VOOZH about

URL: https://willitrunai.com/gpus/b100-192gb

โ‡ฑ AI Models for B100 192GB โ€” What Runs on 192GB VRAM


NVIDIA

B100 192GB

Data CenterBlackwellNVLINKCUDA
192GB
VRAM
8kGB/s
Bandwidth
1.8kTFLOPS
FP16 Compute
3.5kTOPS
INT8 Inference
$35,000 MSRP
B100 192GBCategory AvgAMD Instinct MI325X 256GB

Operating mode

Choose the operating mode for this hardware

Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

See Full AI Tier List for B100 192GB โ†’

About this GPU for AI

The NVIDIA B100 is a Blackwell datacenter GPU designed as a drop-in upgrade for existing HGX H100 infrastructure, targeting 192 GB of HBM3e at 8,000 GB/s bandwidth and 1,750 TFLOPS FP16. As a lower-power Blackwell variant at 700W, it fits within the same thermal envelope as existing H100 SXM racks while delivering substantially more VRAM and higher compute. Note: as of early 2025, NVIDIA has reportedly scaled back B100 production in favor of B200 and GB200 allocations, making availability limited. If it ships, it would be the most VRAM-accessible Blackwell GPU at the 8-GPU HGX level.

Beyond LLMs

AI Capability Matrix

What AI tasks this GPU can handle โ€” from text generation to image and video creation.

CapabilityStatusRepresentative ModelDetail
LLM Chat (7B)Runs nativelyLlama 3.1 8B Q4โ€”
LLM Coding (30B)Runs nativelyQwen 3 30B Q4โ€”
LLM Large (70B)Runs nativelyLlama 3.1 70B Q4โ€”
Image Gen (SDXL)Runs nativelySDXL 1.0 FP16~200ms per image
Image Gen (Flux)Runs nativelyFlux.1 Dev FP16~800ms per image
Image Gen (SD 3.5)Runs nativelySD 3.5 Large FP16~900ms per image
Video Short (25f)Runs nativelyLTX Video 2B~100ms/frame
Video Long (100f)Runs nativelyWan Video 14B~400ms/frame
hbm-memorymassive-vramblackwell-architecturedatacenter-gradehigh-bandwidth

Specifications

Compute
FP161750 TFLOPS
INT83500 TOPS
ArchitectureBlackwell
Memory
VRAM192 GB
Bandwidth8000 GB/s
General
FamilyData Center
SegmentData Center
InterconnectNVLINK
Compute PlatformCUDA
MSRP$35,000

Key Features

192 GB HBM3e per card โ€” 8,000 GB/s bandwidth1,750 TFLOPS FP16 / 3,500 INT8 TOPS with FP4 Tensor Core support700W TDP โ€” designed as drop-in replacement for H100 SXM racksNVLink 5.0 with 1.8 TB/s per-GPU bandwidth2nd-gen Transformer Engine with FP4/FP8 supportHGX-compatible baseboard (plug-in H100 upgrade)

For AI Workloads

Strengths
  • 192 GB HBM3e fits 70B models at FP16 with ample KV cache, or small-batched 405B models with Q4
  • Drop-in H100 SXM infrastructure compatibility โ€” upgrade existing systems without new racks
  • 4x estimated inference speedup vs. H100 due to doubled silicon area and FP4 support
  • 8,000 GB/s bandwidth enables very fast token generation for large models
Considerations
  • Availability uncertain โ€” NVIDIA reportedly deprioritized B100 in favor of B200/GB200; limited supply
  • 700W TDP still requires robust cooling, despite being lower than B200
  • FP4 software ecosystem still maturing โ€” framework support for FP4 inference only recently landed
  • Surpassed on performance-per-dollar by B200 for new deployments if cooling infrastructure allows

Architecture

Blackwell

Blackwell is NVIDIA's fifth-generation RTX architecture, built on TSMC's 4NP process. It introduces 5th-generation Tensor Cores with native FP4 precision support, enabling double the inference throughput per watt compared to Ada Lovelace's FP8 operations. Key innovations include the Neural Rendering Pipeline for AI-driven shading and the debut of GDDR7 memory in consumer GPUs.

AI Relevance

FP4 Tensor Cores deliver the highest tokens-per-watt efficiency in any consumer architecture. Native FP4 quantization means models can run at lower precision with minimal quality loss, effectively doubling the effective VRAM for model weights.

Process: TSMC 4NPPlatform: CUDATensor Cores: Gen 5Precisions: FP32, FP16, BF16, FP8, FP4, INT8, INT4

Buying advice

Should you buy B100 192GB for local AI?

Excellent choice for local AI

Runs 40 of 50 top models well โ€” a strong all-rounder for local inference.

192.0 GB

VRAM

$35,000

MSRP

$182/GB

Cost per GB VRAM

Best models for this GPU

What will limit you first

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best upgrade itinerary

Unlocks 4 additional models that do not fit on the current setup.

Want more headroom? AMD Instinct MI325X 256GB (256.0 GB VRAM) is the next step up.

Recommendations by Workload

Chat

S

Mistral Small 4 119B

This model is a direct match for chat. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 292.9 tok/s ยท 256K ctx ยท llama.cppEST.
95.4 GB / 192.0 GB VRAM

Coding

S

Devstral 2 123B Instruct

This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 97.4 tok/s ยท 256K ctx ยท llama.cppEST.
100.5 GB / 192.0 GB VRAM

Agentic Coding

S

Devstral 2 123B Instruct

This model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 97.4 tok/s ยท 256K ctx ยท llama.cppEST.
105.9 GB / 192.0 GB VRAM

Reasoning

S

Devstral 2 123B Instruct

This model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 97.4 tok/s ยท 256K ctx ยท llama.cppEST.
100.5 GB / 192.0 GB VRAM

RAG

S

Qwen 3.5 122B A10B

This model is a direct match for rag. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, lm-studio.

Decode 270.2 tok/s ยท 131K ctx ยท llama.cppEST.
99.4 GB / 192.0 GB VRAM

Full Model Compatibility

๐Ÿ‘ Mistral
Devstral 2 123B Instruct
S96
123B100.5 GB97 tok/s256K ctx
dense
๐Ÿ‘ DeepSeek
DeepSeek V4 Flash
S96
284B179.4 GB145 tok/s169K ctx
moe
๐Ÿ‘ Alibaba
Qwen 3.5 122B A10B
S95
122B97.0 GB270 tok/s131K ctx
moe
๐Ÿ‘ Mistral
Mistral Small 4 119B
S93
119B98.1 GB293 tok/s256K ctx
moe
๐Ÿ‘ OpenAI
GPT-OSS 120B
S93
117B96.4 GB102 tok/s131K ctx
dense
๐Ÿ‘ Mistral AI
Pixtral Large 124B
S93
124B101.1 GB97 tok/s131K ctx
dense
๐Ÿ‘ Cohere
Command A 111B
S93
111B91.7 GB108 tok/s262K ctx
dense
๐Ÿ‘ Alibaba
Qwen 3 235B A22B
S92
235B166.3 GB137 tok/s131K ctx
moe
MiniMax M2.7
S90
230B164.2 GB156 tok/s134K ctx
moe
๐Ÿ‘ Alibaba
Qwen 2.5 VL 72B
S90
72B68.9 GB166 tok/s33K ctx
dense
๐Ÿ‘ Alibaba
Qwen3-Coder-Next
S90
80B70.4 GB454 tok/s256K ctx
moe
๐Ÿ‘ Alibaba
Qwen3-Coder 30B A3B Instruct
S89
30.5B40.2 GB1016 tok/s256K ctx
moe
๐Ÿ‘ Alibaba
Qwen 3.6 35B A3B
S89
35B45.6 GB854 tok/s262K ctx
+1moe
๐Ÿ‘ Mistral
Leanstral 119B A6B
S89
119B101.5 GB269 tok/s181K ctx
moe
๐Ÿ‘ Alibaba
Qwen 3.5 27B
S89
27B39.7 GB378 tok/s131K ctx
dense
๐Ÿ‘ Alibaba
Qwen3-VL 30B A3B Instruct
S89
30B39.9 GB1051 tok/s256K ctx
moe
๐Ÿ‘ Alibaba
Qwen 3.6 27B
S89
27B37.5 GB275 tok/s262K ctx
+1dense
๐Ÿ‘ Alibaba
Qwen 3.5 35B A3B
S88
35B42.9 GB929 tok/s131K ctx
moe
S88
32B43.5 GB374 tok/s131K ctx
dense
๐Ÿ‘ Mistral
Magistral Small 2507
S87
24B37.2 GB336 tok/s131K ctx
dense
๐Ÿ‘ Mistral
Devstral Small 2 24B Instruct
S87
24B37.2 GB336 tok/s256K ctx
dense
๐Ÿ‘ Alibaba
Qwen 3 30B A3B
S87
30.5B40.2 GB1016 tok/s131K ctx
moe
S87
9B27.8 GB126 tok/s131K ctx
dense
๐Ÿ‘ NVIDIA
Nemotron 3 Nano 30B
S87
30B40.8 GB395 tok/s131K ctx
dense
S86
14B31.1 GB196 tok/s131K ctx
dense
๐Ÿ‘ Mistral
Devstral Small 1.1
S86
24B37.2 GB336 tok/s131K ctx
dense
๐Ÿ‘ Microsoft
Phi-4-reasoning-plus 14B
S85
14.7B32.1 GB206 tok/s33K ctx
dense
A85
8B27.2 GB112 tok/s131K ctx
dense
๐Ÿ‘ Google
Gemma 4 31B
A85
30.7B53.5 GB234 tok/s167K ctx
dense
๐Ÿ‘ OpenAI
GPT-OSS 20B
A84
21B35.4 GB1290 tok/s128K ctx
moe
๐Ÿ‘ NVIDIA
Nemotron Cascade 2 30B A3B
A84
30B41.3 GB1039 tok/s262K ctx
moe
A83
4B24.7 GB56 tok/s131K ctx
dense
๐Ÿ‘ LG AI
EXAONE 4.0 32B
A82
32B43.5 GB372 tok/s131K ctx
dense
๐Ÿ‘ Google
Gemma 4 26B A4B
A81
25.2B39.1 GB1091 tok/s256K ctx
moe
๐Ÿ‘ Mistral
Ministral 3 14B
A80
14B31.1 GB196 tok/s262K ctx
multimodal
๐Ÿ‘ NVIDIA
Nemotron Nano 8B
A80
8B26.9 GB112 tok/s131K ctx
dense
๐Ÿ‘ Microsoft
Phi-4 Mini Reasoning 4B
A80
3.8B23.9 GB53 tok/s131K ctx
dense
๐Ÿ‘ DeepSeek
DeepSeek Coder V2 236B
A79
236B222.7 GB84 tok/s8K ctx
moe
๐Ÿ‘ Jina AI
Jina Embeddings v3
A74
0.57B23.2 GB8 tok/s8K ctx
dense
0.57B22.4 GB8 tok/s8K ctx
dense
๐Ÿ‘ Alibaba
Qwen 3.5 397B A17B
F0
397B265.1 GB46 tok/s4K ctx
moe
1000B637.5 GB6 tok/s4K ctx
moe
1000B637.5 GB6 tok/s4K ctx
+1moe
๐Ÿ‘ DeepSeek
DeepSeek V4 Pro
F0
1600B884.0 GB4 tok/s4K ctx
moe
754B499.1 GB8 tok/s4K ctx
moe
744B493.0 GB8 tok/s4K ctx
moe
๐Ÿ‘ DeepSeek
DeepSeek V3.2
F0
671B429.9 GB12 tok/s4K ctx
moe
๐Ÿ‘ Alibaba
Qwen3-Coder 480B A35B Instruct
F0
480B315.8 GB26 tok/s4K ctx
moe
๐Ÿ‘ DeepSeek
DeepSeek R1 671B
F0
671B489.0 GB9 tok/s4K ctx
moe
๐Ÿ‘ DeepSeek
DeepSeek V3.1 671B
F0
671B489.0 GB9 tok/s4K ctx
moe

Just out of reach

Models you could run with an upgrade

High-quality models that need a bit more memory

Image & Video Generation

Diffusion Model Compatibility

52 of 52 models can generate images or video on your B100 192GB

ModelMax ResolutionGen TimeGrade
SD TurboImage512ร—5120msS
Stable Diffusion 1.5Image512ร—7680msS
Realistic Vision v5.1Image512ร—7680msS
DreamShaper 8Image512ร—7680msS
LCM DreamShaper v7Image512ร—7680msS
PixArt-SigmaImage1024ร—1024200msS
FramePack I2VVideo1280ร—720300ms/frameS
SDXL TurboImage512ร—5120msS
SDXL LightningImage1024ร—1024100msS
Stable Diffusion XL 1.0Image1024ร—1024200msS
Playground v2.5Image1024ร—1024300msS
RealVisXL v5.0Image1024ร—1024200msS
DreamShaper XLImage1024ร—1024200msS
Juggernaut XL v9Image1024ร—1024200msS
Animagine XL 3.1Image1024ร—1024200msS
Pony Diffusion V6 XLImage1024ร—1024200msS
Animagine XL 4.0Image1024ร—1024200msS
Illustrious XLImage1024ร—1024200msS
Wan Video 2.1 1.3BVideo480ร—832100ms/frameS
Stable Diffusion 3.5 MediumImage1024ร—1024300msS
Flux.2 Klein 4BImage1024ร—1024100msS
LTX Video 2BVideo1280ร—720100ms/frameS
KolorsImage1024ร—1024300msS
Stable CascadeImage1024ร—1024400msS
AuraFlow v0.3Image1536ร—1536800msS
Stable Diffusion 3.5 LargeImage1024ร—1024900msS
Stable Diffusion 3.5 Large TurboImage1024ร—1024200msS
CogVideoX 2BVideo720ร—480100ms/frameS
HunyuanVideoVideo720ร—1280300ms/frameS
ChromaImage1024ร—1024200msS
Z-Image TurboImage1536ร—1536200msS
Flux.1 DevImage1024ร—1024800msS
Flux.1 SchnellImage1024ร—1024100msS
LTX Video 13BVideo1280ร—720300ms/frameS
Flux.1 Kontext DevImage1024ร—1024800msS
AnimateDiff v1.5.3Video512ร—768100ms/frameS
Cosmos Diffusion 7BVideo1024ร—576200ms/frameS
CogVideoX 5BVideo720ร—480200ms/frameS
Wan2.2 TI2V 5BVideo832ร—480200ms/frameS
Flux.2 Klein 9BImage1024ร—1024100msS
Flux.1 Fill DevImage1024ร—1024700msS
Mochi 1 PreviewVideo848ร—480300ms/frameS
HunyuanVideo 1.5Video720ร—1280300ms/frameS
Helios 14BVideo1280ร—720300ms/frameS
SkyReels V2 14BVideo1280ร—720300ms/frameS
Wan Video 2.1 14BVideo720ร—1280300ms/frameS
Wan Video 2.2 14BVideo720ร—1280300ms/frameS
Qwen ImageImage1024ร—1024300msS
Qwen Image EditImage1024ร—1024300msS
Flux.2 DevImage1024ร—1024~8sS
MAGI-1Video1280ร—720400ms/frameS
HunyuanImage 3.0Image1024ร—1024500msB

Image models estimated at 1024ร—1024 (28 steps, FP16). Video models estimated at 768ร—512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.

Multi-GPU scaling

B100 192GB โ€” Up to 8ร— via NVLink

Scale out with multiple GPUs for larger models. NVLink provides 1800 GB/s inter-GPU bandwidth with 7% overhead.

ConfigEffective memoryModels that fitEst. bandwidth
1ร— B100192 GB359/3748,000 GB/s
2ร— B100384 GB366/37414,880 GB/s
4ร— B100768 GB373/37429,760 GB/s
8ร— B1001536 GB374/37459,520 GB/s

Model counts use default quantization at coding workload settings. Multi-GPU scaling factor: 0.93ร— per additional GPU.

Upgrade paths

Upgrade from B100 192GB

See what you unlock with more powerful hardware

Upgrade options

Upgrade options

๐Ÿ‘ NVIDIA
8ร— B100 192GBMulti-GPU
8 ร— 192 GB = 1536 GB effectivevia NVLink (1800 GB/s)
B
Unlocks 15 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, Kimi K2.5, Kimi K2.6+12 more ยท +171% faster avg

Unlocks 15 additional models that do not fit on the current setup.

Lifts average decode speed across fitting models by about 171%.

NVLink gives this scale-out path a cleaner inter-GPU story than PCIe-only builds.

~$35,000 MSRP

AMD Instinct MI325X 256GBNext step up
256 GB VRAM (+64)
B
Unlocks 4 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, Llama 4 Maverick 17B 128E, Llama 3.1 405B+1 more

Unlocks 4 additional models that do not fit on the current setup.

~$20,000 MSRP

AMD Instinct MI350X 288GBBest value
288 GB VRAM (+96)
B
Unlocks 5 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, Qwen3-Coder 480B A35B Instruct, Llama 4 Maverick 17B 128E+2 more

Unlocks 5 additional models that do not fit on the current setup.

~$8,000 MSRP

Frequently Asked Questions

Compare with similar

B100 192GB vs AMD Instinct MI300X 192GBB100 192GB vs NVIDIA GB200 192GBB100 192GB vs H100 NVL 188GB
Compare this GPUCompare with another GPU โ†’