Radeon RX 7900 GRE
AMDconsumer| Overview | |
|---|---|
| Architecture | RDNA 3 (2023) |
| Launch date | 2023-07-27 |
| Transistor count | 54B |
| GPU target (gfx) | gfx1100 |
| Memory | |
| VRAM | 16 GB GDDR6 |
| Memory bandwidth | 576 GB/s |
| Cache | 64 MB |
| Compute — vector | |
| FP32 | 46 TFLOPS |
| FP16 | 46 TFLOPS |
| Compute — matrix / tensor | |
| FP16 | 92 TFLOPS |
| INT8 | 92 TOPS |
| INT4 | 184 TOPS |
| Cores & clocks | |
| Compute Units | 80 |
| Shader cores | 5,120 |
| Matrix / Tensor cores | 160 |
| Peak clock | 2,245 MHz |
| Board & system | |
| TDP | 260 W |
| Interconnect | PCIe only (no Infinity Fabric multi-GPU link) |
Compatible ROCm versions
A cut-down Navi 31 part, so it reports gfx1100 and inherits the RX 7900 XTX's ROCm support from 6.0.0 onward — useful to know, since the "GRE" naming gives no hint that it shares a target with the flagship.
transistorCountBillion is 54 B here versus 58 B for the 7900 XTX/XT on the same die family: AMD publishes the lower figure for this SKU, consistent with the GRE shipping fewer memory-cache dies (its 64 MB Infinity Cache and 256-bit interface versus the XTX's 96 MB / 384-bit). Recorded as published rather than normalised to the XTX's number.
fp16TFLOPS (92, dense) is AMD's published "Peak Half Precision (FP16 Matrix) Performance"; the 46 TFLOPs vector rate is captured separately as fp16TFLOPSVector. No structured-sparsity variants are listed (RDNA 3 has no 2:4 sparsity). fp8TFLOPS omitted — no native FP8 matrix support on RDNA 3; INT8/INT4 matrix rates are published instead via the card's 160 "AI Accelerators" (used as matrixCoreCount).
memoryBandwidthGBs uses the raw GDDR6 figure (576 GB/s), not an Infinity Cache-effective number. peakClockMHz uses the published Boost Frequency ("Up to 2245 MHz"), not the 1880 MHz Game Frequency. AMD publishes no MSRP on this page, so msrpUSD is omitted. Source: amd.com/en/products/graphics/desktops/radeon/7000-series/amd-radeon-rx-7900-gre.html ("Expand All" spec accordion), pulled 2026-09-20.