Radeon RX 9070 XT Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
AMD's RX 9070 XT (2025) is a 16GB gaming card that reads long prompts fast, Gemma 4 12B at 4-bit about 61 tokens/s. Some use 12V-2x6 power.
eBay
Radeon RX 9070 XT price history
eBay
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
Full Radeon RX 9070 XT specifications
- Memory16 GB · GDDR6
- Bandwidth640 GB/s
- FP16 / BF16195 TFLOPS
- FP8 / INT8389 / 389 TOPS
- Bus width
- 256-bit
- Architecture
- RDNA 4 (Navi 48)
- Cores
- 4,096
- Tensor cores
- 128
- Ecosystem
- ROCm
- Interface
- PCIe 5.0 x16
- Power
- 304 W
- Power connector
- 2× 8-pin
- Type
- Consumer
- Launch date
- Mar 6, 2025
- MSRP
- $599
Best Radeon RX 9070 XT deals on eBay
Hand-picked deals from trusted eBay sellers. Shop with confidence.
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
UsedASROCK CHALLENGER AMD RADEON RX 9070XT 16GB GDDR6 DESKTOP GPU *NO HDMI* *READ*$730Buy on eBay
UsedGIGABYTE AMD Radeon RX 9070 XT GAMING 16GB Graphics Card - Includes Warranty!Top Rated$798Buy on eBay
UsedAsus Radeon RX 9070 XT Prime OC 16GB GDDR6 Graphics Card PRIME-RX9070XT-016GTop Rated$800Buy on eBay
UsedASUS PRIME Radeon RX 9070 XT GDDR6 OC 16GB Graphics Card *READ*$800Buy on eBay
UsedPowerColor Reaper Radeon RX 9070 XT 16BPCI Express 5.0 ATX - Black$830+ shippingBuy on eBay
UsedAMD Radeon RX 9070 XT Sapphire Pulse 16GB GDDR6 Graphics Card 11348-03-20GTop Rated$850Buy on eBay
What LLMs can a single Radeon RX 9070 XT run?
One 9070 XT tops out at 12B-class at 4-bit; Gemma 4 26B A4B is just too big. 30B-class MoE at 2-bit is fastest, about 140 tokens/s.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 107.5 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 99.7 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 87.3 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 120.4 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 38.5 | 64K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 45.6 | 93K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 132.1 | 75K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 37.1 | 41K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 143.7 | 181K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 85.1 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 75.3 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 61.2 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Needs RAM offload | 37.2 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 16.5 GB | Too big | — | — |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Needs RAM offload | 35.7 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Needs RAM offload | 50 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 54.7 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 47.5 | 196K |
| Gemma 4 12B12B | 12.7 GB | Runs | 36 | 183K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 25.1 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Too big | — | — |
| Qwen3.8 27B27.8B | 29 GB | Too big | — | — |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Needs RAM offload | 22.3 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 33.2 | 256K |
RunsFits fully in VRAM; speed is estimated from VRAM bandwidth.
Needs RAM offloadMoE only: expert weights sit in system RAM, so speed is estimated from 70 GB/s RAM bandwidth.
Too bigWon’t fit on one card at this quantization, even with RAM offload.
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Regularly reading long documents or running coding assistants
2.6 times the matrix compute of the 7800 XT, so it is much faster on long prompts
- You want to try FP8
Radeon gaming cards only gained FP8 compute with this generation, and the previous RX 7000 series doesn't have it
- Testing the waters with 16GB and possibly moving to 32GB later
It uses the same chip as the Radeon AI PRO R9700, so switching cards needs no software changes
Not a good fit
- Generation speed is all you care about
Its bandwidth is modest, so it runs the same model at about three-quarters the speed of an RTX 5070 Ti
- Running Gemma 4 31B
At 4-bit it needs 18.3 GB and doesn't fit, so you're limited to 2-bit with up to about 41K of context
Before you buy
Power connectors depend on the model
AMD recommends a 750W power supply (the card draws 304W), and the reference spec is two 8-pin connectors, but manufacturers can change that: the Sapphire Nitro+ switched to a 12V-2x6 connector and includes an adapter to three 8-pin cables. Before buying a partner card, check that model's spec page to confirm your power supply has enough cables.
Software status
Software support for this generation has caught up: ROCm 10.0 officially supports it on both Linux and Windows, the vLLM documentation lists the RX 9000 series, and Ollama's Linux version recognizes it directly through ROCm. AMD's Windows PyTorch support list also includes the RX 9070 XT and notes that FP8 is available only on this generation; llama.cpp works with either the ROCm or the Vulkan backend.
Compared with similar GPUs
Radeon RX 9070 XT vs GeForce RTX 5070 Ti, Radeon AI PRO R9700, and Radeon RX 7800 XT: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | Radeon RX 9070 XT |  GeForce RTX 5070 Ti | Radeon AI PRO R9700 | Radeon RX 7800 XT |
|---|---|---|---|---|
| Memory | 16 GB | 16 GBSame | 32 GB100% more | 16 GBSame |
| Bandwidth | 640 GB/s | 896 GB/s40% higher | 640 GB/sSame | 624 GB/sSame |
| BF16 / FP16 compute | 195 TFLOPS | 176 TFLOPS10% lower | 191 TFLOPSSame | 75 TFLOPS62% lower |
| INT8 compute | 389 TOPS | 352 TOPS10% lower | 383 TOPSSame | 75 TOPS81% lower |
| Qwen3.8 27B speed (2-bit) | 45.6 t/s | 62.3 t/s37% faster | 45.6 t/sSame | 44.6 t/sSame |
| Largest dense model (4-bit) | 23B | 23BSame | 51B122% more | 23BSame |
| Power | 304 W | 300 WSame | 300 WSame | 263 W13% lower |
| Interface | PCIe 5.0 x16 | PCIe 5.0 x16 | PCIe 5.0 x16 | PCIe 4.0 x16 |
| Released | 2025-03 | 2025-02 | 2025-07 | 2023-09 |
| eBay median |
As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.
Amazon.jp
Xianyu