Desktop GPU · AMD
What AI models can a Radeon RX 7900 GRE run?
16 GB covers almost everything people actually use day to day, including the 24B to 32B class that most reviewers consider the practical ceiling for a single consumer card. At 576 GB/s there is enough bandwidth to keep generation responsive for anything that fits.
The short answer
Assuming an 8k context window and default settings, these are the models worth downloading first.
Best all-rounder
GPT OSS 20B
Q5_K_M · 14.63 GB · about 149.5 tokens/s
Best for code
Mistral Small 24B Instruct 2501
IQ4_XS · 13.83 GB · about 32.1 tokens/s
Largest that still runs well
Gemma 3 27B Instruct
27.43B parameters · Q3_K_M · about 30.6 tokens/s
Fastest useful answer
PowerMoE 3B
about 302.5 tokens/s · 4.53 GB
See how fast it feels
Radeon RX 7900 GRE running GPT OSS 20B at Q5_K_M
YouWhy does my model use more memory when the conversation gets longer?
Because of the KV cache. Every token you send leaves behind a key and a value vector in each layer of the model, and those stay in memory for as long as the conversation lasts.
The weights are a fixed cost: load a 4-bit 8B model and that is about 4.8 GB, whether you write one word or ten thousand. The cache is the part that grows, and it grows in a straight line with the number of tokens in the window.
How steeply depends on the model's attention design. With grouped-query attention, several query heads share one key-value pair, which cuts the cache by that ratio. Without it, every head keeps its own, and a long context can cost more memory than the weights themselves.
If you are short on memory, the first thing to try is lowering the context window in your runtime, not the quantisation.
Reading your question
Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.
Every model, scored on this device
Each row uses the highest-quality quantisation that both fits and stays conversational. Speed is a single-stream estimate at 8k context.
| Model | Params | Verdict | Download | Memory used | Speed | Max context |
|---|---|---|---|---|---|---|
| GPT OSS 20B GPT-OSS | 20.91B (4.18B active) | Just fits | Q5_K_M | 14.63 GB | 149.5 t/s | 32k |
| Gemma 4 26B A4B Instruct Gemma | 25.81B | Runs great | IQ4_XS | 13.78 GB | 31.9 t/s | 64k |
| DeepSeek Coder V2 Lite Instruct DeepSeek | 15.71B (2.74B active) | Runs great | Q6_K | 12.96 GB | 178 t/s | 32k |
| Qwen2.5 14B Instruct Qwen | 14.77B | Runs great | Q6_K | 13.69 GB | 32.5 t/s | 8k |
| Qwen3 14B Qwen | 14.77B | Runs great | Q6_K | 13.44 GB | 33.1 t/s | 16k |
| Mistral Small 24B Instruct 2501 Mistral | 23.57B | Runs great | IQ4_XS | 13.83 GB | 32.1 t/s | 8k |
| Phi 4 Phi | 14.66B | Runs great | Q6_K | 13.67 GB | 32.5 t/s | 8k |
| Qwen1.5 MoE A2.7B Qwen | 14.32B (2.69B active) | Runs great | Q6_K | 13.16 GB | 116.7 t/s | 8k |
| Codestral 22B V0.1 Mistral | 22.25B | Runs great | IQ4_XS | 13.73 GB | 32.5 t/s | 8k |
| vLLM Translategemma 12B Instruct Gemma | 13.19B | Just fits | Q8_0 | 14.26 GB | 30.9 t/s | 32k |
| Mistral Nemo Instruct 2407 Mistral | 12.25B | Just fits | Q8_0 | 14.29 GB | 31 t/s | 8k |
| Gemma 3 12B Instruct Gemma | 12.19B | Runs great | Q8_0 | 13.71 GB | 32.2 t/s | 16k |
| Gemma 4 12B Instruct Gemma | 11.96B | Runs great | Q8_0 | 13.05 GB | 34 t/s | 64k |
| Gemma 3 27B Instruct Gemma | 27.43B | Just fits | Q3_K_M | 14.5 GB | 30.6 t/s | 8k |
| OLMo 2 1124 13B Instruct OLMo | 13.72B | Just fits | Q4_K_M | 14.88 GB | 29.7 t/s | 4k |
| Qwen3.5 9B Qwen | 9.65B | Runs great | Q8_0 | 11.4 GB | 39.3 t/s | 32k |
| Gemma 2 9B Instruct Gemma | 9.24B | Runs great | Q8_0 | 11.28 GB | 39.7 t/s | 8k |
| Qwen3.5 35B A3B Qwen | 35.95B (2.9B active) | Runs great | IQ3_XXS ! | 14.16 GB | 250.1 t/s | 16k |
| Granite 4.1 8B Granite | 8.79B | Runs great | Q8_0 | 10.8 GB | 41.7 t/s | 16k |
| Fanar 1 9B Instruct Fanar | 8.78B | Runs great | Q8_0 | 10.82 GB | 41.5 t/s | 4k |
| Internlm3 8B Instruct InternLM | 8.8B | Runs great | Q8_0 | 9.93 GB | 45.7 t/s | 32k |
| LFM2.5 8B A1B Liquid | 8.47B (1.57B active) | Runs great | Q8_0 | 9.48 GB | 215 t/s | 64k |
| Qwen3 32B Qwen | 32.76B | Just fits | IQ3_XXS ! | 14.58 GB | 30.3 t/s | 8k |
| Qwen2.5 32B Instruct Qwen | 32.76B | Just fits | IQ3_XXS ! | 14.58 GB | 30.3 t/s | 8k |
| Qwen2.5 VL 7B Instruct Qwen | 8.29B | Runs great | Q8_0 | 9.46 GB | 48 t/s | 64k |
| OTel 2.0 LLM 31B Instruct Other | 32.11B | Runs great | IQ3_XXS ! | 13.31 GB | 33.5 t/s | 32k |
| Qwen3 8B Qwen | 8.19B | Runs great | Q8_0 | 10.08 GB | 44.9 t/s | 32k |
| Granite 3.0 8B Instruct Granite | 8.17B | Runs great | Q8_0 | 10.18 GB | 44.4 t/s | 4k |
| T Lite Instruct 2.1 T-Lite | 8.19B | Runs great | Q8_0 | 10.08 GB | 44.9 t/s | 32k |
| Gemma 4 31B Instruct Gemma | 31.27B | Runs great | IQ3_XXS ! | 13.01 GB | 34.3 t/s | 32k |
| GLM 4.7 Flash GLM | 31.22B (3.66B active) | Runs great | IQ3_XXS ! | 12.26 GB | 241.6 t/s | 32k |
| Llama 3.1 8B Instruct Llama | 8.03B | Runs great | Q8_0 | 9.8 GB | 46.4 t/s | 32k |
| Apertus 8B Instruct 2509 Apertus | 8.05B | Runs great | Q8_0 | 9.82 GB | 46.3 t/s | 32k |
| Llama 3 Taiwan 8B Instruct Llama | 8.03B | Runs great | Q8_0 | 9.8 GB | 46.4 t/s | 8k |
| Gemma 4 E4B Instruct Gemma | 8B | Runs great | Q8_0 | 8.72 GB | 52.1 t/s | 128k |
| Qwen3 30B A3B Qwen | 30.53B (3.34B active) | Runs great | IQ3_XXS ! | 12.35 GB | 213.8 t/s | 32k |
| Qwen1.5 7B Qwen | 7.72B | Runs great | Q8_0 | 12.49 GB | 35.6 t/s | 8k |
| Granite 4.1 30B Granite | 28.87B | Runs great | IQ3_XXS ! | 13.14 GB | 33.8 t/s | 8k |
| Qwen2.5 7B Instruct Qwen | 7.62B | Runs great | Q8_0 | 8.8 GB | 52 t/s | 32k |
| Qwen3.5 27B Qwen | 27.78B | Runs great | IQ3_XXS ! | 12.81 GB | 34.9 t/s | 16k |
| OLMo 3 7B Instruct OLMo | 7.3B | Runs great | Q8_0 | 10.07 GB | 45 t/s | 64k |
| Mistral 7B Instruct V0.3 Mistral | 7.25B | Runs great | Q8_0 | 9.02 GB | 50.7 t/s | 32k |
| Mistral 7B Instruct V0.2 Mistral | 7.24B | Runs great | Q8_0 | 9.01 GB | 50.8 t/s | 32k |
| Falcon 7B Falcon | 7.22B | Runs great | Q8_0 | 12.46 GB | 35.8 t/s | 8k |
| DeepSeek Coder 7B Instruct V1.5 DeepSeek | 6.91B | Runs great | Q8_0 | 11.44 GB | 39.2 t/s | 4k |
| OLMoE 1B 7B 0125 Instruct OLMo | 6.92B (1.28B active) | Runs great | Q8_0 | 8.57 GB | 183 t/s | 4k |
| CodeLlama 7B Llama | 6.74B | Runs great | Q8_0 | 11.52 GB | 38.9 t/s | 8k |
| DeepSeek Coder 6.7B Instruct DeepSeek | 6.74B | Runs great | Q8_0 | 11.52 GB | 38.9 t/s | 8k |
| Gemma 4 E2B Instruct Gemma | 5.12B | Runs great | Q8_0 | 5.78 GB | 81.6 t/s | 128k |
| Qwen3.5 4B Qwen | 4.66B | Runs great | Q8_0 | 6.37 GB | 73.9 t/s | 64k |
| Agents A1 4B Other | 4.54B | Runs great | Q8_0 | 6.25 GB | 75.5 t/s | 64k |
| Gemma 3 4B Instruct Gemma | 4.3B | Runs great | Q8_0 | 5.31 GB | 91.1 t/s | 128k |
| Phi 3 Vision 128k Instruct Phi | 4.15B | Runs great | Q8_0 | 7.89 GB | 58.4 t/s | 16k |
| Qwen3 4B Qwen | 4.02B | Runs great | Q8_0 | 5.86 GB | 81.3 t/s | 32k |
| Phi 4 Mini Instruct Phi | 3.84B | Runs great | Q8_0 | 5.59 GB | 86.4 t/s | 64k |
| Phi 3 Mini 4k Instruct Phi | 3.82B | Runs great | Q8_0 | 5.32 GB | 91.6 t/s | 4k |
| Mixtral 8x7B Instruct V0.1 Mistral | 46.7B (12.88B active) | Runs great | IQ2_XXS ! | 13.05 GB | 101.4 t/s | 16k |
| PowerLM 3B PowerLM | 3.51B | Runs great | Q8_0 | 7.03 GB | 66 t/s | 4k |
| Granite 4.1 3B Granite | 3.4B | Runs great | Q8_0 | 4.75 GB | 104 t/s | 64k |
| PowerMoE 3B PowerLM | 3.37B (0.88B active) | Runs great | Q8_0 | 4.53 GB | 302.5 t/s | 4k |
| Phi 3.5 MoE Instruct Phi | 41.87B (6.64B active) | Runs great | IQ2_XXS ! | 11.89 GB | 160 t/s | 16k |
| Llama 3.2 3B Instruct Llama | 3.21B | Runs great | Q8_0 | 4.84 GB | 102.4 t/s | 64k |
| Qwen2.5 3B Instruct Qwen | 3.09B | Runs great | Q8_0 | 4.07 GB | 124.2 t/s | 32k |
| SmolLM3 3B Base SmolLM | 3.08B | Runs great | Q8_0 | 4.34 GB | 114.9 t/s | 64k |
| Starcoder2 3B StarCoder | 3.03B | Runs great | Q8_0 | 3.9 GB | 133.1 t/s | 16k |
| Phi 2 Phi | 2.78B | Runs great | Q8_0 | 6.01 GB | 79 t/s | 2k |
| LFM2.5 2.6B Liquid | 2.7B | Runs great | Q8_0 | 3.87 GB | 132.1 t/s | 128k |
| Gemma 2 2B Instruct Gemma | 2.61B | Runs great | Q8_0 | 3.73 GB | 138.8 t/s | 8k |
| GPT NeoX 20B GPT-NeoX | 20.74B | Runs great | IQ2_XXS ! | 14.2 GB | 31.4 t/s | 2k |
| Qwen3.5 2B Qwen | 2.27B | Runs great | Q8_0 | 3.35 GB | 158.2 t/s | 128k |
| OneRec 1.7B Other | 2.13B | Runs great | Q8_0 | 3.71 GB | 139 t/s | 32k |
| Qwen3 1.7B Qwen | 2.03B | Runs great | Q8_0 | 3.61 GB | 143.8 t/s | 32k |
| DeepSeek R1 Distill Qwen 1.5B Qwen | 1.78B | Runs great | Q8_0 | 2.57 GB | 221.7 t/s | 128k |
| Qwen3 1.7B Base Qwen | 1.72B | Runs great | Q8_0 | 3.3 GB | 160.9 t/s | 32k |
| SmolLM2 1.7B SmolLM | 1.71B | Runs great | Q8_0 | 3.92 GB | 129.9 t/s | 8k |
| Qwen2.5 1.5B Instruct Qwen | 1.54B | Runs great | Q8_0 | 2.44 GB | 238 t/s | 32k |
| Pythia 1.4B Pythia | 1.52B | Runs great | Q8_0 | 3.73 GB | 138.1 t/s | 2k |
| OLMo 2 0425 1B OLMo | 1.48B | Runs great | Q8_0 | 3.19 GB | 168.3 t/s | 4k |
| Llama 3.2 1B Instruct Llama | 1.24B | Runs great | Q8_0 | 2.2 GB | 280.8 t/s | 128k |
| LFM2.5 1.2B Instruct Liquid | 1.17B | Runs great | Q8_0 | 2.13 GB | 294.6 t/s | 64k |
| TinyLlama 1.1B Chat V1.0 Llama | 1.1B | Runs great | Q8_0 | 1.99 GB | 329 t/s | 2k |
| MiniCPM5 1B MiniCPM | 1.08B | Runs great | Q8_0 | 1.95 GB | 330.1 t/s | 128k |
| Gemma 3 1B Instruct Gemma | 1B | Runs great | Q8_0 | 1.71 GB | 399.2 t/s | 32k |
| Qwen3.5 0.8B Qwen | 0.87B | Runs great | Q8_0 | 1.9 GB | 335.6 t/s | 128k |
| Sarashina2.2 0.5B Instruct V0.1 Sarashina | 0.79B | Runs great | Q8_0 | 1.93 GB | 331.6 t/s | 8k |
| Qwen3 0.6B Qwen | 0.75B | Runs great | Q8_0 | 2.28 GB | 256.5 t/s | 32k |
| Qwen1.5 0.5B Chat Qwen | 0.62B | Runs great | Q8_0 | 2.03 GB | 304.2 t/s | 32k |
| Qwen3 0.6B Base Qwen | 0.6B | Runs great | Q8_0 | 2.13 GB | 282.4 t/s | 32k |
| Pythia 410m Pythia | 0.51B | Runs great | Q8_0 | 1.92 GB | 330.5 t/s | 2k |
| H2o Danube3 500m Chat Danube | 0.51B | Runs great | Q8_0 | 1.57 GB | 471.5 t/s | 8k |
Showing the 90 best results of 133. The remaining 43 need more memory than this device has, at any quantisation.
What it will not run
2 of the 133 architectures we track are out of reach here, even at two-bit precision. Another 32 run by splitting layers between the card and system RAM, which works but drops generation to single digits.
If you want more headroom
The next steps up in memory, in order. More memory changes which models load at all; more bandwidth changes how fast they answer, so the two are worth weighing separately.
| Device | Memory | Bandwidth | Models that fit | |
|---|---|---|---|---|
| Radeon RX 7900 XT | 20 GB | 800 GB/s | 100 | Check price |
| GeForce RTX 4090 | 24 GB | 1008 GB/s | 105 | Check price |
| GeForce RTX 3090 Ti | 24 GB | 1008 GB/s | 105 | Check price |
| Radeon RX 7900 XTX | 24 GB | 960 GB/s | 105 | Check price |
Price links go to an Amazon search for the model name and are affiliate links: if you buy through one, we earn a commission at no cost to you. We do not take payment for placement, and the ordering above is by memory capacity alone.
Other devices with 16 GB
Same capacity, different speed. Once a model fits, bandwidth is what separates these.