← All GPUs
AMD · Consumer · RDNA 3 (Navi 31, gfx1100)
AMD Radeon RX 7900 XTX
ROCm supported (gfx1100) on Ubuntu only for Radeon RX cards; AMD's page does not state the PCIe link.
VRAM
24 GB
GDDR6
Memory bandwidth
960 GB/s
FP16 compute
123 TFLOPS
dense tensor
Price
~$999
new, maker or suggested price
What fits at 4-bit
The 4-bit file plus 8,192 tokens of context and 1.5 GB of overhead, against the VRAM of 1 and 2 cards. Speeds are measured medians where someone has published a run at that card count; a blank speed means no run, not a slow one. Machine pages carry estimates for the gaps.
| Model | Needs | 1 card (24 GB) | 2 cards (48 GB) |
|---|---|---|---|
| Qwen3.5 4B4B · Q4_K_M | 4.3 GB | fits | fits |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | 42 tok/smedian of 2 runs | fits |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | fits | fits |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GB | fits | fits |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | fits | fits |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GB | fits | fits |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GB | fits | fits |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | fits | fits |
| Phi-414.7B · Q4_K_M | 11.3 GB | fits | fits |
| gpt-oss-20b21B, 3.6B active · MXFP4 (native) | 12.5 GB | 142 tok/smedian of 2 runs | fits |
| Devstral Small 2 24B (2512)24B · Q4_K_M | 16.1 GB | fits | fits |
| Qwen3.6 27B27B · Q4_K_M | 17.7 GB | fits | fits |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M | 17.4 GB | fits | fits |
| Qwen3.8 27B27B · Q4_K_M | 18.2 GB | fits | fits |
| Granite 4.2 30B29.3B · Q4_K_M | 20.3 GB | fits | fits |
| Gemma 4 31B IT30.7B · Q4_K_M | 18.9 GB | fits | fits |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M | 19.5 GB | fits | fits |
| Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M | 21.9 GB | fits | fits |
| Qwen3.6 35B-A3B35B, 3B active · Q4_K_M | 22.4 GB | fits | fits |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M | 25.3 GB | too large | fits |
| Llama 3.3 70B Instruct70B · Q4_K_M | 43.6 GB | too large | fits |
| Qwen3 Coder Next80B, 3B active · Q4_K_M | 46.9 GB | too large | fits |
Machines built on it
- 2x Radeon RX 7900 XTX~$5,637
Measured speeds
One card
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| gpt-oss-20b | UD-Q4_K_XL | llama.cpp (ROCm, build not stated) | 150 tok/s | – | 0 | llama.cpp benchmark discussion2026-04 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b6520 ROCm | 133 tok/s | 4,545 tok/s2,048-token prompt | 0 | llama.cpp benchmark discussion2025-09 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 45 tok/s | 885 tok/s | 0 | LocalScore2026-10 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 39 tok/s | 865 tok/s | 0 | LocalScore2026-10 |
2 cards
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| Gemma 4 31B IT | Q8_0 | llama.cpp (build not stated) Vulkan | 24 tok/s | 852 tok/s512-token prompt | 0 | llama.cpp benchmark discussion2026-08 |
| Gemma 4 31B IT | Q8_0 | llama.cpp (build not stated) ROCm | 23 tok/s | 921 tok/s512-token prompt | 0 | llama.cpp benchmark discussion2026-08 |
Speculative decoding, several users, or long context
Kept apart because each measures a different workload from one person chatting near the start of a conversation.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, internal AllReduce, RCCL OFF; PR #27825) | 37 tok/s | 1,357 tok/s8,192-token prompt | 0 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, internal AllReduce, RCCL OFF; PR #27825) | 37 tok/s | 1,446 tok/s512-token prompt | 0 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, RCCL ON) | 36 tok/s | 1,580 tok/s512-token prompt | 0 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, RCCL ON) | 36 tok/s | 1,479 tok/s8,192-token prompt | 0 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, internal AllReduce, RCCL OFF; PR #27825) | 32 tok/s | 792 tok/s512-token prompt | 100,000 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
| Qwen3.8 27B | Q8_0 | llama.cpp b10998 (HIP, RCCL ON) | 31 tok/s | 823 tok/s512-token prompt | 100,000 | 2× Radeon RX 7900 XTX | llama.cpp benchmark discussion2026-09 |
Specifications
- Memory
- 24 GB GDDR6
- FP16 basis
- AMD 'Peak Half Precision (FP16 Matrix)' figure.
- Board power
- 355 W
- Slots
- 2.5
- Length
- 287 mm
- NVLink
- No
- Cooling
- Open-air
- Software
- ROCm, Vulkan
- Released
- 2022
- Availability
- sold new
- Price basis
- AMD SEP at launch; AMD still lists the product and has published no revised SEP. AMD's US store does not sell Radeon cards.
- Maker page
- amd.com
Sources
- amd.com: 24 GB GDDR6, 960 GB/s, 355 W TBP, 287 mm, 2.5 slots, 123 TFLOPS FP16 matrix
- amd.com: SEP $999
- rocm.docs.amd.com: ROCm 7.14 supported-GPU list: 7900 XTX/W7900 (gfx1100), 9070 XT/R9700 (gfx1201), MI100 (gfx908) supported; MI50 (gfx906) not supported; Radeon RX cards Ubuntu-only