LLMBOTTLENECK.COM

Can the AMD Radeon RX 7900 XT run Qwen2.5-32B-Instruct?

Qwen2.5-32B-Instruct · AMD Radeon RX 7900 XT · 8,192 ctx · whole model residentQ3_K_M sizepublished dataSpeedestimate

IT FITS

Yes. 2 of 9 formats evaluated fit in 20 GB, at 8,192 tokens. 8 of them are sized from a published file; the rest are rebuilt from the pinned architecture. The best quality that fits is Q3_K_M, needing 18.9 GB and running at an estimated 35 tokens per second.

Fitting in memory is not the same as loading. Whether the runtime and version you have supports this architecture and format on this machine has not been tested here.

Find the AMD Radeon RX 7900 XT: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

2/9formats that fit
20 GBdevice memory
18.9 GBneeded at best quality
~35tokens per second, estimatedFaster than you read
800GB/s bandwidth
Run it in the cloud

Run Qwen2.5-32B-Instruct properly, for cents an hour

The AMD Radeon RX 7900 XT only holds it at Q3_K_M, at an estimated ~35 tokens per second. A rented card runs a better format at full speed, billed by the second.

Your card Or keep Q4_K_M on your own card: with about 4 GB of it in system RAM, the AMD Radeon RX 7900 XT runs it at ~12 tokens per second (about reading pace) — free, if you have the RAM. A rented card below holds all of it on the GPU. Size the offload →

Prices · 18:30 UTC, 22 Sept
Cheapest24 GB

2× RTX 3060

$0.11per hour

Speed
~22 tok/sAbout reading pace
Per million tokens
$1.33

Vast.ai · Marketplace · 98.0% reliable

Rent on Vast.ai

Lowest price per hour

Compare all 26 rentable configurations →

First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch Q4_K_M on it (llama.cpp)
Container image
ghcr.io/ggml-org/llama.cpp:server-cuda
Start command / arguments
-hf Qwen/Qwen2.5-32B-Instruct-GGUF:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080
Expose port
8080 — an OpenAI-compatible API at /v1

Starting from a saved llama.cpp template instead? Leave the arguments empty and set these variables — only the first two change between models:

LLAMA_ARG_HF_REPO=Qwen/Qwen2.5-32B-Instruct-GGUF:Q4_K_M
LLAMA_ARG_CTX_SIZE=8192
LLAMA_ARG_N_GPU_LAYERS=999
LLAMA_ARG_HOST=0.0.0.0
LLAMA_ARG_PORT=8080

Already have the runtime? llama-server -hf Qwen/Qwen2.5-32B-Instruct-GGUF:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080. Weights: Qwen/Qwen2.5-32B-Instruct-GGUF. The CUDA image is for NVIDIA cards; an AMD machine needs llama.cpp's ROCm build.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Runs well on your AMD Radeon RX 7900 XT instead

The models nearest to Qwen2.5-32B-Instruct — same lab first, then closest in size — that the AMD Radeon RX 7900 XT holds whole at the standard format and runs at a usable speed. Nearest in size is not the same as equally good; compare them on the task you care about.

How long a conversation. At Q3_K_M, this pair holds 12,454 tokens of context — roughly 9,341 words — needing 20.0 GB. Past that the AMD Radeon RX 7900 XT runs out of memory, not the model out of context, which would allow 32,768. This is the wall a long chat hits after it has already loaded fine. The nearest round setting below it is 8,192.

Every format evaluated

8 of them are sized from a published file; the rest are rebuilt from the pinned architecture. The Basis column says which is which for each row. Decode and first-token figures are calibrated estimates, not runs on this card.

FormatNeedsOf 20 GBFitsDecodeFirst tokenBasis
FP1668.5 GB342%short by 48.5 GBdoes not run—reconstructed size
Q8_037.8 GB189%short by 17.8 GBdoes not run—published data
Q6_K29.8 GB149%short by 9.83 GBdoes not run—published data
Q5_K_M26.2 GB131%short by 6.21 GBdoes not run—published data
Q5_025.6 GB128%short by 5.59 GBdoes not run—published data
Q4_K_M22.8 GB114%short by 2.80 GBdoes not run—published data
Q4_021.6 GB108%short by 1.59 GBdoes not run—published data
Q3_K_M18.9 GB94%yes~35 tok/sno comparable peakpublished data
Q2_K15.3 GB76%yes~45 tok/sno comparable peakpublished data

Sized at 8,192 tokens of context with the whole model resident — weights, the KV cache and the runtime reserve, offload off. Speed is only quoted for a format that fits: a rate for a configuration that cannot load is not a fact about anything.

Why the first-token figure is the same on every row

First token is modelled from the arithmetic the prompt requires, and that count does not change with the weight format — which is why it reads the same on every row. Real prefill does vary by format, because a quantized matmul is a different kernel; this model does not capture that, and the figure should be read as an order of magnitude rather than a ranking between formats.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.
estimate
Calculated from sourced inputs by a stated method; an estimate, not a measurement.
Change anything

This page fixes the context at 8,192 tokens and one device. Batch, concurrent users, KV-cache format, clusters and rental cost are all in the calculator, already set to this pairing.

Open in the calculator →Everything the AMD Radeon RX 7900 XT runsBest models for 20 GB

Where these numbers come from

Qwen2.5-32B-Instruct’s architecture is read from its publisher’s own config.json at a pinned revision, and the AMD Radeon RX 7900 XT’s 20 GB and 800 GB/s come from the manufacturer’s specification. How far each figure can be trusted is published, per format and worst case included, on the accuracy scorecard.

Building this into your own product? A free API key returns exactly these figures, and the widget puts this answer on a product page with one script tag.

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"qwen-qwen2-5-32b-instruct","quantization":"Q3_K_M","context":8192,"hardware":"amd-radeon-rx-7900-xt"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="qwen-qwen2-5-32b-instruct" data-quantization="Q3_K_M"
  data-hardware="amd-radeon-rx-7900-xt" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

llmbottleneck
catalogue 2026-10-03models 327devices 135