Setting Up a Deep-Learning Workstation
With a local workstation, the GPU is installed in your own machine.
Where Computation Runs
Before running deep-learning examples, make one practical decision: where will the computation take place? You can use a local machine with a recent NVIDIA GPU, use a CPU-only machine, or use a cloud environment that provides access to GPU-based computing. This decision affects how quickly the examples run and whether you need to install or maintain suitable hardware yourself.
Why GPU Speed Matters
A recent NVIDIA GPU is recommended because it can make execution faster. Deep-learning work may involve workloads such as image processing or sequence processing, and the source notes that some of these workloads may take much longer on a CPU-only machine. The recommendation is therefore about the experience of running experiments: a suitable GPU can reduce waiting compared with relying on a CPU alone.
Suppose an experiment processes images or sequences. On a CPU-only machine, the experiment can still be run, but the source warns that some workloads may take much longer. On a local machine with a recent NVIDIA GPU, the same type of example has the recommended faster execution environment. The important distinction is not whether the example can run at all, but how long you may wait for it to complete.
Recommendation Versus Requirement
| Environment | What the source says | Practical meaning |
|---|---|---|
| Recent NVIDIA GPU | Recommended for faster execution | Preferred local setup for a better-performing experience |
| Older or absent GPU | A recent NVIDIA GPU is not required | The examples are not blocked solely because this recommended GPU is unavailable |
| CPU-only machine | Possible, although some workloads may take much longer | Use it when longer execution time is acceptable |
| Cloud GPU instance | Can replace access to a suitable local workstation | Run through AWS EC2 or Google Cloud Platform instead of local GPU hardware |
The recommendation affects speed and convenience; it is not an absolute condition for running the examples.
Selecting a Route for a Student Experiment
A learner wants to run the deep-learning examples but does not have a local machine with a recent NVIDIA GPU. Which routes remain available?
- 1Check the requirement
A recent NVIDIA GPU is recommended for speed, but it is not required for running the examples.
- 2Consider CPU-only execution
The learner can use a CPU-only machine, while recognizing that some workloads may take much longer.
- 3Consider a cloud GPU
The learner can instead use a GPU instance from AWS EC2 or Google Cloud Platform.
- 4Choose based on practical constraints
The learner should weigh acceptable execution time, access to hardware, and the possibility of increasing cloud expense.
The missing local NVIDIA GPU does not prevent the learner from running the examples. The practical alternatives are CPU-only execution or a cloud GPU instance.
Local and Cloud Choices
A local modern NVIDIA GPU and a cloud GPU instance solve the same broad problem: they provide a GPU-based computing environment for experiments. The difference is where that environment exists. With a local workstation, the GPU is installed in your own machine. With AWS EC2 or Google Cloud Platform, the experiment runs through a cloud-based computing environment rather than hardware installed locally.
Jupyter as the Experiment Interface
The computing location and the experiment interface are separate decisions. The workstation or cloud service provides the computing environment, while a Jupyter notebook provides the working format for the experiment. The source identifies Jupyter notebooks as the preferred way to run deep-learning experiments, whether the processing happens locally or through a cloud GPU instance.
A learner can use a Jupyter notebook as the consistent experiment interface while changing the computing location. The notebook may run against a local modern NVIDIA GPU or against a GPU instance provided by AWS EC2 or Google Cloud Platform. The notebook format stays conceptually separate from the hardware location.
Common Setup Mistakes
Common mistake Treating a recent NVIDIA GPU as mandatory
What it looks likeAssuming the examples cannot run unless a recent NVIDIA GPU is installed locally.
Why it failsThe source describes the GPU as recommended for faster execution, not required for running the examples.
Do this insteadUse a CPU-only machine or choose a cloud GPU instance when a suitable local GPU is unavailable.
Common mistake Assuming CPU-only execution is equivalent in speed
What it looks likePlanning an image or sequence workload without allowing for longer execution.
Why it failsThe source notes that some workloads may take much longer on a CPU-only machine.
Do this insteadTreat CPU-only execution as possible, but account for the effect on experiment time.
Common mistake Thinking a cloud GPU is a local workstation
What it looks likeDescribing an AWS EC2 or Google Cloud GPU instance as hardware installed in the learner's own machine.
Why it failsA cloud experiment runs through a cloud-based computing environment rather than locally installed hardware.
Do this insteadKeep the distinction clear: the cloud service supplies the computing environment remotely.
Common mistake Ignoring cloud expense in a long-term plan
What it looks likeChoosing cloud GPU access without considering increasing cloud usage.
Why it failsThe source says the possibility of increasing cloud expense belongs in the long-term setup decision.
Do this insteadInclude ongoing cloud expense when comparing cloud access with a local workstation.
Common mistake Confusing the notebook with the hardware
What it looks likeTreating Jupyter as an alternative to choosing a local or cloud computing environment.
Why it failsJupyter is the preferred experiment interface, while the local machine or cloud service provides the computation.
Do this insteadMake both decisions: select where computation runs, then use the notebook format for the experiment.
Choose Your Route
You are preparing to run deep-learning examples involving image or sequence processing. You have no recent NVIDIA GPU in your local machine, and you want to use Jupyter notebooks for the experiments. Choose between CPU-only execution and a cloud GPU instance. Explain what you gain and what trade-off you must consider.
Hints
- A GPU is recommended for speed, but it is not required.
- AWS EC2 and Google Cloud Platform are named cloud alternatives.
- Consider both execution time and the possibility of increasing cloud expense.
Evaluating the Two Practical Alternatives
The learner does not have a suitable local NVIDIA GPU and wants a faster route than CPU-only execution.
- 1Evaluate CPU-only execution
It remains a possible route, but some image or sequence workloads may take much longer.
- 2Evaluate cloud GPU access
A GPU instance from AWS EC2 or Google Cloud Platform can replace access to a suitable local workstation.
- 3Check the long-term trade-off
Cloud access avoids the need for suitable local GPU hardware, but increasing cloud expense must be included in the decision.
Choose the cloud GPU route when access to faster GPU-based computation is important and its ongoing expense is acceptable. Choose CPU-only execution when avoiding cloud use matters more than the possibility of longer runtimes.
Key Takeaways
- 1A recent NVIDIA GPU is recommended because it provides faster execution for the examples.
- 2The GPU recommendation is not a mandatory requirement; CPU-only execution remains possible, although some workloads may take much longer.
- 3AWS EC2 and Google Cloud Platform can provide cloud GPU alternatives when a suitable local workstation is unavailable.
- 4A local GPU and a cloud GPU differ mainly in where the computing environment is provided, while cloud usage introduces a possible increasing expense.
- 5Jupyter notebooks are the preferred experiment interface, separate from the choice of local or cloud computing hardware.
At a glance
- Use a recent NVIDIA GPU when you want the recommended faster local execution.
- Do not confuse a recommendation with a requirement: the examples can run without that GPU.
- Use CPU-only execution if longer runtimes are acceptable, or use AWS EC2 or Google Cloud Platform for cloud GPU access.
- Include the possibility of increasing cloud expense in a long-term setup decision.
- Treat Jupyter notebooks as the experiment interface and the local or cloud GPU as the computing environment.