RTX 4090 provides Ada Lovelace compute with 24 GB GDDR6X. It is a candidate for development, image generation and memory-bounded inference experiments.
On-demand prices are Arvica's live catalogue rates (USD per GPU-hour), shown at deploy time. Reserved and enterprise prices are set in an accepted Arvica proposal.
Term, volume and dedicated-capacity prices are set in an accepted Arvica proposal. Configuration, region and delivery are confirmed in your proposal.
Availability confirmed before order acceptance. Taxes and additional services are confirmed before payment.
Choose around your workload
01
When to choose this GPU
Evaluate tasks whose weights, activations and caches fit within one GPU. For experiments, compare completed-job cost and reproducibility across available configurations.
02
When to consider alternatives
Consider RTX 5090 for more memory and a Blackwell software path. For 48 GB workloads, compare RTX 6000 Ada and L40S before increasing GPU count.
03
Memory planning
The standard RTX 4090 has 24 GB GDDR6X. Model size alone does not account for batch size, attention cache and intermediate allocations.
04
Interconnect and scale
Confirm the server's PCIe and networking layout for multiple GPUs. Separate graphics capability from any promised multi-node training capability.
05
Software compatibility
Check Ada-compatible drivers and containers, plus application licensing. Request the exact OS and isolation configuration needed for your workflow.
Specify the OS, framework and container image version. Arvica confirms the driver and GPU compatibility before activation. A template name is not a software licence or a guaranteed preinstalled image.
SSH & private access
Save an SSH public key during configuration. Keep the private key on your own device. The verified deployment record provides the actual host and login user once access is ready.
Storage & checkpoints
Separate boot disk, datasets, model weights and checkpoints in your capacity plan. Confirm persistence, backup ownership, attachment rules and deletion behaviour before relying on a volume.
Network & data location
Confirm ingress, egress, public IP access and private networking where required. For regulated or location-sensitive data, specify the required region in advanced settings before submitting.
First connection checklist
Once Arvica supplies verified access, connect using your private key and the exact host/user shown in your instance. Then inspect the allocated GPUs and topology.
Replace USER and HOST with your own deployment details. These commands inspect an existing machine; they do not provision one.
Size the job before committing
Model memory
Parameter count × bytes per weight estimates weights only. Add KV cache, activations, temporary tensors and framework overhead. Training additionally needs gradients and optimiser state.
Measure useful throughput
Record model version, precision, input/output length, concurrency, latency and error rate together. Compare the cost of a completed job or successful request under the same conditions.
Complete cost
Include all allocated GPUs, runtime, storage, networking, taxes and support. Monthly projections use 730 hours for comparison; your contract defines the billable period and minimum term.
Model fit and throughput depend on precision, software, batch size, context and system topology. Compare workload estimates rather than an unsupported tokens-per-second promise.
Evaluate tasks whose weights, activations and caches fit within one GPU. For experiments, compare completed-job cost and reproducibility across available configurations.
What should I check about memory?
The standard RTX 4090 has 24 GB GDDR6X. Model size alone does not account for batch size, attention cache and intermediate allocations.
What software should I prepare?
Check Ada-compatible drivers and containers, plus application licensing. Request the exact OS and isolation configuration needed for your workflow.
What interconnect is included?
Confirm the server's PCIe and networking layout for multiple GPUs. Separate graphics capability from any promised multi-node training capability.
Can I stop billing by shutting down the operating system?
Do not assume so. Resource allocation, storage and reserved commitments may continue billing. Check the contract and request the appropriate lifecycle action through support.
Is my data retained after termination?
Retention depends on the agreed storage configuration. Export data and confirm backup and volume retention before requesting termination.
Can I deploy immediately?
Current configurations enter Arvica capacity review. Running status is shown only after infrastructure is verified.
Is the live GPU-hour price my total cost?
Not necessarily. On-demand usage is billed at Arvica's live catalogue rate; storage, network, tax and service charges are itemised before you commit, and reserved or enterprise terms are set in your proposal.
Can I use a larger cluster?
Use the enterprise capacity path for multi-node networking and reserved allocations.
Evaluate tasks whose weights, activations and caches fit within one GPU. For experiments, compare completed-job cost and reproducibility across available configurations.
Evaluate L40S for image generation, rendering and inference applications that can use a graphics-capable GPU. Test the actual model, resolution and concurrency you intend to serve.
Evaluate a single GPU for prototypes and bounded inference jobs. Measure the chosen model, quantisation and concurrency instead of assuming suitability from the generation name.
Evaluate rendering, visual AI and inference pipelines needing more memory than consumer RTX configurations. Validate your renderer and model-serving stack together.
What changes between configurations
Configuration
Practical difference
Confirm before ordering
PCIe · VM · 24 GB GDDR6X
Guest OS and assigned resources within a virtual machine
Exact card, GPU allocation, peer-to-peer connectivity, CPU and disk limits
PCIe · Bare Metal · 24 GB GDDR6X
Dedicated physical host configuration; host access follows the confirmed service
Exact card, GPU allocation, peer-to-peer connectivity, CPU and disk limits
Size the working set
Weights alone are approximately parameters × bytes per parameter. A 7B model at BF16 needs about 14 GB just for weights; a 70B model about 140 GB. Add KV cache, activations, framework buffers and training state. These are arithmetic estimates, not fit guarantees.
Measure useful performance
Compare the same model revision, precision, sequence length and concurrency. Record time to first token, p95 latency, output tokens/s, peak memory and full instance cost. Training comparisons should include step time and convergence at the same quality target.
Evidence status
Manufacturer specifications are linked below. No Arvica application-throughput benchmark is published for this configuration yet. Request a workload-specific acceptance test before making performance commitments.
Configure dedicated PCIe / SXM, virtual-machine and bare-metal capacity with Arvica. Submit your requirements for a deployment proposal.
Request your preferred configuration. Host sizes are illustrative; the written proposal confirms capacity, location, final pricing and delivery. Submitting an enquiry is free and does not reserve capacity.
2 configurations available for enquiry
Capacity confirmation requiredVirtual machine
RTX 4090 PCIe
Price in your proposalYour configuration and term are priced in an Arvica proposal. Delivery schedule confirmed in proposal.
Illustrative configuration
GPU
1 × RTX 4090
GPU memory
24 GB GDDR6X
vCPU
16
RAM
128 GB
Storage
500 GB
AvailabilityConfirmation requiredRegionConfirm in proposalServiceArvica Cloud
