llmrun

AMD · RDNA 2 · 8 GB VRAM tier

Best AI Models for AMD Radeon RX 6650 XT (8 GB)

Runs 7–9B models well; 14B doesn't fit in 8 GB.

8 GB GDDR6280 GB/s22 TFLOPS FP16180 WRDNA 2 2022
Buyabout $399

8 GB at 280 GB/s fits ~7-8B Q4 models, no more. Not on AMD's official ROCm list; the Vulkan backend (llama.cpp, LM Studio) is the easy path.

What LLMs can the RX 6650 XT run?

#ModelSpeedllmrun ScoreQuant · VRAM
1
Qwen3.5 9BAlibabaHow to run →
Q4_K_M · 6.6 GB · 26.4 t/s
26.4 tok/sQ4_K_M · 6.6 GB
2
Qwen3.5 4BAlibaba
Q4_K_M · 3.4 GB · 51.5 t/s
51.5 tok/sQ4_K_M · 3.4 GB
3
Gemma 2 9B ITGoogle
Q4_K_M · 7.3 GB · 27.5 t/s
27.5 tok/sQ4_K_M · 7.3 GB
4
Gemma 3 4B ITGoogle
Q4_K_M · 5.0 GB · 59.2 t/s
59.2 tok/sQ4_K_M · 5.0 GB
5
Ling 3.0 TinyInclusion AI
Q4_K_M · 5.5 GB · 96.5 t/s
96.5 tok/sQ4_K_M · 5.5 GB
See all 218 models on the RX 6650 XT →Best quant that fits · speeds estimated on this card

About the RX 6650 XT

The RX 6650 XT is an entry card in the 8 GB class: it runs 7–9B models well. It is power-hungry for its speed, and fair value at ≈$250 used. RDNA 2 generation, released 2022.

Speed in 8 GB#16280 GB/s
Per watt#190.15 tok/s per W on Qwen3.5 9B (9B params)
Per dollar#310.6 tok/s per $100 on Qwen3.5 9B (9B params)
GenerationRDNA 2released 2022

RX 6650 XT strengths and weaknesses

Strengths

  • Runs 7–9B models: Qwen3.5 9B at 26.4 tok/s in 6.6 GB

Weaknesses

  • 14B models don't fit: Qwen3 14B needs 11.9 GB at Q4 vs 8 GB here
  • Little room for context: Qwen3.5 9B uses 6.6 of 8 GB, 1.4 GB left
  • Slow prompt reading: A 4K-token prompt on Qwen3.5 9B waits 16.7 s for the first token
  • ROCm, not CUDA: No FP8 weights; MXFP4 only through llama.cpp, not vLLM

Speed per dollar and per watt

On Qwen3.5 9B (9B params) at Q4_K_M. New prices are compared with new, used with used. The last two columns are the speed you get per $100 of price and per 100 W of board power; higher is better. The RX 6650 XT has no new price, so the New view shows it at its MSRP.

CardNew pricetok/stok/s per $100tok/s per 100 W
RTX 50608 GBBest new value
about $46045.8
10.0
31.6
RTX 507012 GB
about $84068.7
8.2
27.5
RX 6650 XT8 GB
about $39926.4
6.6
14.7
RTX 40608 GB
about $52027.8
5.3
24.2
CardUsed pricetok/stok/s per $100tok/s per 100 W
RTX 3060 12GB12 GB
about $32936.8
11.2
21.6
RTX 50608 GBBest used value
about $43045.8
10.7
31.6
RX 6650 XT8 GB
about $25026.4
10.6
14.7
RTX 407012 GB
about $57551.5
9.0
25.8
RTX 40608 GB
about $33527.8
8.3
24.2
RTX 4070 SUPER12 GB
about $67551.5
7.6
23.4
RTX 4060 Ti 8GB8 GB
about $40029.4
7.4
18.4
RTX 4070 Ti12 GB
about $72251.5
7.1
18.1

Cost per token: the RX 6650 XT against 8–12 GB cards

On Qwen3.5 9B (9B params, Q4_K_M, 4K context). Speed, cost per token and power per token depend on the model.

8 GB12 GBthis cardat MSRP
$0.00$0.50$1.00$1.50$0$1,000$ per 1M output tokens · lower is betterused price →RTX 4060 · 8 GB · ≈$335 · $0.94 per 1M tokensRTX 3060 12GB · 12 GB · ≈$329 · $0.76 per 1M tokensRTX 5060 · 8 GB · ≈$430 · $0.73 per 1M tokensRTX 5070 · 12 GB · ≈$736 · $0.83 per 1M tokensRTX 4070 SUPER · 12 GB · ≈$675 · $1.01 per 1M tokensRTX 4070 Ti · 12 GB · ≈$722 · $1.12 per 1M tokensRTX 4070 · 12 GB · ≈$575 · $0.87 per 1M tokensRTX 4060 Ti 8GB · 8 GB · ≈$400 · $1.09 per 1M tokensRX 6650 XT · 8 GB · ≈$250 · $0.88 per 1M tokensRX 6650 XTRTX 5060RTX 5070RTX 4070 SUPERRTX 4070 TiRTX 4070RTX 3060 12GB
$0.00$0.50$1.00$1.50$0$1k$ per 1M output tokens · lower is betterused price →RTX 4060 · 8 GB · ≈$335 · $0.94 per 1M tokensRTX 3060 12GB · 12 GB · ≈$329 · $0.76 per 1M tokensRTX 5060 · 8 GB · ≈$430 · $0.73 per 1M tokensRTX 5070 · 12 GB · ≈$736 · $0.83 per 1M tokensRTX 4070 SUPER · 12 GB · ≈$675 · $1.01 per 1M tokensRTX 4070 Ti · 12 GB · ≈$722 · $1.12 per 1M tokensRTX 4070 · 12 GB · ≈$575 · $0.87 per 1M tokensRTX 4060 Ti 8GB · 8 GB · ≈$400 · $1.09 per 1M tokensRX 6650 XT · 8 GB · ≈$250 · $0.88 per 1M tokensRX 6650 XTRTX 5060RTX 5070
Cost per 1M output tokenson Qwen3.5 9B · card over 3 years + power
$0.88
Card, used price
$0.60
Power
$0.28
3-year cost of ownership$368
Or rent an RTX 5080 in the cloud$0.22/hr

Renting costs as much as buying after ≈1,100 hours (≈0.8 years at 4 h a day).

Include card price

Each dot is a card; hover or tap it for its numbers, click to open it. Cards without a price for the chosen basis are left out. How we calculate cost per token

Similar GPUs

Cards people weigh against the RX 6650 XT, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.

4 GB more and 39% fasterRTX 3060 12GBNVIDIA · 12 GB · Ampere
▲ 39% speed▲ 4 GBabout $329rent $0.07/hr
74% faster, same 8 GBRTX 5060NVIDIA · 8 GB · Blackwell
▲ 74% speedsame 8 GBabout $460
Same 8 GB, new with warrantyRTX 4060NVIDIA · 8 GB · Ada Lovelace
▲ 5% speedsame 8 GBabout $520rent $0.09/hr
4 GB more and 55% fasterRX 9070 GREAMD · 12 GB · RDNA 4
▲ 55% speed▲ 4 GBabout $549
53% faster, same 8 GBA580Intel · 8 GB · Alchemist
▲ 53% speedsame 8 GBabout $179
8 GB more and 2.8× the speedRTX 4080NVIDIA · 16 GB · Ada Lovelace
▲ 178% speed▲ 8 GBabout $1,199rent $0.27/hr

RX 6650 XT tokens per second by model size

Output speed depends on memory bandwidth (280 GB/s). Grey rows need more than 8 GB.

  1. 9BQwen3.5 9B · Q4_K_M · 6.6 GB
    26.4tok/s
  2. 14BQwen3 14B · not at 4-bit
    needs 11.9 GB
  3. 27–31BQwen3.8 27B · not at 4-bit
    needs 17.9 GB
  4. 35B MoEQwen3.6 35B A3B · not at 4-bit
    needs 22.0 GB
  5. 70BLlama 3.3 70B Instruct · not at 4-bit
    needs 51.3 GB
  6. 120B MoEGPT OSS 120B · not at 4-bit
    needs 70.8 GB

Estimated speeds: Q4_K_M, 4K context, one user. How we estimate

All 842 models on the RX 6650 XT

Featured only

Best quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with AMD Radeon RX 6650 XT — ranked by Score, recency and Fit
ModelVRAMScoreFit
Tiny GPT2LMHeadModel
Q4_K_M·16800.0 t/s
0.0 GBS100, Runs great
Qwen2.5 Omni 3B
Q4_K_M·46.0 t/s
6.0 GBA78, Runs well
OpenMath Nemotron 1.5B
Q4_K_M·131.3 t/s
1.7 GBS95, Runs great
Codeguard Jev Style 1.5B
Q4_K_M·131.3 t/s
1.7 GBS95, Runs great
Tiny GPTNeoXForCausalLM
Q4_K_M·560.0 t/s
0.3 GBS100, Runs great
Tiny MistralForCausalLM 0.2
Q4_K_M·560.0 t/s
0.5 GBS100, Runs great
Qwen3 Embedding 0.6B
Q4_K_M·218.2 t/s
1.6 GBS100, Runs great
TwIL 8B
Q4_K_S·30.1 t/s
7.9 GBC41, Usable
Tiny MistralForCausalLM 0.1
Q4_K_M·560.0 t/s
0.5 GBS100, Runs great
Scout 4B
Q4_K_M·51.9 t/s
3.6 GBA81, Runs well
Gemma3 1B IT
BF16·76.4 t/s
3.6 GBS87, Runs great
Phi 4 Mini Flash Reasoning
Q4_K_M·56.9 t/s
5.3 GBA82, Runs well
Huihui Qwen3 4B Abliterated v2
Q4_K_M·57.9 t/s
4.2 GBA83, Runs well
Tiny GemmaForCausalLM
Q4_K_M·560.0 t/s
0.4 GBS100, Runs great
Tiny FalconMambaForCausalLM
BF16·16800.0 t/s
0.2 GBS100, Runs great

RX 6650 XT specs, power and PSU

Specs

Architecture
RDNA 2
VRAM
8 GB GDDR6
Memory bandwidth
280.0 GB/s
FP16 (dense)
21.59 TFLOPS
Stream processors
2,048
TDP
180 W
Released
2022-05-10
MSRP
$399

RX 6650 XT FAQ

Is the RX 6650 XT still worth it in 2026?

At ≈$250 used you get 10.6 tok/s per $100 on Qwen3.5 9B, which ranks 3rd of the 8 cards compared here; it is also the 16th fastest out of 20 cards with 8 GB. It runs 7–9B models well; 14B doesn't fit in 8 GB.

What is the largest model the RX 6650 XT can run?

Qwen3.5 9B (9.7B) is the largest at 4-bit or better: at Q4_K_M it uses 6.6 GB of 8 GB, and runs at about 26.4 tok/s.

Can the RX 6650 XT run 14B models?

No. Qwen3 14B needs 11.9 GB at Q4, more than the 8 GB on this card.

How many tokens per second does the RX 6650 XT give?

About 26.4 tok/s on Qwen3.5 9B at Q4_K_M, with the first token after 16.7 s on a 4K-token prompt.

What power supply does the RX 6650 XT need?

Plan a 400 W PSU (2× the 180 W TDP).

Is renting cheaper than buying an RX 6650 XT?

An RTX 5080 (16 GB) rents from $0.22/hr. Renting costs as much as buying after ≈1,100 hours (≈0.8 years at 4 h a day).

Related

Best coding LLMs8 GB VRAM12 GB VRAMGPUs with 8 GB or moreRTX 5080RTX 3080 Ti