Radeon PRO W7800 Price & Used Prices: Which Local LLMs Can It Run?
Price alert
We'll email you when the eBay (US) price reaches your target.
The Radeon PRO W7800 is AMD's 2023 32GB workstation card; modest bandwidth makes MoE fast and dense models slow. A 48GB version exists, so check listings.
eBay
Radeon PRO W7800 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 PRO W7800 specifications
- Memory32 GB · GDDR6 ECC
- Bandwidth576 GB/s
- FP16 / BF1690.4 TFLOPS
- INT890.4 TOPS
- Bus width
- 256-bit
- Architecture
- RDNA 3 (Navi 31)
- Cores
- 4,480
- Tensor cores
- 140
- Ecosystem
- ROCm
- Interface
- PCIe 4.0 x16
- Power
- 260 W
- Power connector
- 2× 8-pin
- Slots
- 2
- Length
- 280 mm
- Type
- Workstation
- Launch date
- Apr 13, 2023
- MSRP
- $2,499
Best Radeon PRO W7800 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.
NewAMD Radeon PRO W7800 32GB AI TOP Graphics Card (US Stock)$1,799Buy on eBay
NewAMD Radeon PRO W7800 AI TOP 32GB GPU Graphics card$2,000Buy on eBay- NewAMD Radeon PRO W7800 AI TOP 32GB GPU Graphics card$2,010Buy on eBay
NewGIGABYTE AMD Radeon Pro W7800 Graphic Card - 32 GB GDDR6$2,492Buy on eBay- NewAMD Radeon PRO W7800 AI TOP 32GB GPU Graphics card$2,689Buy on eBay
NewGIGABYTE Radeon AI PRO W7800 32GB GDDR6 PCI Express 4.0 x16 ATX Graphics CardTop Rated$3,361$8,404-60%Buy on eBay
What LLMs can a single Radeon PRO W7800 run?
A single W7800 runs up to 35B at 4-bit: 30B-class MoE at about 110 tokens/s, 27B dense only about 25. 8-bit is cramped; stay at 4-bit.
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
 Gemma 4 E4B8B | 3.8 GB | Runs | 98.2 | 128K |
 Qwen3.5 9B9.7B | 4.1 GB | Runs | 91 | 256K |
| Gemma 4 12B12B | 4.7 GB | Runs | 79.6 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 10.5 GB | Runs | 114.9 | 256K |
| Qwen3.6 27B27.8B | 11.8 GB | Runs | 34.8 | 256K |
| Qwen3.8 27B27.8B | 9.8 GB | Runs | 41.3 | 256K |
 GLM 4.7 Flash30B (3B active) | 11.9 GB | Runs | 126.8 | 198K |
| Gemma 4 31B30.7B | 11.8 GB | Runs | 33.6 | 246K |
| Qwen3.6 35B A3B36B (3B active) | 12.3 GB | Runs | 138.7 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 5 GB | Runs | 77.5 | 128K |
| Qwen3.5 9B9.7B | 5.7 GB | Runs | 68.5 | 256K |
| Gemma 4 12B12B | 7.1 GB | Runs | 55.6 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 16.9 GB | Runs | 91.8 | 256K |
| Qwen3.6 27B27.8B | 16.8 GB | Runs | 25 | 245K |
| Qwen3.8 27B27.8B | 16.5 GB | Runs | 25.4 | 250K |
| GLM 4.7 Flash30B (3B active) | 18.3 GB | Runs | 107.1 | 198K |
| Gemma 4 31B30.7B | 18.3 GB | Runs | 22.5 | 168K |
| Qwen3.6 35B A3B36B (3B active) | 22.1 GB | Runs | 110.3 | 256K |
| Model | Size | Runs? | Speed t/s | Max context |
|---|---|---|---|---|
| Gemma 4 E4B8B | 8.2 GB | Runs | 49.6 | 128K |
| Qwen3.5 9B9.7B | 9.5 GB | Runs | 43.1 | 256K |
| Gemma 4 12B12B | 12.7 GB | Runs | 32.6 | 256K |
| Gemma 4 26B A4B25.2B (3.8B active) | 26.9 GB | Runs | 69.9 | 256K |
| Qwen3.6 27B27.8B | 28.6 GB | Runs | 15 | 69K |
| Qwen3.8 27B27.8B | 29 GB | Runs | 14.8 | 63K |
| GLM 4.7 Flash30B (3B active) | 31.8 GB | Runs | 80.7 | 26K |
| Gemma 4 31B30.7B | 32.6 GB | Too big | — | — |
| Qwen3.6 35B A3B36B (3B active) | 36.9 GB | Needs RAM offload | 32.7 | 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 PRO W7800 cards run?
Four W7800s (128GB) run Qwen3.8 Flash Next at 4-bit at about 71 tokens/s; two suit gpt-oss 120B at 2-bit, around 94. No NVLink.
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 78.9 GB | 4 cards | $9,968 | 128 GB | 85.6 | 1,040 W |
 DeepSeek V4 Flash284B (13B active) | 96.8 GB | 4 cards | $9,968 | 128 GB | 66.9 | 1,040 W |
 Hy3295B (21B active) | 107.2 GB | 4 cards | $9,968 | 128 GB | 29.5 | 1,040 W |
| GLM 5.3 Flash320B (18B active) | 108.7 GB | 4 cards | $9,968 | 128 GB | 52.7 | 1,040 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| Qwen3.8 Flash Next176B (6B active) | 111.3 GB | 4 cards | $9,968 | 128 GB | 70.5 | 1,040 W |
| DeepSeek V4 Flash284B (13B active) | 155.1 GB | 8 cards | $19,936 | 256 GB | 47.7 | 2,080 W |
| Hy3295B (21B active) | 182.2 GB | 8 cards | $19,936 | 256 GB | 21.8 | 2,080 W |
| GLM 5.3 Flash320B (18B active) | 199.7 GB | 8 cards | $19,936 | 256 GB | 32.7 | 2,080 W |
| Model | Size | Cards | Total price | Total VRAM | Speed t/s | Total power |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash284B (13B active) | 161.9 GB | 8 cards | $19,936 | 256 GB | 46.1 | 2,080 W |
| Qwen3.8 Flash Next176B (6B active) | 188.2 GB | 8 cards | $19,936 | 256 GB | 49.7 | 2,080 W |
Full calculation rules are on the Methodology page.
Who it suits
Good fit
- You use MoE models with few active parameters day to day
GLM 4.7 Flash and Qwen3.6 35B A3B use only 3B parameters per token, and both run above 100 tokens/s at 4-bit
- You want 32GB with mature software
It's the same chip generation as the RX 7900 XTX, and the W7800 is on Ollama's support lists for both Linux and Windows
- Your system's power supply is limited
260W, two 8-pin connectors, and AMD recommends a 700W power supply
Not a good fit
- You want dense 31B models to run fast
Gemma 4 31B at 4-bit runs at about 23 tokens/s, 40% slower than the RX 7900 XTX
- You often process long prompts or batched requests
It has less than half the R9700's matrix compute and no FP8
Before you buy
32GB and 48GB versions
AMD later released a 48GB W7800, and listings often mix the two. The data on this page is for the original 32GB version; when buying used, check the VRAM capacity, and look up the 48GB version's bandwidth and power separately.
Size and cooling
Dual slot, 280 mm long, with blower cooling that exhausts hot air straight out through the bracket, so with several cards side by side, temperatures are much easier to keep in check than with open-air gaming cards.
Software status
It uses the same chip as the RX 7900 XTX and has the same software support: ROCm 10.0 supports it on both Linux and Windows, and it's on Ollama's lists for both platforms. When reading older tutorials, note that in the ROCm 7.2.1 era, AMD's Windows PyTorch support table listed only the W7900, not the W7800; the current compatibility table has added it.
Compared with similar GPUs
Radeon PRO W7800 vs Radeon AI PRO R9700, Radeon PRO W7900, and Arc Pro B70: price, VRAM, bandwidth, compute and local LLM speed compared.
| Spec | Radeon PRO W7800 | Radeon AI PRO R9700 | Radeon PRO W7900 |  Arc Pro B70 |
|---|---|---|---|---|
| Memory | 32 GB | 32 GBSame | 48 GB50% more | 32 GBSame |
| Bandwidth | 576 GB/s | 640 GB/s11% higher | 864 GB/s50% higher | 608 GB/s6% higher |
| BF16 / FP16 compute | 90 TFLOPS | 191 TFLOPS111% higher | 123 TFLOPS36% higher | 184 TFLOPS103% higher |
| INT8 compute | 90 TOPS | 383 TOPS324% higher | 123 TOPS36% higher | 367 TOPS306% higher |
| Qwen3.8 27B speed (2-bit) | 41.3 t/s | 45.6 t/s10% faster | 60.3 t/s46% faster | 43.5 t/s5% faster |
| Largest dense model (4-bit) | 51B | 51BSame | 79B55% more | 51BSame |
| Power | 260 W | 300 W15% higher | 295 W13% higher | 230 W12% lower |
| Interface | PCIe 4.0 x16 | PCIe 5.0 x16 | PCIe 4.0 x16 | PCIe 5.0 x16 |
| Released | 2023-04 | 2025-07 | 2023-04 | 2026-03 |
| 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