Radeon RX 7900 XTX
AMDconsumer| Overview | |
|---|---|
| Architecture | RDNA 3 (2022) |
| Launch date | 2022-12-13 |
| Transistor count | 58B |
| GPU target (gfx) | gfx1100 |
| Memory | |
| VRAM | 24 GB GDDR6 |
| Memory bandwidth | 960 GB/s |
| Cache | 96 MB |
| Compute — vector | |
| FP32 | 61.4 TFLOPS |
| FP16 | 61.4 TFLOPS |
| Compute — matrix / tensor | |
| FP16 | 123 TFLOPS |
| INT8 | 123 TOPS |
| INT4 | 246 TOPS |
| Cores & clocks | |
| Compute Units | 96 |
| Shader cores | 6,144 |
| Matrix / Tensor cores | 192 |
| Peak clock | 2,500 MHz |
| Board & system | |
| TDP | 355 W |
| Interconnect | PCIe only (no Infinity Fabric multi-GPU link) |
| MSRP | $999 |
Compatible ROCm versions
fp16TFLOPS (123) is directly published as "Peak Half Precision (FP16 Matrix) Performance" on AMD's own product page, not plain vector-shader FP32 (61.4 TFLOPs, much higher but not the matrix throughput number this collection uses; captured separately as fp16TFLOPSVector — also 61.4). No structured-sparsity variant is listed for this card (unlike RDNA4 parts), so there's only one figure to report. fp8TFLOPS omitted — RDNA 3 has no native FP8 matrix support (added in RDNA 4); AMD's page instead lists INT8/INT4 matrix rates (int8TOPS/int4TOPS) via "AI Accelerators" (192, used as matrixCoreCount). memoryBandwidthGBs uses the raw GDDR6 figure (960 GB/s), not the page's "Effective Memory Bandwidth" of 3,500 GB/s (Infinity Cache-inflated), for consistency with the rest of this collection. Source: amd.com/en/products/graphics/desktops/radeon/7000-series/amd-radeon-rx-7900xtx.html ("Expand All" spec accordion), pulled 2026-09-01.