AMD Radeon RX 7900 XT
RDNA 3 consumer GPU for local AI inference with 20GB of memory.
- Arch
- RDNA 3
- Memory
- 20 GB GDDR6
- Bandwidth
- 800 GB/s
- Released
- Q4 2022
At a glance
We don't currently track any Radeon RX 7900 XT offers. See alternative GPUs.
Technical Specifications
AMD Radeon RX 7900 XT · Per GPU
| Compute · dense | |
|---|---|
| FP16 / BF16 | 103 TFLOPS |
| INT8 | 103 TOPS |
| FP32 | 51.6 TFLOPS |
| Precision support | FP16BF16FP32INT4INT8 |
| Memory | |
|---|---|
| Capacity | 20 GB GDDR6 |
| Bandwidth | 800 GB/s |
| Bus width | 320-bit |
| Silicon | |
|---|---|
| Architecture | RDNA 3 |
| Process | 5nm / 6nm |
| Transistors | 58 billion |
| Shader cores | 5,376 Stream processors |
| Matrix cores | 168 |
| Compute units | 84 |
| Power | |
|---|---|
| Board power | 315 W |
Frequently Asked Questions
Why choose the Radeon RX 7900 XT?
20GB on RDNA 3, more than the 16GB cards for a 14B model at 8-bit or a 7B model at FP16 with longer context. AMD lists it as supported by ROCm on Linux.
When is the Radeon RX 7900 XT not a good fit?
No FP8 or structured sparsity on RDNA 3. If the model fits in 16GB, the RX 9070 XT adds FP8 and a higher dense FP16 rate (195 against 103 TFLOPS). Where a framework needs CUDA, the RTX 4080 is a 16GB consumer alternative. The ROCm software ecosystem, while rapidly improving, is narrower than CUDA.
Alternatives to AMD Radeon RX 7900 XT
Last updated