Start Building
In stock · Provisions in ~5 min

NVIDIA Ada Lovelace, 24 GB GDDR6X, PCIe Gen4

NVIDIA RTX 4090The fastest consumer GPU for AI workloads.

The NVIDIA RTX 4090 is the flagship Ada Lovelace consumer GPU with 24 GB of GDDR6X memory at 1.01 TB/s of bandwidth, 4th-generation Tensor Cores with FP8, and 16,384 CUDA cores. Available on packet.ai from $0.39/GPU-hour dedicated.

From $0.39/GPU-hrBest value Ada GPU

Dedicated $0.39/hr · Monthly $263/mo

24GB
GDDR6X memory
1.01TB/s
Memory bandwidth
82.6TFLOPS
FP32 compute
450W
TDP
Architecture

Ada Lovelace, built for AI.

The RTX 4090 brings 4th-gen Tensor Cores, FP8 support, and 24 GB GDDR6X to a PCIe card. The most capable consumer GPU for inference and generative AI.

4th-gen Tensor Cores + FP8

Ada Tensor Cores with FP8 deliver up to 1,321 AI TOPS. Strong enough for 7B model serving and LoRA fine-tuning.

24 GB GDDR6X at 1.01 TB/s

Fits 7B models at FP16 and 13B at 4-bit on one card. The sweet spot for consumer-grade LLM inference.

PCIe Gen4 form factor

No SXM board required. Drops into any Gen4 server for fast, flexible deployment.

DLSS 3 + AV1 encode

Hardware AV1 encode and DLSS 3 Frame Generation make the RTX 4090 ideal for FLUX, SDXL, and video AI pipelines.

Technical specs

NVIDIA RTX 4090 specifications.

SpecificationValueGreat for
GPU architecture
NVIDIA Ada Lovelace
4th-gen Tensor Cores with FP8.
GPU memory
24 GB GDDR6X
7B at FP16 native. 13B at Q4 on one card.
Memory bandwidth
1.01 TB/s
Fast token generation for small-batch inference.
FP32 compute
82.6 TFLOPS
Strong general-purpose and inference compute.
AI compute (FP8 sparse)
1,321 TOPS
Best-in-class consumer AI throughput.
CUDA cores
16,384
Massive parallel compute for training and rendering.
Host interface
PCIe Gen4 x16
No SXM board needed. Fits any Gen4 server.
Power
450W TDP
Peak Ada perf. Needs 850W PSU minimum.
Pricing

Three ways to run RTX 4090.

Dedicated single-tenant, plus multi-node clusters.

DedicatedMonthly · Single-tenant
$263 /month
~43% off hourly rate

Reserved RTX 4090 at a flat monthly rate. Predictable cost, full single-tenant isolation.

Deploy Monthly →
Multi-node Cluster
From 8 GPUs

Scale inference across multiple RTX 4090 nodes with InfiniBand interconnect.

  • 8–512 GPUs per cluster
  • InfiniBand interconnect
  • Provisioned in <1 hr
Get a wholesale quote →
Use cases

What the RTX 4090 is built for.

7B–13B LLM inference

24 GB fits 7B at FP16 natively and 13B at Q4. The most popular model sizes at the lowest cost per token.

  • 7B at FP16 native
  • 13B at Q4 on one card
  • $0.39/hr dedicated

Fine-tuning & LoRA

24 GB gives headroom for LoRA and QLoRA fine-tuning of 7B models at the lowest hourly rate on packet.ai.

  • LoRA / QLoRA native
  • Ada FP8 Tensor Cores
  • Hourly billing, no waste

AI image & video generation

DLSS 3 Frame Generation, hardware AV1 encode, and 24 GB VRAM make the RTX 4090 a workhorse for FLUX, SDXL, and video AI.

  • FLUX / SDXL natively
  • AV1 hardware encode
  • DLSS 3 Frame Gen
FAQ

NVIDIA RTX 4090, answered.

For anything else, reach help@packet.ai.

What is the NVIDIA RTX 4090?

Ada Lovelace flagship consumer GPU: 24 GB GDDR6X, 1.01 TB/s, 1,321 AI TOPS.

How much does RTX 4090 cost on packet.ai?

Dedicated from $0.39/GPU-hour, or $263/month flat rate. Single-tenant, 99.99% SLA.

What models fit in RTX 4090?

7B at FP16 natively; 13B at Q4 on a single card. For larger models, use L40S or H100.

How does RTX 4090 compare to RTX 5090?

RTX 5090 has more memory (32 GB vs 24 GB), faster bandwidth, and 5th-gen Tensor Cores with FP4. RTX 4090 wins on price at $0.39/hr vs $0.59/hr.

Is RTX 4090 good for fine-tuning?

Yes. LoRA and QLoRA fine-tuning of 7B models fits comfortably in 24 GB at the lowest hourly rate on packet.ai.

Does RTX 4090 support NVLink?

No. PCIe Gen4 only. For NVLink multi-GPU, use H100 SXM or B200.

Run the RTX 4090. 24 GB Ada from $0.39/hr.

The best value consumer GPU on packet.ai. $0.39/hr dedicated, or $263/mo flat.

On-demand · hourly billing · US & EU regions

NVIDIA RTX 4090from $0.39/GPU-hr In stock
Deploy RTX 4090 →