FitMyLLM
▸ NVIDIA· HOPPER

NVIDIA H200 NVL

Running LLMs on the H200 NVL — the long read: which models fit at which quantisation, and the settings worth changing. · Or what else $35,000 buys

VRAM
141 GB
FLAGSHIP
BANDWIDTH
4890
GB/S
MODELS Q4
398/449
89%
7B Q4 SPEED
~559
BLAZING
▸ MODEL COVERAGE @ Q489% OF ALL
▸ ESTIMATED SPEED· BY MODEL SIZE @ Q4

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

3B
~1304
TOK/S
7B
~559
TOK/S
14B
~279
TOK/S
32B
~122
TOK/S
70B
~56
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
141 GB
BANDWIDTH
4890 GB/s
FP16 COMPUTE
241.3 TFLOPS
TDP
600W
MEMORY
HBM3e
ARCHITECTURE
Hopper
CUDA CORES
16,896
TENSOR CORES
528
PCIE
Gen 5 x16
MSRP
$35,000
395
FAST MODELS · >30 TOK/S
Real-time chat speed
398
USABLE · >10 TOK/S
Comfortable for all tasks
398
TOTAL COMPATIBLE
Fit in VRAM at Q4
▸ RENT IT IN THE CLOUD

Buying H200 NVL 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· 398
S
Falcon-H1R Tiny 90M0.09B
FALCON·256K CTX· CHAT· REASONING
43467
TOK/S · 0% VRAM
›
S
SmolLM2 135M0.135B
SMOLLM·2K CTX· CHAT
28978
TOK/S · 0% VRAM
›
S
nomic-embed-text-v1.5 100M0.14B
EMBEDDING·8K CTX· CHAT
27943
TOK/S · 0% VRAM
›
S
GPT-2 124M0.14B
GPT2·1K CTX· CHAT
27943
TOK/S · 0% VRAM
›
S
SmolVLM 256M0.256B
SMOLLM·8K CTX· CHAT· VISION
15281
TOK/S · 0% VRAM
›
S
Gemma 3 270M0.27B
GEMMA·32K CTX· CHAT
14489
TOK/S · 0% VRAM
›
S
Snowflake Arctic Embed M v2.00.305B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
12826
TOK/S · 0% VRAM
›
S
bge-large-en-v1.5 335M0.335B
EMBEDDING·1K CTX· CHAT
11678
TOK/S · 0% VRAM
›
S
mxbai-embed-large-v10.335B
EMBEDDING·1K CTX· EMBEDDING
11678
TOK/S · 0% VRAM
›
S
Snowflake Arctic Embed L0.335B
EMBEDDING·1K CTX· EMBEDDING
11678
TOK/S · 0% VRAM
›
S
LFM2 350M0.35B
LFM·125K CTX· CHAT· TOOL_USE
11177
TOK/S · 0% VRAM
›
S
SmolLM2 360M0.36B
SMOLLM·8K CTX· CHAT
10867
TOK/S · 1% VRAM
›
S
GPT-2 Medium 345M0.38B
GPT2·1K CTX· CHAT
10295
TOK/S · 1% VRAM
›
S
Qwen 2.5 0.5B0.5B
QWEN·32K CTX· CHAT
7824
TOK/S · 1% VRAM
›
S
SmolVLM 500M0.5B
SMOLLM·8K CTX· CHAT· VISION
7824
TOK/S · 1% VRAM
›
S
Falcon-H1 0.5B0.52B
FALCON·128K CTX· CHAT
7523
TOK/S · 1% VRAM
›
S
BGE-M30.568B
EMBEDDING·8K CTX· EMBEDDING
6887
TOK/S · 1% VRAM
›
S
Snowflake Arctic Embed L v2.00.568B
EMBEDDING·8K CTX· EMBEDDING· MULTILINGUAL
6887
TOK/S · 1% VRAM
›
S
Qwen3-Embedding 0.6B0.6B
EMBEDDING·32K CTX· EMBEDDING· MULTILINGUAL
6520
TOK/S · 1% VRAM
›
S
Falcon-H1R Tiny 0.6B0.6B
FALCON·32K CTX· CHAT· REASONING
6520
TOK/S · 1% VRAM
›
S
Falcon Perception 0.6B0.6B
FALCON·4K CTX· VISION
6520
TOK/S · 1% VRAM
›
S
Qwen 1.5 0.5B0.62B
QWEN·32K CTX· CHAT
6310
TOK/S · 1% VRAM
›
S
LFM2 700M0.74B
LFM·125K CTX· CHAT· TOOL_USE
5286
TOK/S · 1% VRAM
›
S
Qwen3 0.6B0.75B
QWEN·32K CTX· CHAT· REASONING
5216
TOK/S · 1% VRAM
›
S
GPT-2 Large 774M0.81B
GPT2·1K CTX· CHAT
4830
TOK/S · 1% VRAM
›
S
Qwen 3.5 0.8B0.87B
QWEN·256K CTX· CHAT· CODING· MULTILINGUAL
4497
TOK/S · 1% VRAM
›
S
InternVL3 1B0.94B
OTHER·32K CTX· CHAT· VISION
4162
TOK/S · 1% VRAM
›
S
MiniCPM5 1B1.08B
MINICPM·128K CTX· CHAT· REASONING· MULTILINGUAL
3622
TOK/S · 1% VRAM
›
S
TinyLlama 1.1B1.1B
LLAMA·2K CTX· CHAT
3556
TOK/S · 1% VRAM
›
S
LFM2.5-1.2B-Thinking1.2B
LFM·122K CTX· CHAT· REASONING· TOOL_USE
3260
TOK/S · 1% VRAM
›
S
Llama-3.2-1B1.2B
LLAMA·4K CTX· CHAT
3260
TOK/S · 1% VRAM
›
S
LFM2 1.2B1.2B
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
3260
TOK/S · 1% VRAM
›
S
Zamba2 1.2B1.2B
OTHER·4K CTX· CHAT
3260
TOK/S · 1% VRAM
›
S
EXAONE-4.0-1.2B1.3B
EXAONE·64K CTX· CHAT
3009
TOK/S · 1% VRAM
›
S
OPT 1.3B1.3B
OPT·2K CTX· CHAT
3009
TOK/S · 1% VRAM
›
S
MiniCPM-V 4.61.3B
OTHER·256K CTX· CHAT· VISION
3009
TOK/S · 1% VRAM
›
S
DeepSeek Coder 1.3B1.35B
DEEPSEEK·16K CTX· CODING
2898
TOK/S · 1% VRAM
›
S
Phi-1 1.3B1.42B
PHI·2K CTX· CODING
2755
TOK/S · 1% VRAM
›
S
Phi-1.5 1.3B1.42B
PHI·2K CTX· CHAT· CODING
2755
TOK/S · 1% VRAM
›
S
LFM2 8B A1B8.3BMoE
LFM·125K CTX· CHAT· TOOL_USE· MULTILINGUAL
2608
TOK/S · 4% VRAM
›
▸ NEXT STEP

Get personalized recommendations.

See ranked models with benchmark scores, run commands, and precise speed estimates for your H200 NVL.

WHAT THIS CARD IS WORTH

H200 NVL holds 398 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is Command A+ at IQ4_XS.

TOKENS/SEC PER $100
0.1
8B at Q4_K_M, so cards compare like for like
VRAM PER $100
0.40 GB
what memory costs on this card
THE BIGGEST IT TAKES
Command A+218B · IQ4_XS124.5 GB11 tok/sEST
Step 3.7 Flash201.37B · Q4_K_M125.8 GB22 tok/sEST
Falcon 180B180B · Q4_K_M113.8 GB1 tok/sEST
bloom 176.2B176.2B · Q4_K_M125.5 GB1 tok/sEST
dots.llm1.inst 142.8B142.8B · Q6_K123.5 GB1 tok/sEST
WizardLM 2 8x22B141B · Q6_K119.0 GB5 tok/sEST
Mixtral-8x22B140.6B · Q6_K118.7 GB5 tok/sEST
DBRX 132B132B · Q6_K111.4 GB5 tok/sEST
AGAINST CARDS OF SIMILAR MEMORY
M4 Ultra (192GB)144 GB$7,4991.2 tok/s per $100
M2 Ultra (192GB)144 GB$5,4991.3 tok/s per $100
M3 Ultra (192GB)144 GB$6,9991.0 tok/s per $100
B300144 GB$35,0000.1 tok/s per $100
H200 NVL141 GB$35,0000.1 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

NVIDIA H200 NVL — 141 GB VRAM.

▸ H200 NVL SPEC
BRAND
NVIDIA
VRAM
141 GB HBM3e
BANDWIDTH
4890 GB/s
FP16 COMPUTE
241.3 TFLOPS
FP32 COMPUTE
60.3 TFLOPS
CUDA CORES
16,896
TENSOR CORES
528
TDP
600 W
ARCHITECTURE
Hopper
MSRP
$35000
▸ AI CAPABILITY
398/ 449 models @ Q4

With 141 GB VRAM and 4890 GB/s bandwidth, this GPU handles models up to 201.37B parameters.

Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~559 tok/s.

RENT IT — LIVE PRICES
Checking what the clouds are charging…
§ 01TOP MODELS FOR H200 NVL
398 FIT · SHOWING 20
MODELSIZEVRAM Q4TOK/SAVG
Step 3.7 Flash201.37B123.6 GB35653.8
Falcon 180B180B110.5 GB2251.3
bloom 176.2B176.2B108.2 GB2215.0
dots.llm1.inst 142.8B142.8B87.8 GB27—
WizardLM 2 8x22B141B86.7 GB10042.4
Mixtral-8x22B140.6B86.4 GB10031.9
DBRX 132B132B81.2 GB10946.3
Mistral Medium 3.5128B78.7 GB3146.6
Pixtral Large 124B124B76.3 GB3239.3
Nemotron 3 Super 120B-A12B123.61B76.0 GB32653.2
Mistral-Large 123B123B75.7 GB3233.5
Devstral 2 123B123B75.7 GB3238.1
Qwen 3.5 122B A10B122B75.1 GB39156.8
Nemotron 3 Super 120B120B73.8 GB32657.3
Mistral Small 4 119B119B73.2 GB60250.2
GPT-OSS 120B117B72.0 GB76754.1
Command A 111B111B68.3 GB3527.6
GLM 4.5 Air110B67.7 GB32651.0
Qwen 1.5 110B110B67.7 GB3633.4
Llama 4 Scout 17B-16E109B67.1 GB23033.9