FitMyLLM
AMD· CDNA 4.0

AMD Radeon Instinct MI355X

Running LLMs on the Radeon Instinct MI355X — the long read: which models fit at which quantisation, and the settings worth changing. · Or what else $30,000 buys

VRAM
288 GB
FLAGSHIP
BANDWIDTH
8190
GB/S
MODELS Q4
425/449
95%
7B Q4 SPEED
~1040
BLAZING
▸ MODEL COVERAGE @ Q495% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

3B
~2427
TOK/S
7B
~1040
TOK/S
14B
~520
TOK/S
32B
~228
TOK/S
70B
~104
TOK/S
▸ MEASURED RIG REPORTS

We rent the machine and time every model on it: decode, VRAM peak, concurrency, watts and cost per million tokens. Yours may already be one of them — and two are free to read in full.

SEE THE REPORTS
▸ SPECIFICATIONS
VRAM
288 GB
BANDWIDTH
8190 GB/s
FP16 COMPUTE
629.1 TFLOPS
TDP
1400W
MEMORY
HBM3e
ARCHITECTURE
CDNA 4.0
STREAM PROCESSORS
16,384
COMPUTE UNITS
128
PCIE
Gen 5 x16
MSRP
$30,000
420
FAST MODELS · >30 TOK/S
Real-time chat speed
425
USABLE · >10 TOK/S
Comfortable for all tasks
425
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ RENT IT IN THE CLOUD

Buying Radeon Instinct MI355X costs $15–$40k and isn’t practical for most teams. Spin one up by the hour instead:

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· 425
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
80889
TOK/S · 0% VRAM
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
53926
TOK/S · 0% VRAM
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
52000
TOK/S · 0% VRAM
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
52000
TOK/S · 0% VRAM
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
28437
TOK/S · 0% VRAM
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
26963
TOK/S · 0% VRAM
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
23869
TOK/S · 0% VRAM
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
21731
TOK/S · 0% VRAM
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
21731
TOK/S · 0% VRAM
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
21731
TOK/S · 0% VRAM
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
20800
TOK/S · 0% VRAM
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
20222
TOK/S · 0% VRAM
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
19158
TOK/S · 0% VRAM
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
14560
TOK/S · 0% VRAM
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
14560
TOK/S · 0% VRAM
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
14000
TOK/S · 0% VRAM
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
12817
TOK/S · 0% VRAM
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
12817
TOK/S · 0% VRAM
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
12133
TOK/S · 0% VRAM
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
12133
TOK/S · 0% VRAM
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
12133
TOK/S · 0% VRAM
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
11742
TOK/S · 0% VRAM
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
9838
TOK/S · 0% VRAM
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
9707
TOK/S · 0% VRAM
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
8988
TOK/S · 0% VRAM
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
8368
TOK/S · 0% VRAM
S
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
7745
TOK/S · 0% VRAM
S
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
6741
TOK/S · 0% VRAM
S
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
6618
TOK/S · 0% VRAM
S
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
6067
TOK/S · 0% VRAM
S
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
6067
TOK/S · 0% VRAM
S
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
6067
TOK/S · 0% VRAM
S
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
6067
TOK/S · 0% VRAM
S
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
5600
TOK/S · 0% VRAM
S
OPT 1.3B1.3B
OPT·2K CTX· CHAT
5600
TOK/S · 0% VRAM
S
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
5600
TOK/S · 0% VRAM
S
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
5393
TOK/S · 0% VRAM
S
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
5127
TOK/S · 0% VRAM
S
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
5127
TOK/S · 0% VRAM
S
LFM2 8B A1B8.3BMoE
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
4853
TOK/S · 2% VRAM
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your Radeon Instinct MI355X.

WHAT THIS CARD IS WORTH

Radeon Instinct MI355X holds 425 of the models in our catalogue and is, in practice, a Q4_K_S card — the largest it takes is MiniMax-M3 at Q4_K_S.

TOKENS/SEC PER $100
2.9
8B at Q4_K_M, so cards compare like for like
VRAM PER $100
0.96 GB
what memory costs on this card
THE BIGGEST IT TAKES
MiniMax-M3427.04B · Q4_K_S252.3 GB363 tok/sEST
Llama 3.1 405B405B · Q4_K_M252.0 GB21 tok/sEST
Hermes 4 405B405B · Q4_K_M252.0 GB21 tok/sEST
Cogito v2 405B405B · Q4_K_M252.0 GB21 tok/sEST
Llama-4-Maverick-17B-128E400B · Q4_K_M247.7 GB448 tok/sEST
Jamba Large 1.6398B · Q4_K_M246.9 GB92 tok/sEST
Qwen3.5-397B-A17B397B · Q4_K_M245.3 GB448 tok/sEST
Hunyuan-Large389B · Q4_K_M241.3 GB167 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
M4 Ultra (384GB)288 GB$9,9990.9 tok/s per $100
Radeon Instinct MI325X288 GB$20,0005.5 tok/s per $100
Radeon Instinct MI355X288 GB$30,0002.9 tok/s per $100
Radeon Instinct MI350X288 GB$25,0003.5 tok/s per $100
Radeon Instinct MI300A192 GB$12,0009.2 tok/s per $100

Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.

▸ DEVICE UNDER TEST

AMD Radeon Instinct MI355X — 288 GB VRAM.

RADEON INSTINCT MI355X SPEC
BRAND
AMD
VRAM
288 GB HBM3e
BANDWIDTH
8190 GB/s
FP16 COMPUTE
629.1 TFLOPS
FP32 COMPUTE
78.6 TFLOPS
STREAM PROCESSORS
16,384
TDP
1400 W
ARCHITECTURE
CDNA 4.0
MSRP
$30000
▸ AI CAPABILITY
425/ 449 models @ Q4

With 288 GB VRAM and 8190 GB/s bandwidth, this GPU handles models up to 405B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~936 tok/s.

RENT IT — LIVE PRICES
Checking what the clouds are charging…
§ 01TOP MODELS FOR RADEON INSTINCT MI355X
425 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
Llama 3.1 405B405B248.0 GB1848.4
Hermes 4 405B405B248.0 GB1833.4
Cogito v2 405B405B248.0 GB18
Llama-4-Maverick-17B-128E400B245.0 GB42846.1
Jamba Large 1.6398B243.8 GB7711.0
Qwen3.5-397B-A17B397B243.2 GB42868.4
Hunyuan-Large389B238.3 GB14073.9
GLM 4.7358B219.3 GB22864.4
GLM 4.6357B218.7 GB60741.4
GLM 4.5355B217.5 GB60747.5
Nemotron-4 340B340B208.3 GB2165.4
MiMo V2.5310.78B190.5 GB48557.5
MiMo V2 Flash 309B309B189.4 GB48572.4
DeepSeek V4 Flash 0731304.18B186.4 GB56062.4
ERNIE-4.5-300B-A47B-Paddle300.5B184.2 GB2437.4
ERNIE 4.5 300B-A47B300B183.9 GB15549.8
Hunyuan Hy3-preview295B180.8 GB34762.7
DeepSeek V4 Flash290.94B178.3 GB56059.3
DeepSeek V2 236B236B144.7 GB34740.4
DeepSeek-V2.5 236B236B144.7 GB34764.0