What can AMD Radeon RX 7900 XTX run for long context?

Build: RX 7900 XTX + Ryzen 9 7950X + 64GB DDR5 (Linux/ROCm)

Memory: 24 GB VRAM + 64 GB system RAM
Runner: llama.cpp (ROCm)
AnyChatCodingAgentsReasoningVisionLong contextCreative

Runs comfortably
84 models

Ranked by fit for long context use case + predicted speed. Click a row for VRAM breakdown.

#1Gemma 4 E4B (Effective 4B)
4B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 8.6 GBHeadroom: 15.4 GBTTFT: instant
ollama run gemma4:e4b
124
tok/s
Estimated
Weights
4.40 GB
KV cache
2.00 GB
Activations
0.23 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~83 ms (instant)
Model details →
#2Phi-3.5 Mini Instruct
3.8B
phi
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 8.2 GBHeadroom: 15.8 GBTTFT: instant
ollama run phi3.5:3.8b
131
tok/s
Estimated
Weights
4.10 GB
KV cache
1.90 GB
Activations
0.21 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~79 ms (instant)
Model details →
#3Codestral Mamba 7B
7B
mistral
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 9.9 GBHeadroom: 14.1 GBTTFT: fast
125
tok/s
Estimated
Weights
4.20 GB
KV cache
3.50 GB
Activations
0.22 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~146 ms (fast)
Model details →
#4Falcon Mamba 7B
7B
falcon
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 9.9 GBHeadroom: 14.1 GBTTFT: fast
125
tok/s
Estimated
Weights
4.20 GB
KV cache
3.50 GB
Activations
0.22 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~146 ms (fast)
Model details →
#5Command R7B (12-2024)
8B
command-r
Quant: Q4_K_MContext: 8,192VRAM: 10.6 GBHeadroom: 13.4 GBTTFT: fast
109
tok/s
Estimated
Weights
4.40 GB
KV cache
4.00 GB
Activations
0.23 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~167 ms (fast)
Model details →
#6Ministral 8B Instruct
8B
mistral
Quant: Q4_K_MContext: 8,192VRAM: 11.3 GBHeadroom: 12.7 GBTTFT: fast
109
tok/s
Estimated
Weights
5.00 GB
KV cache
4.00 GB
Activations
0.26 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~167 ms (fast)
Model details →
#7Qwen 2.5 7B Instruct
7B
qwen
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 11.0 GBHeadroom: 13.0 GBTTFT: fast
ollama run qwen2.5:7b
71
tok/s
Estimated
Weights
8.10 GB
KV cache
0.47 GB
Activations
0.41 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~146 ms (fast)
Model details →
#8Mistral Nemo 12B Instruct
12B
mistral
Commercial OK
Quant: Q5_K_MContext: 8,192VRAM: 17.1 GBHeadroom: 6.9 GBTTFT: fast
ollama run mistral-nemo:12b
64
tok/s
Estimated
Weights
8.70 GB
KV cache
6.00 GB
Activations
0.44 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~250 ms (fast)
Model details →
#9Qwen 3 14B
14B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 17.8 GBHeadroom: 6.2 GBTTFT: fast
ollama run qwen3:14b
62
tok/s
Estimated
Weights
8.40 GB
KV cache
7.00 GB
Activations
0.43 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~292 ms (fast)
Model details →
#10Qwen 3 8B
8B
qwen
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 14.6 GBHeadroom: 9.4 GBTTFT: fast
ollama run qwen3:8b
62
tok/s
Estimated
Weights
8.20 GB
KV cache
4.00 GB
Activations
0.42 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~167 ms (fast)
Model details →
#11InternLM 2.5 7B Chat
7B
internlm
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 10.1 GBHeadroom: 13.9 GBTTFT: fast
125
tok/s
Estimated
Weights
4.40 GB
KV cache
3.50 GB
Activations
0.23 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~146 ms (fast)
Model details →
#12DeepSeek V2 Lite Chat
15.7B
deepseek
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 18.9 GBHeadroom: 5.1 GBTTFT: instant
364
tok/s
Estimated
Weights
8.60 GB
KV cache
7.85 GB
Activations
0.44 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~50 ms (instant)
Model details →

Runs with tradeoffs
36 models

Tight VRAM, partial CPU offload, or context-limited.

Qwen 2.5 14B Instruct
14B
qwen
Commercial OK
Quant: Q5_K_MContext: 8,192VRAM: 20.0 GBHeadroom: 4.0 GBTTFT: fast
  • • Tight VRAM fit — only 4.0 GB headroom left for context growth
ollama run qwen2.5:14b
55
tok/s
Estimated
Weights
10.50 GB
KV cache
7.00 GB
Activations
0.53 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~292 ms (fast)
Model details →
Gemma 4 Turkish 26B (4B active)
26B
gemma
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 20.3 GBHeadroom: 3.7 GBTTFT: noticeable
  • • Tight VRAM fit — only 3.7 GB headroom left for context growth
34
tok/s
Estimated
Weights
14.30 GB
KV cache
3.25 GB
Activations
0.72 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~542 ms (noticeable)
Model details →
Gemma 3 27B
27B
gemma
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 22.2 GBHeadroom: 1.8 GBTTFT: noticeable
  • • Tight VRAM fit — only 1.8 GB headroom left for context growth
ollama run gemma3:27b
32
tok/s
Estimated
Weights
16.00 GB
KV cache
3.38 GB
Activations
0.80 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~563 ms (noticeable)
Model details →
Qwen3.8 27B
27B
qwen
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 22.7 GBHeadroom: 1.3 GBTTFT: noticeable
  • • Tight VRAM fit — only 1.3 GB headroom left for context growth
ollama run qwen3.8:27b
32
tok/s
Estimated
Weights
16.50 GB
KV cache
3.38 GB
Activations
0.83 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~563 ms (noticeable)
Model details →
Jamba 1.5 Mini
52B
other
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 59.5 GBHeadroom: 2.9 GBTTFT: fast
  • • Partial CPU offload: ~60% of layers run on CPU
  • • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
8
tok/s
Estimated
Weights
30.00 GB
KV cache
26.00 GB
Activations
1.51 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~250 ms (fast)
Model details →
Command R 35B
35B
command-r
Quant: Q4_K_MContext: 8,192VRAM: 41.6 GBHeadroom: 20.8 GBTTFT: noticeable
  • • Partial CPU offload: ~42% of layers run on CPU
  • • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
ollama run command-r:35b
4
tok/s
Estimated
Weights
21.00 GB
KV cache
17.50 GB
Activations
1.06 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~730 ms (noticeable)
Model details →
Llama 3.2 11B Vision Instruct
11B
llama
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 20.6 GBHeadroom: 3.4 GBTTFT: fast
  • • Tight VRAM fit — only 3.4 GB headroom left for context growth
ollama run llama3.2-vision:11b
45
tok/s
Estimated
Weights
12.50 GB
KV cache
5.50 GB
Activations
0.63 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~229 ms (fast)
Model details →
MedGemma 27B
27B
gemma
Quant: Q4_K_MContext: 2,048VRAM: 22.2 GBHeadroom: 1.8 GBTTFT: noticeable
  • • Tight VRAM fit — only 1.8 GB headroom left for context growth
32
tok/s
Estimated
Weights
16.00 GB
KV cache
3.38 GB
Activations
0.80 GB
Runtime
2.00 GB
Time to first token (prefill, 512-token prompt): ~563 ms (noticeable)
Model details →

What if you upgraded?

Hypothetical scenarios. We re-ran the compatibility engine for each.

+32 GB system RAM

Check the current price

Adds 32 GB to your CPU-offload working set. Helps when models don't quite fit in VRAM.

Unlocks: 28 new comfortable, 40 new tradeoff

  • • Qwen 3 0.6B
  • • Qwen 3 1.7B
  • • Gemma 3 270M
  • • SmolLM2 135M Instruct
Shop this upgrade↗

Upgrade to AMD Instinct MI210

Launch MSRP $8,500 (2022). Check the current price.

64 GB VRAM (vs your 24 GB) plus a bandwidth jump from ~960 GB/s to ~? GB/s.

Unlocks: 61 new comfortable

  • • Qwen 3 0.6B
  • • Qwen 3 30B-A3B
  • • Llama 3.3 70B Instruct
  • • Qwen 2.5 Coder 32B Instruct
Shop this upgrade↗

Add a second AMD Radeon RX 7900 XTX

Launch MSRP $999 (2022). Check the current price.

Tensor parallelism splits the model across both cards, effectively doubling VRAM. Bandwidth doesn't double — runs ~1.5× the single-card speed in practice.

Unlocks: 52 new comfortable

  • • Qwen 3 0.6B
  • • Qwen 3 30B-A3B
  • • Qwen 2.5 Coder 32B Instruct
  • • Qwen 3 32B
Shop this upgrade↗

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

Won't run
top 5 popular models

Need more memory than you have. Shown for orientation.

DeepSeek V4 Pro (1.6T MoE)
1600B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (24 GB) + 60% of system RAM (38 GB) combined.

—
Qwen 3 235B-A22B
235B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (24 GB) + 60% of system RAM (38 GB) combined.

—
DeepSeek R1 (671B reasoning)
671B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (24 GB) + 60% of system RAM (38 GB) combined.

—
Llama 4 Scout
109B
llama
Commercial OK

Even with CPU offload, needs more memory than your VRAM (24 GB) + 60% of system RAM (38 GB) combined.

—
DeepSeek V4 Flash (284B MoE)
284B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (24 GB) + 60% of system RAM (38 GB) combined.

—
Cloud GPU rental

Not enough VRAM for these models? Rent a GPU by the hour

The models above need more memory than 24 GB VRAM + 64 GB system RAM. You can rent a GPU with enough VRAM by the hour to run one of them, or to test it before you buy hardware. For daily use, owning a card usually costs less over time.

RunPod

billed per GPU-hour

Hourly GPU pods, in community-cloud and secure-cloud tiers.

The two tiers run on different hosts and terms. Check which tier a pod is on before a long job.

Rent on RunPod

Vast.ai

billed per GPU-hour

Marketplace of host-listed GPUs; price and reliability vary by host.

Host quality varies. Filter for verified hosts and check the host's reliability before a long run.

Rent on Vast.ai

Referral links: RunLocalAI may earn a commission or account credits if you sign up and rent, at no extra cost to you. Hourly rates and which GPUs are listed change all the time, so check the provider's page before you start. Editorial opinions are independent of the referral relationship; see how we make money.

How to read these numbers

Measured here
Measured here - RunLocalAI ran this exact combo on owner hardware with public evidence.

Source-backed
Source-backed / community - a reproduced public source supports the speed, but it is not labeled as owner-measured.

Extrapolated
Extrapolated - predicted from a measured benchmark on similar-bandwidth hardware.

Estimated
Estimated - formula based on VRAM bandwidth and model architecture; not a benchmark row.

RunLocalAI Will-It-Run Framework →

Want a specific benchmark we don't have? Email Contact support and we'll prioritize it.