VOOZH about

URL: https://willitrunai.com/gpus/tesla-p40-24gb

⇱ AI Models for Tesla P40 24GB β€” What Runs on 24GB VRAM


NVIDIA

Tesla P40 24GB

Pascal DatacenterDatacenterPascalPCIe 3CUDA
24GB
VRAM
346GB/s
Bandwidth
24TFLOPS
FP16 Compute
47TOPS
INT8 Inference
$5,699 MSRP
Tesla P40 24GBCategory AvgMacBook Pro M4 Max 36GB

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 Tesla P40 24GB β†’

About this GPU for AI

The Tesla P40 is a Pascal-generation datacenter GPU from 2016, built for inference workloads before the era of large language models. At 24 GB of GDDR5X, it was notable as one of the first high-VRAM inference accelerators, and it saw renewed interest from the LLM community when NVLink 3090-class cards were scarce. It can run 7B models at Q4 quantization but generation will be slow by current standards. Available on the used market for very low prices, it remains a viable ultra-budget option for hobbyists building an inference server, though modern alternatives are strongly preferred.

Beyond LLMs

AI Capability Matrix

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

CapabilityStatusRepresentative ModelDetail
LLM Chat (7B)Runs nativelyLlama 3.1 8B Q4β€”
LLM Coding (30B)Runs nativelyQwen 3 30B Q4β€”
LLM Large (70B)Won’t fitLlama 3.1 70B Q4β€”
Image Gen (SDXL)Runs nativelySDXL 1.0 FP16~~17.6s per image
Image Gen (Flux)Runs with offloadFlux.1 Dev FP16~~1m 19s per image
Image Gen (SD 3.5)Runs nativelySD 3.5 Large FP16~~1m 37s per image
Video Short (25f)Runs nativelyLTX Video 2B~~15.3s/frame
Video Long (100f)Won't fitWan Video 14B~~45s/frame
legacy-datacenterbudget-friendlylarge-vramend-of-life

Specifications

Compute
FP1624 TFLOPS
INT847 TOPS
ArchitecturePascal
Memory
VRAM24 GB
Bandwidth346 GB/s
General
FamilyPascal Datacenter
SegmentDatacenter
InterconnectPCIe 3
Compute PlatformCUDA
MSRP$5,699

Key Features

24 GB GDDR5X VRAM346 GB/s memory bandwidth24 TFLOPS FP16 / Pascal compute architecturePCIe 3.0 x16, 250W TDPPassive cooling β€” requires active airflow in server chassisCUDA Compute Capability 6.1

For AI Workloads

Strengths
  • 24 GB VRAM at extremely low used-market prices β€” among the cheapest options for loading 13B Q4 models
  • Widely available used; originally deployed in large hyperscale data centers and now decommissioned in volume
  • Passive form factor works in standard server chassis with adequate airflow
  • Runs 7B models at Q4 β€” functional for light local inference on a budget
Considerations
  • Pascal architecture from 2016 β€” no Tensor Cores, no FP16 hardware acceleration beyond basic CUDA
  • 346 GB/s bandwidth makes token generation noticeably slow even for 7B models
  • No INT8 hardware acceleration β€” quantized inference falls back to software paths
  • End of driver support approaching; not guaranteed to work with latest inference frameworks

Architecture

Pascal

Pascal is NVIDIA's first 16nm FinFET GPU architecture, powering the GTX 10-series consumer cards and Tesla P100/P40 datacenter accelerators. It introduced unified memory architecture and NVLink interconnect for datacenter GPUs.

AI Relevance

No dedicated Tensor Cores β€” all AI inference runs on standard CUDA cores at FP16 or FP32 precision. Still usable for small models (7B Q4) on cards with sufficient VRAM like the GTX 1080 Ti (11 GB) or P40 (24 GB), but significantly slower than Turing and newer.

Process: TSMC 16nmPlatform: CUDAPrecisions: FP32, FP16

Buying advice

Should you buy Tesla P40 24GB for local AI?

Excellent choice for local AI

Runs 26 of 50 top models well β€” a strong all-rounder for local inference.

24.0 GB

VRAM

$5,699

MSRP

$237/GB

Cost per GB VRAM

Best models for this GPU

What will limit you first

This setup is broadly balanced for this model.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Older PCIe generation

PCIe 3.0 is workable, but it compounds the penalty when you offload heavily or try to scale across multiple cards.

Best upgrade itinerary

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Unlocks 1 additional models that do not fit on the current setup.

Want more headroom? MacBook Pro M4 Max 36GB (36.0 GB unified memory) is the next step up.

Recommendations by Workload

Chat

S

Qwen 3 14B

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, ollama, lm-studio.

Decode 25.8 tok/s Β· 80K ctx Β· llama.cppEST.
13.1 GB / 24.0 GB VRAM

Coding

S

Devstral Small 2 24B Instruct

This model is a direct match for coding. It belongs to a current frontier family for local AI. It should run, but memory headroom will be limited. Known channels: huggingface, ollama, lm-studio.

Decode 15.0 tok/s Β· 40K ctx Β· llama.cppEST.
20.4 GB / 24.0 GB VRAM

Agentic Coding

S

Qwen 3.6 27B

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 should run, but memory headroom will be limited. Known channels: huggingface, lm-studio.

Decode 10.2 tok/s Β· 69K ctx Β· llama.cppEST.
21.7 GB / 24.0 GB VRAM

Reasoning

S

Qwen 3 14B

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, ollama, lm-studio.

Decode 25.8 tok/s Β· 80K ctx Β· llama.cppEST.
14.3 GB / 24.0 GB VRAM

RAG

A

Granite 4.1 8B

This model is a direct match for rag. It sits in the middle of the current model mix. It fits natively with comfortable headroom. Known channels: huggingface, ollama.

Decode 45.0 tok/s Β· 104K ctx Β· llama.cppEST.
13.1 GB / 24.0 GB VRAM

Full Model Compatibility

πŸ‘ Alibaba
Qwen3-Coder 30B A3B Instruct
S93
30.5B23.4 GB31 tok/s23K ctx
moe
πŸ‘ Alibaba
Qwen3-VL 30B A3B Instruct
S93
30B23.1 GB32 tok/s26K ctx
moe
πŸ‘ OpenAI
GPT-OSS 20B
S93
21B18.6 GB39 tok/s52K ctx
moe
S91
14B14.3 GB26 tok/s80K ctx
dense
πŸ‘ Microsoft
Phi-4-reasoning-plus 14B
S91
14.7B15.3 GB25 tok/s33K ctx
dense
πŸ‘ Alibaba
Qwen 3 30B A3B
S91
30.5B23.4 GB31 tok/s23K ctx
moe
πŸ‘ Alibaba
Qwen 3.5 9B
S91
9B11.0 GB40 tok/s111K ctx
dense
πŸ‘ Alibaba
Qwen 3.5 27B
S90
27B22.9 GB13 tok/s21K ctx
dense
πŸ‘ Mistral
Magistral Small 2507
S90
24B20.4 GB15 tok/s40K ctx
dense
πŸ‘ Mistral
Devstral Small 2 24B Instruct
S89
24B20.4 GB15 tok/s40K ctx
dense
πŸ‘ Alibaba
Qwen 3.6 27B
S89
27B20.7 GB10 tok/s69K ctx
+1dense
S89
8B10.4 GB45 tok/s115K ctx
dense
πŸ‘ Mistral
Devstral Small 1.1
S88
24B20.4 GB15 tok/s40K ctx
dense
πŸ‘ Alibaba
Qwen 3.5 4B
S87
4B7.9 GB56 tok/s131K ctx
dense
πŸ‘ NVIDIA
Nemotron Cascade 2 30B A3B
S87
30B24.5 GB22 tok/s13K ctx
moe
πŸ‘ NVIDIA
Nemotron 3 Nano 30B
S87
30B24.0 GB9 tok/s16K ctx
dense
πŸ‘ Google
Gemma 4 26B A4B
S86
25.2B22.3 GB33 tok/s23K ctx
moe
πŸ‘ Mistral
Ministral 3 14B
S86
14B14.3 GB26 tok/s80K ctx
multimodal
πŸ‘ NVIDIA
Nemotron Nano 8B
A84
8B10.1 GB45 tok/s130K ctx
dense
πŸ‘ Microsoft
Phi-4 Mini Reasoning 4B
A84
3.8B7.1 GB53 tok/s131K ctx
dense
πŸ‘ Alibaba
Qwen 3.5 35B A3B
A80
35B26.1 GB17 tok/s4K ctx
moe
πŸ‘ Jina AI
Jina Embeddings v3
A77
0.57B6.4 GB8 tok/s8K ctx
dense
A77
32B26.7 GB7 tok/s5K ctx
dense
A75
0.57B5.6 GB8 tok/s8K ctx
dense
πŸ‘ Alibaba
Qwen 3.6 35B A3B
A74
35B28.8 GB13 tok/s4K ctx
+1moe
πŸ‘ LG AI
EXAONE 4.0 32B
A71
32B26.7 GB7 tok/s5K ctx
dense
πŸ‘ Alibaba
Qwen 3.5 397B A17B
F0
397B248.3 GB2 tok/s4K ctx
moe
πŸ‘ Mistral
Devstral 2 123B Instruct
F0
123B83.7 GB2 tok/s4K ctx
dense
1000B620.7 GB2 tok/s4K ctx
moe
1000B620.7 GB2 tok/s4K ctx
+1moe
πŸ‘ DeepSeek
DeepSeek V4 Pro
F0
1600B867.2 GB2 tok/s4K ctx
moe
πŸ‘ Alibaba
Qwen 3.5 122B A10B
F0
122B80.2 GB2 tok/s4K ctx
moe
πŸ‘ DeepSeek
DeepSeek V4 Flash
F0
284B162.6 GB2 tok/s4K ctx
moe
πŸ‘ Mistral
Mistral Small 4 119B
F0
119B81.3 GB2 tok/s4K ctx
moe
πŸ‘ Cohere
Command A 111B
F0
111B74.9 GB2 tok/s4K ctx
dense
πŸ‘ Alibaba
Qwen 2.5 VL 72B
F0
72B52.1 GB2 tok/s4K ctx
dense
πŸ‘ OpenAI
GPT-OSS 120B
F0
117B79.6 GB2 tok/s4K ctx
dense
πŸ‘ Alibaba
Qwen3-Coder-Next
F0
80B53.6 GB2 tok/s4K ctx
moe
F0
754B482.3 GB2 tok/s4K ctx
moe
πŸ‘ Mistral AI
Pixtral Large 124B
F0
124B84.3 GB2 tok/s4K ctx
dense
744B476.2 GB2 tok/s4K ctx
moe
πŸ‘ DeepSeek
DeepSeek V3.2
F0
671B413.1 GB2 tok/s4K ctx
moe
πŸ‘ Alibaba
Qwen 3 235B A22B
F0
235B149.5 GB2 tok/s4K ctx
moe
πŸ‘ Alibaba
Qwen3-Coder 480B A35B Instruct
F0
480B299.0 GB2 tok/s4K ctx
moe
πŸ‘ Google
Gemma 4 31B
F0
30.7B36.7 GB3 tok/s4K ctx
dense
MiniMax M2.7
F0
230B147.4 GB2 tok/s4K ctx
moe
πŸ‘ Mistral
Leanstral 119B A6B
F0
119B84.7 GB2 tok/s4K ctx
moe
πŸ‘ DeepSeek
DeepSeek Coder V2 236B
F0
236B205.9 GB2 tok/s4K ctx
moe
πŸ‘ DeepSeek
DeepSeek R1 671B
F0
671B472.2 GB2 tok/s4K ctx
moe
πŸ‘ DeepSeek
DeepSeek V3.1 671B
F0
671B472.2 GB2 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

41 of 52 models can generate images or video on your Tesla P40 24GB

ModelMax ResolutionGen TimeGrade
SD TurboImage512Γ—512~2.2sS
Stable Diffusion 1.5Image512Γ—768~4.4sS
Realistic Vision v5.1Image512Γ—768~4.4sS
DreamShaper 8Image512Γ—768~4.4sS
LCM DreamShaper v7Image512Γ—768~1.3sS
PixArt-SigmaImage1024Γ—1024~17.6sS
FramePack I2VVideo256Γ—256~32.3s/frameS
SDXL TurboImage512Γ—512~2.2sS
SDXL LightningImage1024Γ—1024~6.6sS
Stable Diffusion XL 1.0Image1024Γ—1024~17.6sS
Playground v2.5Image1024Γ—1024~26.4sS
RealVisXL v5.0Image1024Γ—1024~19.8sS
DreamShaper XLImage1024Γ—1024~19.8sS
Juggernaut XL v9Image1024Γ—1024~19.8sS
Animagine XL 3.1Image1024Γ—1024~19.8sS
Pony Diffusion V6 XLImage1024Γ—1024~19.8sS
Animagine XL 4.0Image1024Γ—1024~19.8sS
Illustrious XLImage1024Γ—1024~19.8sS
Wan Video 2.1 1.3BVideo256Γ—256~12.9s/frameS
Stable Diffusion 3.5 MediumImage1024Γ—1024~30.8sS
Flux.2 Klein 4BImage1024Γ—1024~5.3sS
LTX Video 2BVideo768Γ—512~15.3s/frameS
KolorsImage1024Γ—1024~35.2sS
Stable CascadeImage1024Γ—1024~44sS
AuraFlow v0.3Image1536Γ—1536~1m 19sS
Stable Diffusion 3.5 LargeImage1024Γ—1024~1m 37sS
Stable Diffusion 3.5 Large TurboImage1024Γ—1024~17.6sS
CogVideoX 2BVideo720Γ—480~15.3s/frameA
HunyuanVideoVideo256Γ—256~32.3s/frameA
ChromaImage256Γ—256~32.3sA
Z-Image TurboImage1536Γ—1536~18.2sB
Flux.1 DevImage256Γ—256~1m 19sB
Flux.1 SchnellImage256Γ—256~15.4sB
LTX Video 13BVideo256Γ—256~32.3s/frameB
Flux.1 Kontext DevImage256Γ—256~1m 28sB
AnimateDiff v1.5.3Video512Γ—768~8s/frameB
Cosmos Diffusion 7BVideo256Γ—256~48.7s/frameB
CogVideoX 5BVideo256Γ—256~46.3s/frameB
Wan2.2 TI2V 5BVideo256Γ—256~46.3s/frameB
Flux.2 Klein 9BImage256Γ—256~16.1sD
Flux.1 Fill DevImage256Γ—256~1m 15sD
Mochi 1 PreviewVideo256Γ—256~29.1s/frameF
HunyuanVideo 1.5Video256Γ—256~27s/frameF
Helios 14BVideo256Γ—256~33.3s/frameF
SkyReels V2 14BVideo256Γ—256~33.3s/frameF
Wan Video 2.1 14BVideo256Γ—256~33.3s/frameF
Wan Video 2.2 14BVideo256Γ—256~33.3s/frameF
Qwen ImageImage256Γ—256~29.6sF
Qwen Image EditImage256Γ—256~29.6sF
Flux.2 DevImage256Γ—256~13m 53sF
MAGI-1Video256Γ—256~41.3s/frameF
HunyuanImage 3.0Image256Γ—256~52.2sF

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

Upgrade paths

Upgrade from Tesla P40 24GB

See what you unlock with more powerful hardware

Upgrade options

Upgrade options

MacBook Pro M4 Max 36GBNext step up
36 GB Unified (+12)410 GB/s (+64)
A
Unlocks 1 additional models that do not fit on the current setup.Unlocks Gemma 4 31B+33% faster avg

Unlocks 1 additional models that do not fit on the current setup.

Lifts average decode speed across fitting models by about 33%.

~$2,499 MSRP

πŸ‘ NVIDIA
RTX 5000 Ada 32GBNVIDIA upgrade
32 GB VRAM (+8)576 GB/s (+230)
A
Unlocks 6 additional models that do not fit on the current setup.Unlocks Gemma 4 31B, Kimi Linear 48B A3B, Falcon 40B Instruct+3 more Β· +85% faster avg

Unlocks 6 additional models that do not fit on the current setup.

Lifts average decode speed across fitting models by about 85%.

~$4,000 MSRP

Mac mini M4 64GBBest value
64 GB Unified (+40)
B
Unlocks 17 additional models that do not fit on the current setup.Unlocks Qwen 2.5 VL 72B, Gemma 4 31B, Llama 3.3 70B+14 more

Unlocks 17 additional models that do not fit on the current setup.

~$1,099 MSRP

AMD Instinct MI350X 288GBBiggest leap
288 GB VRAM (+264)8000 GB/s (+7654)
B
Unlocks 45 additional models that do not fit on the current setup.Unlocks Qwen 3.5 397B A17B, Devstral 2 123B Instruct, Qwen 3.5 122B A10B+42 more Β· +372% faster avg

Unlocks 45 additional models that do not fit on the current setup.

Lifts average decode speed across fitting models by about 372%.

~$8,000 MSRP

Frequently Asked Questions

Compare with similar

Tesla P40 24GB vs RTX 3090 24GBTesla P40 24GB vs RTX 3090 Ti 24GBTesla P40 24GB vs RTX 4090 24GB
Compare this GPUCompare with another GPU β†’