llmrun

AMD · RDNA 2 · 16 GB VRAM tier

Best AI Models for AMD Radeon RX 6800 (16 GB)

Runs 14B models well; 27–31B only just fits.

16 GB GDDR6512 GB/s32 TFLOPS FP16250 WRDNA 2 2020
Buyabout $579

What LLMs can the RX 6800 run?

#ModelSpeedllmrun ScoreQuant · VRAM
1
Qwen3.5 9BAlibabaHow to run →
Q4_K_M · 6.6 GB · 48.3 t/s
48.3 tok/sQ4_K_M · 6.6 GB
2
Qwen3.5 4BAlibaba
Q4_K_M · 3.4 GB · 94.2 t/s
94.2 tok/sQ4_K_M · 3.4 GB
3
GPT OSS 20BOpenAI
Q4_K_M · 13.1 GB · 75.6 t/s
75.6 tok/sQ4_K_M · 13.1 GB
4
Phi 4Microsoft
Q4_K_M · 12.5 GB · 32.3 t/s
32.3 tok/sQ4_K_M · 12.5 GB
5
Gemma 4 12B ITGoogle
Q4_K_M · 9.0 GB · 37.3 t/s
37.3 tok/sQ4_K_M · 9.0 GB
See all 307 models on the RX 6800 →Best quant that fits · speeds estimated on this card

About the RX 6800

The RX 6800 is an entry card in the 16 GB class: it runs 14B models well. It is power-hungry for its speed, and at its $579 MSRP it gave the best speed per dollar of the cards compared. RDNA 2 generation, released 2020.

Speed in 16 GB#10512 GB/s
Per watt#120.19 tok/s per W on Qwen3.5 9B (9B params)
Per dollar#18.3 tok/s per $100 on Qwen3.5 9B (9B params)
GenerationRDNA 2released 2020

RX 6800 strengths and weaknesses

Strengths

  • Runs 35B MoE models: Qwen3.6 35B A3B at 67.9 tok/s in 15.7 GB

Weaknesses

  • 70B models don't fit: Llama 3.3 70B Instruct needs 51.3 GB at Q4 vs 16 GB here
  • Slow prompt reading: A 4K-token prompt on Qwen3.5 9B waits 11.1 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 6800 has no current price, so it is shown at its MSRP.

CardNew pricetok/stok/s per $100tok/s per 100 W
RX 680016 GB
about $57948.3
8.3
19.3
RTX 507012 GB
about $84068.7
8.2
27.5
RX 7900 XTX24 GB
about $1,13090.6
8.0
25.5
RTX 5070 Ti16 GBBest new value
about $1,17091.6
7.8
30.5
RX 907016 GB
about $80060.4
7.6
27.5
RX 9070 XT16 GB
about $81560.4
7.4
19.9
RTX 508016 GB
about $1,69098.1
5.8
27.3
RX 9060 XT 16GB16 GB
about $59030.2
5.1
18.9
CardUsed pricetok/stok/s per $100tok/s per 100 W
RTX 3060 12GB12 GBBest used value
about $32936.8
11.2
21.6
RX 907016 GB
about $65060.4
9.3
27.5
RX 7900 XTX24 GB
about $1,00090.6
9.1
25.5
RTX 407012 GB
about $57551.5
9.0
25.8
RTX 5070 Ti16 GB
about $1,10091.6
8.3
30.5
RTX 4070 SUPER12 GB
about $67551.5
7.6
23.4
RTX 4070 Ti12 GB
about $72251.5
7.1
18.1
RTX 408016 GB
about $1,10073.3
6.7
22.9

Cost per token: the RX 6800 against 12–32 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.

12 GB16 GB24 GB32 GBthis cardat MSRP
$0.00$0.50$1.00$1.50$2.00$2.50$0$2,000$4,000$6,000$ per 1M output tokens · lower is betternew price →RX 9070 · 16 GB · ≈$800 · $0.99 per 1M tokensRX 7900 XTX · 24 GB · ≈$1,130 · $0.95 per 1M tokensRTX 5070 Ti · 16 GB · ≈$1,170 · $0.95 per 1M tokensRTX 5090 · 32 GB · ≈$5,376 · $1.99 per 1M tokensRTX 5080 · 16 GB · ≈$1,690 · $1.25 per 1M tokensRTX 5070 · 12 GB · ≈$840 · $0.93 per 1M tokensRX 9070 XT · 16 GB · ≈$815 · $1.07 per 1M tokensRX 9060 XT 16GB · 16 GB · ≈$590 · $1.46 per 1M tokensRX 6800 · 16 GB · $579 MSRP · $0.98 per 1M tokensRX 6800RTX 5070 TiRTX 5090RTX 5080RTX 5070RX 9070 XTRX 9060 XT 16GB
$0.00$0.50$1.00$1.50$2.00$2.50$0$2k$4k$6k$ per 1M output tokens · lower is betternew price →RX 9070 · 16 GB · ≈$800 · $0.99 per 1M tokensRX 7900 XTX · 24 GB · ≈$1,130 · $0.95 per 1M tokensRTX 5070 Ti · 16 GB · ≈$1,170 · $0.95 per 1M tokensRTX 5090 · 32 GB · ≈$5,376 · $1.99 per 1M tokensRTX 5080 · 16 GB · ≈$1,690 · $1.25 per 1M tokensRTX 5070 · 12 GB · ≈$840 · $0.93 per 1M tokensRX 9070 XT · 16 GB · ≈$815 · $1.07 per 1M tokensRX 9060 XT 16GB · 16 GB · ≈$590 · $1.46 per 1M tokensRX 6800 · 16 GB · $579 MSRP · $0.98 per 1M tokensRX 6800RTX 5070 TiRTX 5090
Cost per 1M output tokenson Qwen3.5 9B · card over 3 years + power
$0.98
Card, MSRP
$0.76
Power
$0.22
3-year cost of ownership$743
Or rent an RTX 3090 in the cloud$0.19/hr

Renting costs as much as buying after ≈3,000 hours (≈2.1 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 6800, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.

25% faster, same 16 GBRX 9070AMD · 16 GB · RDNA 4
▲ 25% speedsame 16 GBabout $800
52% faster, same 16 GBRTX 4080NVIDIA · 16 GB · Ada Lovelace
▲ 52% speedsame 16 GBabout $1,199rent $0.24/hr
8 GB more and nearly 2× the speedRX 7900 XTXAMD · 24 GB · RDNA 3
▲ 88% speed▲ 8 GBabout $1,130
22% faster, same 16 GBRX 7800 XTAMD · 16 GB · RDNA 3
▲ 22% speedsame 16 GBabout $499
4 GB more and 56% fasterRX 7900 XTAMD · 20 GB · RDNA 3
▲ 56% speed▲ 4 GBabout $899
More efficient per wattRTX 5090 Laptop GPUNVIDIA · 24 GB · Blackwell
▲ 90% speed▲ 8 GB

RX 6800 tokens per second by model size

Output speed depends on memory bandwidth (512 GB/s). Grey rows need more than 16 GB.

  1. 9BQwen3.5 9B · Q4_K_M · 6.6 GB
    48.3tok/s
  2. 14BQwen3 14B · Q4_K_M · 11.9 GB
    32.3tok/s
  3. 27–31BQwen3.8 27B · Q3_K_L · 15.4 GB tight · Q3_K_L
    20.5tok/s
  4. 35B MoEQwen3.6 35B A3B · IQ3_S · 15.7 GB tight · IQ3_S
    67.9tok/s
  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 1057 models on the RX 6800

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 6800 — ranked by Score, recency and Fit
ModelVRAMScoreFit
LFM2 1.2B
Q4_K_M·287.1 t/s
1.2 GBS100, Runs great
Granite 3.1 2B Instruct
Q4_K_M·154.4 t/s
3.2 GBS97, Runs great
T5gemma 2B 2B Ul2 IT
Q4_K_M·83.3 t/s
6.0 GBS88, Runs great
Luciole 8B Base
Q4_K_M·55.1 t/s
5.6 GBA82, Runs well
GPT Neox Japanese 2.7B
BF16·51.7 t/s
5.9 GBA80, Runs well
Smol Llama 101M GQA
Q4_K_M·830.3 t/s
0.4 GBS100, Runs great
Qwen3 30B A3B
Q3_K_M·66.4 t/s
16.0 GBC44, Usable
Neural Chat 7B v3 3
Q4_K_M·62.6 t/s
6.8 GBA84, Runs well
Nemotron H 8B Base 8K
Q4_K_M·57.4 t/s
7.9 GBA82, Runs well
Tiny Qwen2ForCausalLM 2.5
Q4_K_M·1024.0 t/s
0.6 GBS100, Runs great
Qwen3 32B
IQ3_XXS·23.0 t/s
15.7 GBC35, Usable
EuroMoE 2.6B A0.6B 2512
Q4_K_M·166.9 t/s
2.0 GBS98, Runs great
GhostFace 24B V1
Q3_K_M·25.2 t/s
15.2 GBC43, Usable
Granite 4.1 30B
IQ3_XXS·25.6 t/s
15.8 GBC36, Usable
Sarvam 30B
Q2_K·98.0 t/s
14.3 GBA79, Runs well

RX 6800 specs, power and PSU

Specs

Architecture
RDNA 2
VRAM
16 GB GDDR6
Memory bandwidth
512.0 GB/s
FP16 (dense)
32.33 TFLOPS
Stream processors
3,840
TDP
250 W
Released
2020-11-18
MSRP
$579

RX 6800 FAQ

Is the RX 6800 still worth it in 2026?

At its MSRP of $579 you get 8.3 tok/s per $100 on Qwen3.5 9B, which ranks 1st of the 8 cards compared here; it is also the 10th fastest out of 17 cards with 16 GB. It runs 14B models well; 27–31B only just fits.

What is the largest model the RX 6800 can run?

Gemma 4 26B A4B IT (25.8B) is the largest at 4-bit or better: at Q4_K_S it uses 15.5 GB of 16 GB, and runs at about 69.4 tok/s. At 3-bit, a bigger model fits too: Gemma 2 27B IT (27.2B) at IQ3_M, 15.0 GB.

Can the RX 6800 run Qwen3.5 9B?

Yes. Qwen3.5 9B fits at Q4_K_M, using 6.6 GB of 16 GB, and it is the best model that runs well on this card.

Can the RX 6800 run 70B models?

No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than the 16 GB on this card.

How many tokens per second does the RX 6800 give?

About 48.3 tok/s on Qwen3.5 9B at Q4_K_M, with the first token after 11.1 s on a 4K-token prompt.

What power supply does the RX 6800 need?

Plan a 500 W PSU (2× the 250 W TDP).

Is 16 GB enough for coding and agents?

Yes for coding: GPT OSS 20B, the best of the 2 coding models that fit, runs at about 76 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.

Is renting cheaper than buying an RX 6800?

An RTX 3090 (24 GB) rents from $0.19/hr. Renting costs as much as buying after ≈3,000 hours (≈2.1 years at 4 h a day).

Related

Best coding LLMs16 GB VRAM24 GB VRAMGPUs with 16 GB or moreRTX 5090RTX 3090 Ti