Radeon RX 9060 XT (16GB)
AMDconsumer| Overview | |
|---|---|
| Architecture | RDNA 4 (2025) |
| Transistor count | 29.7B |
| GPU target (gfx) | gfx1200 |
| Memory | |
| VRAM | 16 GB GDDR6 |
| Memory bandwidth | 320 GB/s |
| Cache | 32 MB |
| Compute — vector | |
| FP32 | 25.6 TFLOPS |
| FP16 | 25.6 TFLOPS |
| Compute — matrix / tensor | |
| FP16 (dense / sparse) | 103 / 205 TFLOPS |
| FP8 (dense / sparse) | 205 / 410 TFLOPS |
| INT8 (dense / sparse) | 205 / 410 TOPS |
| INT4 | 410 TOPS |
| Cores & clocks | |
| Compute Units | 32 |
| Shader cores | 2,048 |
| Matrix / Tensor cores | 64 |
| Peak clock | 3,130 MHz |
| Board & system | |
| TDP | 160 W |
| PCIe | PCIe 5.0 x16 |
| Interconnect | PCIe only (no Infinity Fabric multi-GPU link) |
Compatible ROCm versions
Closes the coverage gap flagged in the RX 9060 XT (8GB) entry, whose notes recorded that the 16 GB SKU's URL 404'd during the 2026-09-01 pull. AMD serves the 16 GB variant at the unsuffixed URL (amd-radeon-rx-9060xt.html, page title "AMD Radeon RX 9060 XT (16GB)") while the 8 GB variant lives at amd-radeon-rx-9060xt-8gb.html — worth remembering, since guessing a "-16gb" suffix returns a 404.
Compute specs are identical to the 8 GB variant across the board (32 CUs, 2048 stream processors, 64 AI accelerators, 3130 MHz boost, same FP32/FP16/FP8/INT8/INT4 figures) — the two SKUs differ only in memory capacity (16 GB vs 8 GB) and board power (160 W vs 150 W). Memory bandwidth is the same 320 GB/s on a 128-bit bus at 20 Gbps, so the 16 GB card fits larger models without feeding them any faster. For local LLM use that still makes it the more sensible of the two: 8 GB is the binding constraint on the smaller SKU well before compute is.
Shares gfx1200 (Navi 44) with the 8 GB variant, so ROCm support is the same — from 6.4.0 (2025-04) onward.
fp16TFLOPS (103, dense) and fp8TFLOPS (205, dense) are both directly published with explicit "with Structured Sparsity" variants (205 / 410) recorded in the ...Sparse fields. int8TOPS/int8TOPSSparse mirror the FP8 figures exactly (205/410); int4TOPS is listed but its sparsity variant (821 TOPs) has no schema field. matrixCoreCount uses the page's "AI Accelerators" count (64).
peakClockMHz uses the published Boost Frequency ("Up to 3130 MHz"), not the 2530 MHz Game Frequency. No Launch Date field is published on this page and AMD's COMPUTEX 2025 announcement (2025-05-20) gives no specific availability date, so launchDate is omitted; releaseYear (2025) is unaffected. AMD publishes no MSRP on the spec page, so msrpUSD is omitted. Source: amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9060xt.html ("Expand All" spec accordion), pulled 2026-09-20.