per node | 2 | 4 |
| Hourly price (live) | from $0.58/hr | Not in stock right now |
Questions
RTX 4090 or MI210. What buyers ask first.
Other comparisons