Radeon RX 7900 XTX for AI (2026): LLM Performance, ROCm and Price

Is the RX 7900 XTX good for AI? Yes, for running models. It has 24 GB of VRAM and 960 GB/s of memory bandwidth, within 5% of an RTX 4090, and since LLM token generation is bandwidth bound its speed ceiling on Llama 3.1 8B (a 4.9 GB file) is ~195 tok/s against ~205 on the 4090. On November 2, 2026 it was also the cheapest 24 GB card we could find: $1,175 used on eBay and $1,820.99 new (Amazon, via videocardprices.com), against $1,299 and up for a used RTX 3090. ROCm 7.14.0 lists it as officially supported, and Ollama, llama.cpp, vLLM, ComfyUI and PyTorch all work on RDNA 3. What you give up is the CUDA-only layer (TensorRT-LLM, ExLlamaV2, some ComfyUI nodes) and FP8 hardware.

This page is about the buying decision and the ROCm setup. For which models to run on the card, see best AI models for the RX 7900 XTX; for the head-to-head with NVIDIA's used 24 GB card, see RX 7900 XTX vs RTX 3090.

November 2026 update: (1) ROCm is on the 7.x line — 7.14.0 is the production release and 7.13.0 a technology preview; gfx1100 stays officially supported, and RDNA 4 (gfx1201, the RX 9070 series) is on the supported list too. (2) Prices rose with the DRAM/GDDR shortage — the 7900 XTX's lowest new price climbed from $1,049.99 in June to $1,820.99 by October 17 (videocardprices.com's tracker), while a used card was $1,175; NVIDIA's 24 GB+ options moved as much or more (used RTX 3090 $1,299-1,606, RTX 5090 $3,340 in stock). (3) FlashAttention's ROCm fork spans RDNA 3 and RDNA 4 (forward pass on gfx11; full backward only on CDNA). (4) Every speed on this page is now a labelled bandwidth ceiling, because we do not own this card and will not print numbers we did not measure.

This guide covers ROCm setup specifically for the 7900 XTX, building FlashAttention-2 for gfx1100, how it compares on paper with the RTX 3090, 4090, 5080 and 5090, multi-GPU configs, undervolting, the use cases where AMD wins and where NVIDIA still wins, and tuning recipes for Ollama, vLLM, llama.cpp, and ComfyUI.

Table of Contents

1. Why the 7900 XTX Matters in 2026
2. November 2026 Update: ROCm 7.14, RDNA 4, and Prices
3. Hardware Specs
4. vs RTX 3090 / 4090 / 5080 / 5090
5. 2026 Model Compatibility (Qwen3, Llama 4, Gemma 4, DeepSeek)
6. ROCm 7.x Setup (Step by Step)
7. FlashAttention-2 for gfx1100
8. Ollama on 7900 XTX
9. vLLM on 7900 XTX
10. llama.cpp on 7900 XTX
11. Stable Diffusion / Flux / ComfyUI
12. Speed Ceilings by Model
13. What 7900 XTX Cannot Do (vs NVIDIA)
14. Multi-GPU 7900 XTX Configurations
15. Undervolting and Power Limit
16. Cooling and Acoustics
17. Mixing with NVIDIA in One System
18. Buying Advice
19. Troubleshooting
20. FAQ

---

Why the 7900 XTX Matters in 2026

Three things changed in 2024-2025 that made the 7900 XTX viable for local AI, and a fourth landed in 2026:

1. ROCm 6.x stabilization — official RDNA 3 support landed in ROCm 5.7, became production-grade in 6.x.
2. Major framework support — Ollama, vLLM-ROCm, llama.cpp HIP, PyTorch ROCm, ComfyUI all ship working RDNA 3 paths.
3. FlashAttention-2 fork — AMD's port to gfx1100 closed the long-context performance gap.
4. ROCm 7.x (2026) — the stack moved to the 7.x line (7.14.0 production, 7.13.0 preview as of November 2026), adding RDNA 4 (gfx1201) support, official AMD-authored llama.cpp install docs, and pre-built vLLM ROCm wheels. Nothing in 7.x breaks the 7900 XTX; it stays a first-class gfx1100 target.

Pricing tells a more interesting story now. The 7900 XTX launched at $999 in December 2022. The 2026 DRAM/GDDR shortage pushed it well past that: videocardprices.com tracked the lowest new price at $1,049.99 on June 10, 2026 and $1,820.99 on October 17, with a used card at $1,175 on eBay when we read it on November 2. NVIDIA's 24 GB+ cards moved too: a used RTX 3090 averaged $1,568 on resaleprices.com (1,499-1,606 fair range, November 1), the RTX 4090 is out of production, and the cheapest in-stock 32 GB RTX 5090 we found was $3,340 at Walmart. So on November readings the 7900 XTX, new or used, is the cheapest way into 24 GB, for buyers who can tolerate the AMD ecosystem. Our GPU prices and the memory shortage page tracks the wider market. If you are weighing the second-hand route instead, read our used GPU AI buying guide.

---

November 2026 Update: ROCm 7.14, RDNA 4, and Prices

If you read an older version of this guide, here is exactly what is different by November 2026 and why none of it changes the verdict:

ROCm moved to the 7.x line. The stack that was "6.x" through most of 2025 is now 7.14.0 (production) with a 7.13.0 technology preview, per AMD's ROCm documentation on November 2, 2026. Practical effects for a 7900 XTX owner: nothing breaks — gfx1100 remains a first-class supported target — and you gain AMD-authored official llama.cpp install docs, pre-built vLLM ROCm wheels, and better Windows support via the dedicated "ROCm on Radeon and Ryzen" docs path. The one thing to watch is OS pinning: ROCm 7.x certifies the 7900 XTX only on Ubuntu 22.04.5 / 24.04.4 / RHEL 9.7 / RHEL 10.1, so install one of those point releases rather than a random interim build.

RDNA 4 arrived — but it is not a 7900 XTX replacement for AI. The RX 9070 / 9070 XT (gfx1201) are now officially supported in ROCm 7.2 and run Ollama, vLLM, and llama.cpp. However, they ship with 16 GB of VRAM, not 24 GB, and 640 GB/s of bandwidth (a ~130 tok/s ceiling on Llama 3.1 8B Q4, against ~195 on the 7900 XTX). The cheapest 9070 XT on videocardprices.com was $839.99 on November 2. For local AI specifically, VRAM capacity beats raw RDNA 4 efficiency: the 7900 XTX's 24 GB still fits models the 9070 XT cannot. If you want a single AMD card for LLMs in 2026, the 7900 XTX remains the pick. (If your interest is a unified-memory APU box instead, see the Strix Halo / Ryzen AI Max+ 395 guide.)

The price spike hit everyone. A 2026 DRAM/GDDR shortage inflated every GPU. The 7900 XTX's lowest new price went from $1,049.99 (June 10) to $1,820.99 (October 17), and a used RTX 3090 now asks $1,299-1,606. The RTX 4090 went out of production in 2024. The result: the 7900 XTX is still the only new 24 GB consumer GPU you can buy in stock, and a used one ($1,175 on eBay on November 2) undercuts a used 3090.

FlashAttention's ROCm fork widened. The CK backend now covers RDNA 3, RDNA 4, and CDNA (MI200/MI300/MI355). For the 7900 XTX, inference (forward pass) is fully supported; the backward pass is still missing on gfx11, which only matters if you fine-tune on the card (use the Triton backend or a rocWMMA build for training).

---

Hardware Specs

Compute capability vs RTX 4090: lower FP16 / BF16 raw throughput (122 vs 165 TFLOPS), similar memory bandwidth (960 vs 1,008 GB/s), no FP8 hardware (RTX 4090 has 660 FP8 TFLOPS).

---

vs RTX 3090 / 4090 / 5080 / 5090

Prices are the readings from November 2, 2026 named under the table, in a market inflated by the DRAM/GDDR shortage. Treat them as a snapshot. Speed rows are bandwidth ceilings (vendor bandwidth ÷ the 4.9 GB llama3.1:8b file), not measurements.

Prices: 7900 XTX from videocardprices.com (new as of October 17, used read November 2); RTX 3090 eBay listing via videocardprices.com and resaleprices.com fair range (November 1); RTX 5080 and 5090 cheapest in-stock listings at Walmart, November 2. Bandwidth from AMD's and NVIDIA's spec pages.

The 7900 XTX wins on $/VRAM: same 24 GB and nearly the same bandwidth as the 3090 and 4090 for less money on November readings. It loses on FP8/FP4 and Blackwell-class features, compute-bound work like image generation, and TensorRT-LLM / ExLlamaV2 access. With the RTX 4090 out of production, the 7900 XTX is the only new-in-box 24 GB GPU in stock. For the full cross-vendor breakdown including Intel Arc, see AMD vs NVIDIA vs Intel AI GPU.

---

2026 Model Compatibility (Qwen3, Llama 4, Gemma 4, DeepSeek)

The 24 GB on the 7900 XTX is the deciding factor for which 2026 models you can actually run. For ranked picks with exact Ollama tags, use the RX 7900 XTX model picks page; here is what fits comfortably (fully on-GPU) versus what needs offload, based on the current generation of open models. All figures assume FlashAttention-2 built for gfx1100 and a 4K-8K working context.

The practical headline for 2026: the 7900 XTX is ideal for the 8B-to-32B model tier — exactly where Qwen3, Gemma 4, and the Qwen3-Coder MoE deliver the best quality-per-VRAM. The new wave of large MoE models (Llama 4 Scout, full DeepSeek-R1) exceeds 24 GB and wants either heavy offload or a multi-GPU rig. For a continuously-updated view of what runs where, cross-reference our Ollama model RAM/VRAM table and the best Ollama models roundup. If image generation is your priority instead, the best GPU for image generation comparison covers where the 7900 XTX lands on Flux and SDXL.

Quant tip: the 7900 XTX has no FP8/FP4 hardware, so prefer GGUF Q4_K_M/Q5_K_M or AWQ-INT4 weights. FP8-only checkpoints (some 2026 Llama/DeepSeek releases) will run dequantized and slower — pick a GGUF or AWQ build of the same model where one exists.

---

ROCm 7.x Setup (Step by Step)

Ubuntu 22.04.5 / 24.04.4

ROCm 7.x certifies the 7900 XTX (gfx1100) specifically on Ubuntu 22.04.5, Ubuntu 24.04.4, RHEL 9.7, and RHEL 10.1 — match one of those point releases to avoid driver pain. Install the latest stable ROCm (7.14.0 on November 2, 2026; check repo.radeon.com for the current minor):

# Add AMD GPU repo and install ROCm 7.x (use the current version path from repo.radeon.com)
wget https://repo.radeon.com/amdgpu-install/latest/ubuntu/jammy/amdgpu-install_latest_all.deb
sudo apt install ./amdgpu-install_latest_all.deb
sudo amdgpu-install --usecase=rocm,hiplibsdk -y

# Add user to required groups
sudo usermod -aG render,video $USER

# Reboot
sudo reboot

# Verify (after reboot)
rocminfo | grep -A 2 "Agent"
# Should show: Name: gfx1100, Marketing Name: Radeon RX 7900 XTX
rocm-smi

For Ubuntu 24.04, replace jammy with noble in the URL. (The old ROCm 6.2 amdgpu-install_6.2.60200-1 package still works if you need to pin an older stack, but 7.x is recommended for new installs.)

Fedora 40+

sudo dnf install rocm-hip rocm-hip-devel rocm-comgr rocm-runtime
sudo usermod -aG render,video $USER
sudo reboot

Verify GPU works for inference

# Quick PyTorch sanity check (use the ROCm wheel matching your installed ROCm minor)
pip install torch --index-url https://download.pytorch.org/whl/rocm6.3
python -c "import torch; print(torch.cuda.is_available(), torch.cuda.get_device_name(0))"
# Expected: True Radeon RX 7900 XTX

The PyTorch ROCm wheel index lags the latest ROCm system version slightly — pick the highest rocmX.Y wheel available that is ≤ your installed ROCm. gfx1100 is supported by every recent wheel.

For full ROCm walkthrough including ROCm Vulkan path for unsupported GPUs, see AMD ROCm Setup for Local LLMs.

---

FlashAttention-2 for gfx1100

git clone https://github.com/ROCm/flash-attention
cd flash-attention
GPU_ARCHS="gfx1100" python setup.py install

Verify:

import flash_attn
print(flash_attn.__version__)

Once installed, llama.cpp -fa, vLLM, and PyTorch SDPA auto-use it.

What it changes, worked through for Llama 3.1 8B (32 layers, 8 KV heads, head dimension 128, so 128 KiB of f16 KV cache per token, from the model's published config). We have not timed FA on/off on this card, so the table shows memory, which is arithmetic, not speed:

FlashAttention matters most for long-context workloads on the 7900 XTX, and Ollama and llama.cpp need it on before they will quantize the KV cache.

---

Ollama on 7900 XTX

curl -fsSL https://ollama.com/install.sh | sh
ollama run llama3.1:8b

Auto-detects ROCm. Verify GPU is being used:

ollama run llama3.1:8b "hi"
# In another terminal:
rocm-smi
# Should show ~99% GPU utilization

For tuning Modelfile parameters, see Ollama Modelfile Guide.

---

vLLM on 7900 XTX

docker pull rocm/vllm:latest

docker run --device /dev/kfd --device /dev/dri \
    --group-add video --group-add render \
    --security-opt seccomp=unconfined \
    --shm-size 16G \
    -p 8000:8000 \
    rocm/vllm:latest \
    vllm serve casperhansen/llama-3.1-8b-instruct-awq \
    --quantization awq \
    --max-model-len 16384 \
    --gpu-memory-utilization 0.92

vLLM-ROCm on the 7900 XTX is functional but generally behind CUDA builds on equivalent NVIDIA hardware, because fewer kernels are tuned for gfx1100. Continuous batching and PagedAttention work. FP8 weights do not (no Ada-class FP8 hardware on RDNA 3); use AWQ-INT4 instead. As of ROCm 7.x, AMD ships pre-built vLLM ROCm wheels, so you no longer have to compile from source for gfx1100. For the full server-side configuration (batching, KV cache sizing, OpenAI-compatible API), see the vLLM complete setup guide.

---

llama.cpp on 7900 XTX

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp

HIPCXX="$(hipconfig -l)/clang" \
HIP_PATH="$(hipconfig -R)" \
cmake -B build \
    -DGGML_HIP=ON \
    -DAMDGPU_TARGETS=gfx1100 \
    -DCMAKE_BUILD_TYPE=Release
cmake --build build -j

./build/bin/llama-cli -m model.gguf -ngl 999 -fa

For multi-GPU 7900 XTX:

./build/bin/llama-cli -m model.gguf -ngl 999 -fa --tensor-split 24,24

---

Stable Diffusion / Flux / ComfyUI

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
python3.11 -m venv venv && source venv/bin/activate
pip install torch torchvision --index-url https://download.pytorch.org/whl/rocm6.3
pip install -r requirements.txt
python main.py --listen 0.0.0.0

For Automatic1111: clone the AMD-friendly fork (lshqqytiger/stable-diffusion-webui-amdgpu) which has ROCm-specific fixes.

What fits (1024x1024). Image generation is compute bound, the RTX 4090 has more FP16 throughput (165 vs 122 TFLOPS, vendor specs), so expect it to finish each image sooner; we do not publish seconds-per-image figures we have not measured:

Flux Dev BF16 doesn't fit in 24 GB (needs ~24 GB just for the model); use FP8 or GGUF Q8 quants. Same applies to RTX 4090.

For ControlNet / IPAdapter / video generation: most workflows work, but a few custom nodes that depend on xformers attention or specific CUDA kernels won't. See ComfyUI Complete Guide.

---

Speed Ceilings by Model

We do not own a 7900 XTX, so this section prints no measured rates. Each LLM figure is a bandwidth ceiling: AMD's 960 GB/s divided by the file size of the default Ollama tag (Q4_K_M), read from the Ollama registry on November 2, 2026. Real engines land below the ceiling: on our own Apple M3 Pro (Ollama 0.40.13), 7-8B models reached 84-90% of it, and on a rented NVIDIA L4 (24 GB, 300 GB/s) they reached 77-81% (every row is on /benchmarks), while DatabaseMart's published RTX 5090 Ollama runs reached about half on 32B models and less on an 8B, where per-token overhead dominates on a fast card.

LLM inference

The MoE pattern (Qwen3-Coder, Gemma 4) is a notable 2026 win for the 7900 XTX: because only a few billion parameters activate per token, throughput stays high even though total model size is large — exactly the kind of architecture that plays to a 24 GB card. Dense 70B models remain partial-offload territory on a single card.

Image generation

Embeddings

If you own the card, Ollama prints the real rate with ollama run qwen3:14b --verbose (the "eval rate" line), and llama-bench gives a repeatable table.

---

What 7900 XTX Cannot Do (vs NVIDIA)

Honest list:

1. No FP8 — FP8-only checkpoints (some newer Llama / DeepSeek releases) run dequantized and slower than on cards with FP8 hardware.
2. No FlashAttention-3 — it targets Hopper (H100); no consumer card, AMD or NVIDIA, runs it.
3. No NVLink — multi-GPU bound by PCIe (~32 GB/s) vs NVIDIA NVLink consumer (3090: ~112 GB/s).
4. No TensorRT-LLM — NVIDIA only.
5. No ExLlamaV2 — CUDA only; you cannot use the fastest single-GPU INT4 inference.
6. Narrower image-gen ecosystem — some xformers / CUDA-kernel-dependent custom nodes don't work.
7. No Stable Video Diffusion at full speed — kernels less optimized.
8. Slower for fine-tuning — bitsandbytes-rocm fork lags upstream.

If any of those matter for your specific workload, NVIDIA is the right choice despite the price premium.

---

Multi-GPU 7900 XTX Configurations

2x 7900 XTX (48 GB total)

PCIe-only (no NVLink/Infinity Fabric link on consumer Radeon). Tensor parallel via vLLM works:

vllm serve casperhansen/llama-3.1-70b-instruct-awq \
    --quantization awq \
    --tensor-parallel-size 2 \
    --max-model-len 16384

Tensor parallelism splits each layer across both cards, so both read weights at once; PCIe 4.0 x16 (~32 GB/s each way) is the bottleneck for the all-reduce traffic. Measure your own speedup with vLLM's benchmark script before buying a second card for speed alone; for capacity, the second card is what lets a 70B fit at all.

7900 XTX + 7900 XTX + Radeon Pro W7900 (96 GB total)

For 70B at Q8 (about 75 GB) fully on GPUs: tensor split [24, 24, 48]. (70B at BF16 is about 141 GB and does not fit in 96 GB.) llama.cpp --tensor-split handles asymmetric splits. Run dual-card vLLM if W7900 is in a separate node, otherwise three-way TP works in vLLM-ROCm with caveats.

For broader multi-GPU patterns including mixed AMD/NVIDIA via GPUStack.

---

Undervolting and Power Limit

Stock TBP is 355 W. Sweet spot for inference: 290-310 W.

Linux

# Power cap to 290W
sudo rocm-smi --setpoweroverdrive 290

# Lock GPU clock for stable inference latency
sudo rocm-smi --setperflevel high

# Persistent at boot via systemd
sudo tee /etc/systemd/system/rocm-power-limit.service <<'EOF'
[Unit]
Description=Set RX 7900 XTX power limit
After=multi-user.target

[Service]
Type=oneshot
ExecStart=/usr/bin/rocm-smi --setpoweroverdrive 290
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target
EOF
sudo systemctl enable rocm-power-limit

For undervolting via the curve, use LACT or CoreCtrl GUI tools.

Windows

AMD Software → Performance → Tuning → Manual → set Power Limit to -15% to -20%. Apply and run llama-bench for an hour to verify stability.

Because LLM decode is bandwidth bound, a 290 W cap usually costs little throughput while running cooler and quieter; confirm on your own card by running llama-bench at stock and capped.

---

Cooling and Acoustics

The 7900 XTX runs hot. Reference (MBA) coolers are loud at sustained load; AIB triple-fan models are much quieter.

For 24/7 inference rigs:

• Power cap to 290W (above)
• Custom fan curve via LACT — start ramp at 50°C, full speed at 85°C
• Case airflow: front intake + top/rear exhaust, no front blockers
• Open-frame mining-style chassis works well for multi-GPU

See the case and cooling notes in the AI PC build guide for whole-system airflow.

---

Mixing with NVIDIA in One System

You can install a 7900 XTX and an RTX 4090 in the same machine. Both drivers coexist on Linux (with care). Use cases:

• 7900 XTX for LLM inference + 4090 for image gen / fine-tuning
• 7900 XTX for embeddings + 4090 for hot-path chat
• Both for independent users / containers

Tools that route across vendors: GPUStack, LiteLLM with separate Ollama / vLLM endpoints per GPU. There is no cross-vendor tensor parallel — they are independent compute resources.

---

Buying Advice

Buy the 7900 XTX if:

• You want the best $/perf for new 24 GB GPUs
• You run mainstream LLMs (Llama, Qwen, Mistral, Gemma) at Q4-Q5
• You do some image generation but it's not the primary use
• You're on Linux and comfortable with open-source tooling
• You don't need FP8 / TensorRT-LLM / ExLlamaV2

Buy used RTX 3090 instead if:

• You want NVLink for multi-GPU 70B
• You do heavy fine-tuning (NVIDIA ecosystem matters)
• You're in a market with abundant used 3090s ($1,299-1,606 on November 2026 readings, the cheapest CUDA + NVLink 24 GB path)

Buy RTX 4090 (used; out of production) instead if:

• You need FP8 (newer model checkpoints, vLLM FP8 throughput)
• Image generation is your primary workload
• You want maximum single-card LLM speed

Buy 5090 / W7900 / professional cards if:

• You need 32+ GB VRAM in a single card
• You want the latest features (FP4 on 5090, FP8 on Hopper)

---

Troubleshooting

---

FAQ

See answers to common Radeon RX 7900 XTX questions below.

---

Sources: AMD Radeon RX 7900 XTX product page (24 GB, 960 GB/s, 355 W) | ROCm system requirements (ROCm 7.14.0, RX 7900 XTX gfx1100 supported; read November 2, 2026) | Ollama GPU support | NVIDIA GeForce comparison (RTX 3090, 5080, 5090 bandwidth) | videocardprices.com RX 7900 XTX (new $1,820.99 as of October 17, used $1,175, read November 2, 2026) | Ollama library (model file sizes). No figure on this page is our own measurement of this card.

Related guides:

• AMD ROCm Setup for Local LLMs
• AMD vs NVIDIA vs Intel AI GPU
• Best GPUs for AI 2025
• CUDA Optimization for Local LLMs
• AI PC Build Guide
• Best AI Models for the RX 7900 XTX
• RX 7900 XTX vs RTX 3090
• GPUStack Setup Guide
• ComfyUI Complete Guide