RX 7900 XT
An RX 7900 XT runs 19 of the 34 open models tracked here at Q4_K_M with 8k context, computed on 16 October 2026. The largest is Muse Glimmer 30B, needing 18.7 GB and generating about 32 tokens per second.
- Memory
- 20 GB VRAM
- Bandwidth
- 800 GB/s
- Board power
- 315 W
- Released
- 2022
- Availability
- Sold new
Open the RX 7900 XT in the LLM checker
What an RX 7900 XT runs, model by model
RX 7900 XT (20 GB) addresses 20 GB at 800 GB/s. Every model below is at Q4_K_M with 8k of context, the setting most people actually use. Once the 5% safety margin comes off, the card has 19 GB to spend: memory decides what loads, bandwidth how fast it answers. Both are estimates from stated formulas, and the VRAM calculator shows the working.
On this card a model fits at or under 19 GB, and above 17 GB it is tight: it loads, but there is little left for the conversation, and a desktop on the same 20 GB can push it over. Tokens per second are single-stream estimates from memory bandwidth; ≤ marks a ceiling real runtimes fall well below.
Computed 16 October 2026The model to actually run on it
Mistral Small 3.1 24B is the best use of this card: 16.3 GB of the 20 GB available, an estimated 40 tokens per second, comfortable for chat, and 23,552 tokens of context still available.
How this is chosen, since no benchmark is quoted: models are compared by size, a mixture-of-experts model counting at the geometric mean of its total and active parameters (a rule of thumb, not a measurement). The pick is the biggest class that fits with memory to spare and answers at 25 tokens per second or better, and within that class the most recent general-purpose release; coding and reasoning specialists are listed above but not recommended by default.
Muse Glimmer 30B is the largest model the card will hold, at 18.7 GB, with room for 32,768 tokens of context even at 94% of memory, because its attention design caches so little per token.
If quality matters more than parameter count, Gemma 4 12B fits at Q8_0 in about 14.2 GB. Q8 is near-lossless; a smaller model at Q8 often beats a larger one squeezed into Q3.
How much context actually fits
Model size is the question people ask; context is the one that bites. The KV cache grows linearly with the conversation, so a model that loads comfortably can still run out of memory halfway through a long document. On 20 GB, the context you get is whatever the weights leave behind, which is why two cards that run the same model can feel completely different in use. These are the largest contexts this card holds, at Q4_K_M weights.
| Model | Max context, FP16 KV | With Q8 KV cache |
|---|---|---|
| Mistral Small 3.1 24B | 23,552 tokens | 48,128 tokens |
| Muse Glimmer 30B | 32,768 tokens | 71,680 tokens |
| Qwen3.6 27B | 23,552 tokens | 48,128 tokens |
| Gemma 3 27B | 18,432 tokens | 41,984 tokens |
Quantising the KV cache to Q8 roughly doubles what you can hold, at a quality cost most people never notice: Mistral Small 3.1 24B on this card goes from 23,552 to 48,128 tokens. Capped at 128k tokens, or at the model's own native window where that is smaller, marked "model limit": free memory beyond that point buys nothing. Rows that hold far more context than their size suggests are hybrid, sliding-window or latent-attention models, which cache only a fraction of what a standard transformer does; each model page shows the working.
Where this card stops
The first model out of reach is Granite 4.2 30B: about 20.3 GB at Q4_K_M and 8k context, against 20 GB of usable memory. Dropping to Q3_K_M would need about 17 GB, which fits, though Q3 loses enough quality that a smaller model at Q4 is usually the better trade. Offloading the remainder to system RAM works and is ten to fifty times slower; it is a way to see a model run, not a way to use one.
Buy or rent
Worth it if 20 GB at 800 GB/s covers what you run and you are comfortable checking each tool for ROCm. Against an RX 9070 XT it holds more and reads faster; the 9070 XT is newer and runs FP8. If you need 24 GB, an RX 7900 XTX or a used RTX 3090 is the step up, and renting by the hour covers the occasional larger model.
Retail prices drift, and a figure written today would be wrong within a month, so this page does not carry one. Put the figure you are actually looking at, with your electricity rate and your real monthly hours, into the build vs rent calculator, which already knows this card draws 315 W under load. It returns the month owning becomes cheaper, or tells you it never does. Hours per month decides it far more often than the price does.
The honest take
Twenty gigabytes at 800 GB/s: the XTX's smaller sibling, with more memory than any 16 GB card.
It sits between the 16 GB cards and the 24 GB tier: four gigabytes more than an RX 9070 XT, four fewer than an RX 7900 XTX. At 800 GB/s it reads memory a quarter faster than the 9070 XT and a sixth slower than the XTX. It is RDNA 3, so no FP8, but unlike the RX 9000 cards it is on Ollama's ROCm list for Windows as well as Linux.
What it is good at
- 20 GB at 800 GB/s: more room than any 16 GB card, and a quarter faster than an RX 9070 XT.
- On AMD's ROCm 10.1 matrix for Linux and Windows 11, and on Ollama's ROCm lists for both.
- AMD publishes validated llama.cpp builds for RDNA 3 and RDNA 4 cards on Linux.
What it is not
- No FP8: AMD documents FP8 for RDNA 4 only, so FP8 checkpoints are converted rather than run natively.
- Twenty gigabytes is four short of the 24 GB tier, so the largest models a 3090 or 7900 XTX holds do not fit.
- 315 W of board power on a two-and-a-half-slot card, and no CUDA.
The thing people get wrong: AMD's page prints two bandwidth figures. The much larger "effective" one counts the on-chip cache, which model weights do not fit in; for language models, 800 GB/s is the number that matters.
Compared with the alternatives
- RX 7900 XTX: 24 GB · 960 GB/s · best fit Muse Glimmer 30B at ~39 tok/s. AMD's 24 GB card: 3090-class memory and bandwidth, without CUDA.
- RX 9070 XT: 16 GB · 640 GB/s · best fit Gemma 4 12B at ~64 tok/s. AMD's RDNA 4 gaming card: 16 GB at 640 GB/s, with ROCm on Linux and Windows, and no CUDA.
- RTX 3090: 24 GB · 936 GB/s · best fit Muse Glimmer 30B at ~38 tok/s. The value benchmark for local models, and it has been for years.
All 39 cards are compared side by side on the GPU index, and the same numbers from the model's point of view are on each model fit table. To check one specific pairing rather than read a table, use the can-I-run-it checker.
Questions
- Can an RX 7900 XT run a 70B model?
- Not at Q4_K_M in one card. Llama 3.3 70B needs about 45.8 GB at 8k context and this card can address 20 GB. Your options are a smaller model, a harsher quantisation with very little context, splitting across two cards, or renting a larger one for the hours you need it.
- What is the best model to run on an RX 7900 XT?
- Mistral Small 3.1 24B. At Q4_K_M and 8k context it needs about 16.3 GB of the 20 GB available and generates an estimated 40 tokens per second, which is comfortable for chat. If quality matters more than size, Gemma 4 12B fits at Q8 in about 14.2 GB.
- How many tokens per second does an RX 7900 XT generate?
- It depends entirely on the size of the model, because generating a token means reading every active weight out of memory. At 800 GB/s and 70% efficiency, this card produces an estimated 121 tokens per second on an 8B model at Q4, and about 32 on the largest model it holds, Muse Glimmer 30B. Reading speed is roughly 5 to 8 words per second, so anything above 25 feels immediate. Treat 121 as a ceiling on 800 GB/s rather than a measurement; the higher a figure, the further real runtimes fall below it.
- Is 20 GB enough for running LLMs locally?
- It runs 19 of the 34 models tracked here at Q4_K_M with 8k of context, up to 29.8B parameters. The honest test is not the model list but the context: Mistral Small 3.1 24B on this card holds about 23,552 tokens before memory runs out.
- RX 7900 XT or RX 7900 XTX for local models?
- The RX 7900 XTX holds more: 24 GB against 20 GB, so it runs 28 of these models to this card's 19. On speed, the RX 7900 XTX is faster: 960 GB/s against 800 GB/s, and bandwidth is what sets chat speed.
- RX 7900 XT or RX 9070 XT for local LLMs?
- The 7900 XT holds 20 GB against 16 GB and reads it at 800 GB/s against 640 GB/s, so it fits more and has a speed ceiling a quarter higher. The 9070 XT is the newer RDNA 4 card: it runs FP8, which the 7900 XT does not, and draws 304 W against 315 W. For chat with the largest model that fits, the 7900 XT; for newer formats, the 9070 XT.
- Does the RX 7900 XT work with ROCm?
- Yes. AMD's ROCm 10.1 compatibility matrix lists it (as gfx1100) for Ubuntu, RHEL and Windows 11, and Ollama lists it for ROCm on both Linux and Windows. Tools that only ship CUDA builds still need a ROCm port.
- Memory: weights (parameters × bytes per parameter) + KV cache (2 × layers × KV heads × head dim × context × bytes for a standard transformer; sliding-window, hybrid and latent-attention models counted as they actually cache) + 0.5 GB runtime + 4% of weights. Full derivation in the VRAM calculator.
- Speed: 70% × bandwidth ÷ active weight bytes, single stream, no batching, prompt processing excluded. A ceiling, not a measurement: the 7 of 19 figures marked ≤ are where real runtimes fall furthest below the number, because so few weights are read per token. See the tokens-per-second estimator.
- Card specification (manufacturer figures): 20 GB, 800 GB/s, 315 W; see AMD's own page for this card.
- Model architecture values come from each model's published config; the self-hosted LLM guide explains what each one changes.