Radeon RX 7800 XT Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
The RX 7800 XT is AMD's 2023 16GB gaming card, smooth on 9B to 12B models, Gemma 4 12B at about 60 tokens/s. ROCm supports it; Ollama just works.
eBay
Radeon RX 7800 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 7800 XT specifications
- Memory16 GB · GDDR6
- Bandwidth624 GB/s
- FP16 / BF1674.7 TFLOPS
- INT874.7 TOPS
- Bus width
- 256-bit
- Architecture
- RDNA 3 (Navi 32)
- Cores
- 3,840
- Tensor cores
- 120
- Ecosystem
- ROCm
- Interface
- PCIe 4.0 x16
- Power
- 263 W
- Power connector
- 2× 8-pin
- Length
- 267 mm
- Type
- Consumer
- Launch date
- Sep 6, 2023
- MSRP
- $499
Best Radeon RX 7800 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.
UsedXFX Speedster QICK 319 Radeon RX 7800 XT Core Edition 16GB GDDR6 Graphics CardTop Rated$559$999-44%Buy on eBay
UsedXFX Speedster MERC 319 Radeon RX 7800 XT 16GB - Graphics Card - ExcellentTop Rated$600Buy on eBay
UsedXFX Radeon RX 7800 XT Speedster QICK319 Core Edition 16GB GDDR6 Graphics CardTop Rated$600Buy on eBay
Used(Factory Refurbished) ASUS RX7800XT O16G PCIe 4.0 DUAL-RX7800XT-O16G Video Card$600Buy on eBay
UsedXFX Radeon RX 7800 XT 16GB Speedster SWFT210 Core GPU | 1yr Warranty, Fast Ship!Top Rated$610Buy on eBay
UsedXFX Radeon RX 7800 XT 16GB Merc319 Black GPU | 1yr Warranty, Fast Ship!Top Rated$620Buy on eBay
What LLMs can a single Radeon RX 7800 XT run?
A single 7800 XT is happiest with 9B to 12B at 4-bit. Gemma 4 26B A4B at 2-bit fits on the card at about 119 tokens/s, triple 4-bit with system RAM.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 105.2 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 97.6 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 85.4 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 119.1 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 37.6 | 64K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 44.6 | 93K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 130.9 | 75K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 36.2 | 41K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 142.5 | 181K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 83.3 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 73.6 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 59.8 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Needs RAM offload | 37 | 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 | 49.8 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 53.5 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 46.4 | 196K |
| Gemma 4 12B12B | 12.7 GB | Runs | 35.2 | 183K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 25 | 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 | 33.1 | 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
- Everyday use of Qwen3.5 9B and Gemma 4 12B
60 to 74 tokens/s at 4-bit, with room for 256K of context
- Power draw matters to you
The card draws 263W and AMD recommends a 700W power supply, the lowest among the 16GB AMD cards on our site
- You want ROCm on both Linux and Windows
It is on ROCm 10.0's official support list, and the vLLM documentation lists it too
Not a good fit
- Running models from the 24GB tier
Qwen3.8 27B and Gemma 4 31B both exceed 16GB at 4-bit
- Fitting MoE models entirely on the GPU
Gemma 4 26B A4B at 4-bit is 16.9 GB and Qwen3.6 35B A3B is 22.1 GB, so both have to put experts in system RAM, leaving just 37 to 50 tokens/s
- Regularly reading long documents
Its matrix compute is less than 40% of the RX 9070 XT's, so long prompts are slow
Before you buy
You can skip the HSA_OVERRIDE from old guides
Early guides often had 7800 XT owners set HSA_OVERRIDE_GFX_VERSION=11.0.0 so ROCm would treat the card as a 7900 XTX. ROCm 10.0 now officially supports it, and Ollama's Linux and Windows versions both recognize it directly, so just follow the normal install process. You might still need the override only if you insist on a very old ROCm.
Running on Windows
AMD provides a validated Windows build of llama.cpp: download, unzip, and you have ROCm acceleration. Ollama's Windows version also supports this card directly. If you hit compatibility problems, llama.cpp's Vulkan backend doesn't depend on ROCm and works as a fallback.
Size and power supply
AMD's reference card is 267 mm long, takes 2.5 slots and uses two 8-pin connectors, with a recommended 700W system power supply. The RX 6950 XT and RX 9070 XT, also 16GB, call for 850W and 750W, so if you want to keep your old power supply, this card is the easiest to satisfy.
Compared with similar GPUs
Radeon RX 7800 XT vs GeForce RTX 5060 Ti 16GB, Radeon RX 9070 XT, and Radeon RX 6950 XT: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | Radeon RX 7800 XT |  GeForce RTX 5060 Ti 16GB | Radeon RX 9070 XT | Radeon RX 6950 XT |
|---|---|---|---|---|
| Memory | 16 GB | 16 GBSame | 16 GBSame | 16 GBSame |
| Bandwidth | 624 GB/s | 448 GB/s28% lower | 640 GB/sSame | 576 GB/s8% lower |
| BF16 / FP16 compute | 75 TFLOPS | 95 TFLOPS27% higher | 195 TFLOPS161% higher | 47 TFLOPS37% lower |
| INT8 compute | 75 TOPS | 190 TOPS154% higher | 389 TOPS421% higher | 95 TOPS27% higher |
| Qwen3.8 27B speed (2-bit) | 44.6 t/s | 32.6 t/s27% slower | 45.6 t/sSame | 41.3 t/s7% slower |
| Largest dense model (4-bit) | 23B | 23BSame | 23BSame | 23BSame |
| Power | 263 W | 180 W32% lower | 304 W16% higher | 335 W27% higher |
| Interface | PCIe 4.0 x16 | PCIe 5.0 x8 | PCIe 5.0 x16 | PCIe 4.0 x16 |
| Released | 2023-09 | 2025-04 | 2025-03 | 2022-05 |
| 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