Is the Tesla P40 good for training modern models?

Not really. Pascal has no Tensor Cores and no BF16 support, and its FP16 throughput is very low, so mixed-precision training is slow. Use it for quantized inference or FP32 jobs; for training, an RTX 3090 or newer card is a better fit.

Which LLMs run well on 24 GB?

4-bit GGUF models up to about 30B parameters, and 7B–14B models at 8-bit, via Ollama or llama.cpp. Expect moderate speed: 257 GB/s limits tokens per second.

Will current PyTorch and vLLM work on the Tesla P40?

It depends on the build. The P40 has compute capability 6.1 and some recent CUDA and framework builds no longer target older GPUs. Check the CUDA and framework versions in your template or Docker image before you deploy.

Tesla P40 vs RTX 3090: what is the trade-off?

Both have 24 GB. The RTX 3090 is listed at 35.4 TFLOPS and 805 GB/s with modern Tensor Cores, so it is far faster. The P40 is the older option when 24 GB matters more than speed.

GPU Cloud · NVIDIA · Pascal

Rent an NVIDIA Tesla P40 GPU with 24 GB VRAM

Rent Tesla P40 cloud GPU servers by the hour on Cloudj: 24 GB VRAM on NVIDIA Pascal with about 9.4 TFLOPS FP32 and 257 GB/s memory bandwidth. It is a low-frills way to get 24 GB of memory for quantized LLM inference, hobby projects and classic CUDA work, at a 250 W power limit. Check that your CUDA and framework versions still support this GPU, pick a live configuration and connect over SSH with your key.

Rent Tesla P40 All GPUs

Tesla P40 specifications

VRAM per GPU

24 GB

Architecture

NVIDIA Pascal

Compute capability

6.1

FP32 compute

≈ 9.4 TFLOPS

Memory bandwidth

≈ 257 GB/s

Power limit

250 W

Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.

Live Tesla P40 configurations and prices

Out of stock right now

Browse available GPUs

What you can run on the Tesla P40 cloud GPU

Quantized LLMs with Ollama

24 GB fits GGUF models up to roughly 30B parameters at 4-bit, which is the P40's best use today.

Private chat and RAG prototypes

Run Ollama with Open WebUI for a private assistant over your documents without sending data to an API.

Classic FP32 workloads

Run older CUDA codes, simulations and FP32 training that do not depend on Tensor Cores.

Learning and coursework

Get hands-on with GPU servers, drivers and CUDA on 24 GB of memory for small experiments.

Is the Tesla P40 the right GPU for your workload?

Choose the Tesla P40 if your primary requirement is 24 GB of VRAM for memory-hungry inference or development tasks where older Pascal compute is sufficient. Opt for a newer or larger GPU if you need higher memory bandwidth or modern tensor acceleration for heavy training, or a smaller GPU if your workload uses little memory.

Tesla P40 vs similar cloud GPUs

GPU	VRAM per GPU	FP32 compute	Power limit
NVIDIA Tesla P40	24 GB	≈ 9.4 TFLOPS	250 W
NVIDIA L4	24 GB	≈ 24.2 TFLOPS	72 W	Details →
NVIDIA Tesla V100	16 GB	≈ 12.5 TFLOPS	300 W	Details →
NVIDIA A100 PCIE	40 GB	≈ 15.6 TFLOPS	250 W	Details →
NVIDIA A100 SXM4	40 GB	≈ 15.6 TFLOPS	400 W	Details →

Guides for Tesla P40

GPU cloud Asia guide: Japan, Singapore, Korea and nearbyA practical GPU cloud Asia guide: which country codes to search for East and Southeast Asian users, a fallback order, and how to pair a GPU with a VPS.GPU cloud by location: how to pick a country for your GPUHow GPU cloud by location works on Cloudj: filter a live list of machines by country code, decide between near users, lowest price or data rules.GPU cloud Europe: choosing a country for your AI workloadsA practical guide to GPU cloud Europe searches: filter by several country codes at once, weigh Western, Nordic and Eastern hosts, and plan where data lives.GPU cloud North America: renting GPUs in the US and CanadaA GPU cloud North America guide: how to search US and CA hosts, read the results, pick a VPS front end and decide when a US GPU suits users elsewhere.

All articles →

Tesla P40 cloud GPU: frequently asked questions

Rent the Tesla P40 cloud GPU by the hour

Pick a live configuration, choose a template and connect over SSH with your key. Billing stops when you destroy the instance.

Rent Tesla P40 All GPUs