▸ AMD· CDNA 3.0

AMD Radeon Instinct MI300

VRAM

128 GB

FLAGSHIP

BANDWIDTH

6550

GB/S

MODELS Q4

302/331

91%

7B Q4 SPEED

~832

BLAZING

▸ MODEL COVERAGE @ Q491% OF ALL

▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

Average speeds at Q4 quantization. Actual performance varies by model architecture and context length.

3B

~1941

TOK/S

7B

~832

TOK/S

14B

~416

TOK/S

32B

~182

TOK/S

70B

~83

TOK/S

▸ SPECIFICATIONS

VRAM

128 GB

BANDWIDTH

6550 GB/s

FP16 COMPUTE

383 TFLOPS

TDP

600W

MEMORY

HBM3

ARCHITECTURE

CDNA 3.0

STREAM PROCESSORS

14,080

COMPUTE UNITS

110

PCIE

Gen 5 x16

MSRP

$12,000

302

FAST MODELS · >30 TOK/S

Real-time chat speed

302

USABLE · >10 TOK/S

Comfortable for all tasks

302

TOTAL COMPATIBLE

Fit in VRAM at Q4

▸ RENT IT IN THE CLOUD

Buying Radeon Instinct MI300 costs $15–$40k and isn’t practical for most teams. Spin one up by the hour instead:

Rent on RunPod →Or Vast.ai →

Spin up in ~60s. Pay by the second. Cancel anytime.

Some links are affiliate links — we may earn a small commission at no extra cost to you. This helps keep FitMyLLM free and independent.

▸ COMPATIBLE MODELS· 302

nomic-embed-text-v1.5 100M0.1B

EMBEDDING·8K CTX· CHAT

TOK/S · 0% VRAM

GPT-2 124M0.124B

GPT2·1K CTX· CHAT

TOK/S · 0% VRAM

SmolLM2 135M0.135B

SMOLLM·2K CTX· CHAT

TOK/S · 0% VRAM

bge-large-en-v1.5 335M0.335B

EMBEDDING·1K CTX· CHAT

TOK/S · 1% VRAM

mxbai-embed-large-v10.335B

EMBEDDING·1K CTX· EMBEDDING

TOK/S · 1% VRAM

Snowflake Arctic Embed L0.335B

EMBEDDING·1K CTX· EMBEDDING

TOK/S · 1% VRAM

GPT-2 Medium 345M0.345B

GPT2·1K CTX· CHAT

TOK/S · 1% VRAM

SmolLM2 360M0.36B

SMOLLM·8K CTX· CHAT

TOK/S · 1% VRAM

Falcon-H1 0.5B0.5B

FALCON·128K CTX· CHAT

TOK/S · 1% VRAM

Qwen 1.5 0.5B0.5B

QWEN·32K CTX· CHAT

TOK/S · 1% VRAM

Qwen 2.5 0.5B0.5B

QWEN·32K CTX· CHAT

TOK/S · 1% VRAM

EMBEDDING·8K CTX· EMBEDDING

TOK/S · 1% VRAM

QWEN·32K CTX· CHAT· REASONING

TOK/S · 1% VRAM

GPT-2 Large 774M0.774B

GPT2·1K CTX· CHAT

TOK/S · 1% VRAM

Qwen 3.5 0.8B0.8B

QWEN·256K CTX· CHAT· CODING· MULTILINGUAL

TOK/S · 1% VRAM

Qwen3.5-0.8B0.9B

QWEN·256K CTX· CHAT

TOK/S · 1% VRAM

FALCON·32K CTX· CHAT

TOK/S · 1% VRAM

INTERNLM·32K CTX· CHAT

TOK/S · 1% VRAM

TinyLlama 1.1B1.1B

LLAMA·2K CTX· CHAT

TOK/S · 1% VRAM

LFM2.5-1.2B-Thinking1.2B

LFM·122K CTX· CHAT· REASONING· TOOL_USE

TOK/S · 1% VRAM

Llama-3.2-1B1.2B

LLAMA·4K CTX· CHAT

TOK/S · 1% VRAM

DeepSeek Coder 1.3B1.3B

DEEPSEEK·16K CTX· CODING

TOK/S · 1% VRAM

EXAONE-4.0-1.2B1.3B

EXAONE·64K CTX· CHAT

TOK/S · 1% VRAM

OPT·2K CTX· CHAT

TOK/S · 1% VRAM

PHI·2K CTX· CODING

TOK/S · 1% VRAM

Phi-1.5 1.3B1.3B

PHI·2K CTX· CHAT· CODING

TOK/S · 1% VRAM

granite-4.0-h-tiny 6.9B6.9BMoE

GRANITE·128K CTX· CHAT

TOK/S · 4% VRAM

Falcon-H1 1.5B1.5B

FALCON·128K CTX· CHAT· CODING

TOK/S · 1% VRAM

GPT-2 XL 1.5B1.5B

GPT2·1K CTX· CHAT

TOK/S · 1% VRAM

Qwen2.5-Coder-1.5B1.5B

QWEN·32K CTX· CHAT· TOOL_USE· CODING

TOK/S · 1% VRAM

Qwen2 Math 1.5B1.5B

QWEN·4K CTX· REASONING

TOK/S · 1% VRAM

Qwen 2.5 1.5B1.5B

QWEN·32K CTX· CHAT· CODING

TOK/S · 1% VRAM

Yi Coder 1.5B1.5B

YI·125K CTX· CODING

TOK/S · 1% VRAM

stablelm-2-1_6b1.6B

STABLELM·4K CTX· CHAT

TOK/S · 1% VRAM

QWEN·32K CTX· CHAT· REASONING

TOK/S · 1% VRAM

SmolLM2 1.7B1.71B

SMOLLM·8K CTX· CHAT

TOK/S · 1% VRAM

Qwen 1.5 1.8B1.8B

QWEN·32K CTX· CHAT

TOK/S · 1% VRAM

Moondream2 1.9B1.9B

OTHER·2K CTX· VISION· CHAT

TOK/S · 1% VRAM

GEMMA·8K CTX· CHAT

TOK/S · 1% VRAM

Granite 3.0 2B2B

GRANITE·128K CTX· CHAT· CODING

TOK/S · 1% VRAM

▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Radeon Instinct MI300.

▸Find models for Radeon Instinct MI300 Compare with other GPUs

▸ DEVICE UNDER TEST

AMD Radeon Instinct MI300 — 128 GB VRAM.

▸ RADEON INSTINCT MI300 SPEC

BRAND: AMD
VRAM: 128 GB HBM3
BANDWIDTH: 6550 GB/s
FP16 COMPUTE: 383 TFLOPS
FP32 COMPUTE: 47.9 TFLOPS
STREAM PROCESSORS: 14,080
TDP: 600 W
ARCHITECTURE: CDNA 3.0
MSRP: $12000

▸ AI CAPABILITY

302/ 331 models @ Q4

With 128 GB VRAM and 6550 GB/s bandwidth, this GPU handles models up to 180B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~749 tok/s.

§ 01TOP MODELS FOR RADEON INSTINCT MI300

302 FIT · SHOWING 20

MODEL	SIZE	VRAM Q4	TOK/S	AVG
Falcon 180B	180B	110.5 GB	32	51.3
bloom 176.2B	176.2B	108.2 GB	33	15.0
dots.llm1.inst 142.8B	142.8B	87.8 GB	41	—
WizardLM 2 8x22B	141B	86.7 GB	149	42.4
Mixtral-8x22B	140.6B	86.4 GB	149	31.9
DBRX 132B	132B	81.2 GB	162	46.3
Qwen3.5-122B-A10B	125.1B	77.0 GB	47	45.5
Pixtral Large 124B	124B	76.3 GB	47	39.3
Mistral-Large 123B	123B	75.7 GB	47	33.5
Devstral 2 123B	123B	75.7 GB	47	38.1
Qwen 3.5 122B A10B	122B	75.1 GB	582	56.8
Nemotron 3 Super 120B	120B	73.8 GB	485	57.3
Nemotron 3 Super 120B-A12B	120B	73.8 GB	485	53.2
Mistral Small 4 119B	119B	73.2 GB	896	50.2
GPT-OSS 120B	117B	72.0 GB	1142	54.1
Command A 111B	111B	68.3 GB	52	27.6
GLM 4.5 Air	110B	67.7 GB	485	51.0
Qwen 1.5 110B	110B	67.7 GB	53	33.4
Llama 4 Scout 17B-16E	109B	67.1 GB	342	33.9
Sarvam 105B	105B	64.7 GB	55	48.0