Radeon RX 7900 XTX 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 7900 XTX is AMD's 2022 24GB flagship gaming card, running Qwen3.8 27B at 4-bit at about 41 tokens/s. Many vLLM quantized models won't run on it.
eBay
Radeon RX 7900 XTX 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 7900 XTX specifications
- Memory24 GB · GDDR6
- Bandwidth960 GB/s
- FP16 / BF16123 TFLOPS
- INT8123 TOPS
- Bus width
- 384-bit
- Architecture
- RDNA 3 (Navi 31)
- Cores
- 6,144
- Tensor cores
- 192
- Ecosystem
- ROCm
- Interface
- PCIe 4.0 x16
- Power
- 355 W
- Power connector
- 2× 8-pin
- Length
- 287 mm
- Type
- Consumer
- Launch date
- Dec 13, 2022
- MSRP
- $999
Best Radeon RX 7900 XTX 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.
UsedMSI AMD Radeon RX 7900 XTX Gaming 24GB Graphics Card GPUTop Rated$975Buy on eBay
UsedPowerColor AMD Radeon RX 7900 XTX Gaming 24GB Graphics Card GPUTop Rated$975Buy on eBay
UsedXFX Speedster Merc 310 Radeon RX 7900 XTX 24GB GDDR6 Graphics Card$1,140+ shippingBuy on eBay
UsedSAPPHIRE PULSE AMD Radeon RX 7900 XTX 24GB GDDR6 Graphics Card - Excellent$1,175Buy on eBay
UsedPowerColor Red Devil AMD RX 7900 XTX 24GB OC Graphics Card RX7900XTX 24G-E/OCTop Rated$1,300Buy on eBay
UsedXFX Speedster MERC 310 Radeon RX 7900 XTX 24GB - Graphics Card - ExcellentTop Rated$1,320Buy on eBay
What LLMs can a single Radeon RX 7900 XTX run?
A single 7900 XTX is comfortable only at 4-bit, topping out at Qwen3.6 35B A3B at about 137 tokens/s; a 27B dense model runs a third as fast.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 150 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 139.9 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 123.5 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 140.8 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 56.2 | 192K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 66.3 | 221K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 151.2 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 54.2 | 143K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 161.1 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 120.5 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 107.2 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 88 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 118.8 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 40.7 | 117K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 41.4 | 122K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 133.7 | 115K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 36.7 | 66K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 136.7 | 123K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 79.1 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 69 | 256K |
| Gemma 4 12B12B | 12.7 GB | Runs | 52.7 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Needs RAM offload | 26.3 | 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 | 22.8 | 198K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 34.8 | 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.
What LLMs can multiple Radeon RX 7900 XTX cards run?
Large MoE on the 7900 XTX starts at four cards (96GB): Qwen3.8 Flash Next at 2-bit, about 113 tokens/s. No NVLink; use CPU-connected slots.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 4 cards | $5,000 | 96 GB | 112.5 | 1,420 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 8 cards | $10,000 | 192 GB | 92.2 | 2,840 W |
 Hy3295B (21B active) | 107.2 GB | 8 cards | $10,000 | 192 GB | 45 | 2,840 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 8 cards | $10,000 | 192 GB | 75.4 | 2,840 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 8 cards | $10,000 | 192 GB | 96.2 | 2,840 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 8 cards | $10,000 | 192 GB | 69.1 | 2,840 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 8 cards | $10,000 | 192 GB | 67.2 | 2,840 W |
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- Running 27B to 31B dense models on one card
Qwen3.8 27B at 4-bit runs at about 41 tokens/s with up to about 120K of context
- Running MoE models
Qwen3.6 35B A3B at 4-bit (22.1 GB) fits entirely in VRAM, at about 137 tokens/s
- You only use tools like llama.cpp and Ollama, and want official ROCm support on both Linux and Windows
- You want to avoid ex-mining cards
Ethereum had already stopped mining when it launched, so used cards have almost never been mined on
Not a good fit
- Running AWQ or GPTQ quantizations in vLLM
Neither format is supported on AMD cards
- Running 27B models at 8-bit
Qwen3.8 27B needs 29 GB at 8-bit, which 24GB can't hold
- Generation speed is all you care about
With the same 24GB, it roughly matches the RTX 3090, so moving from a 3090 brings no gain
Before you buy
On reference cards, check the hotspot temperature first
Some coolers (vapor chambers) in the launch batch of AMD's reference 7900 XTX didn't have enough liquid, so the hotspot temperature shot up to 110°C under full load and the card throttled. AMD acknowledged the problem at the time and offered replacements; partner cards were not affected. When buying a used reference card, run it at full load for ten minutes and watch the Hotspot temperature: if it sits near 110°C with the fans maxed out, it is very likely one of the affected cards.
Power supply and size
The card draws 355W, and AMD recommends an 800W system power supply. The reference card is 287 mm long, takes 2.5 slots and uses two 8-pin connectors; partner cards vary widely in length and power connectors, so check the specific model.
Choosing software
For personal chat and coding, llama.cpp or Ollama is the easiest route: ROCm officially supports this card on both Linux and Windows, both tools recognize it out of the box, and AMD also provides a validated Windows build of llama.cpp. The vLLM route is narrow: AWQ, GPTQ and bitsandbytes are all unsupported on AMD cards, this generation has no FP8, and 24GB can't hold the original BF16 version of a 27B-class model, which leaves mainly GGUF.
Compared with similar GPUs
Radeon RX 7900 XTX vs GeForce RTX 3090, GeForce RTX 4090, and Radeon PRO W7900: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | Radeon RX 7900 XTX |  GeForce RTX 3090 | GeForce RTX 4090 | Radeon PRO W7900 |
|---|---|---|---|---|
| Memory | 24 GB | 24 GBSame | 24 GBSame | 48 GB100% more |
| Bandwidth | 960 GB/s | 936 GB/sSame | 1,008 GB/s5% higher | 864 GB/s10% lower |
| BF16 / FP16 compute | 123 TFLOPS | 142 TFLOPS16% higher | 330 TFLOPS169% higher | 123 TFLOPSSame |
| INT8 compute | 123 TOPS | 285 TOPS131% higher | 661 TOPS437% higher | 123 TOPSSame |
| Qwen3.8 27B speed (2-bit) | 66.3 t/s | 64.8 t/sSame | 69.3 t/s5% faster | 60.3 t/s9% slower |
| Largest dense model (4-bit) | 37B | 37BSame | 37BSame | 79B114% more |
| Power | 355 W | 350 WSame | 450 W27% higher | 295 W17% lower |
| GPU interconnect | — | NVLink 112.5 GB/s | Not supported | — |
| Interface | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 | PCIe 4.0 x16 |
| Released | 2022-12 | 2020-09 | 2022-10 | 2023-04 |
| 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