Start Building →
Coming soon — join waitlist

NVIDIA Blackwell, 32 GB GDDR7, PCIe Gen5

NVIDIA RTX 5090The most powerful consumer GPU ever built.

The NVIDIA RTX 5090 is NVIDIA’s flagship Blackwell consumer GPU — 32 GB of GDDR7 memory at 1.79 TB/s of bandwidth, 5th-generation Tensor Cores with native FP4 support, and 21,760 CUDA cores. Coming soon to packet.ai from $0.59/GPU-hour dedicated.

From $0.59/GPU-hrComing soon

Dedicated $0.59/hr · Pricing to be confirmed at launch

Join waitlist →See pricing
32GB
GDDR7 memory
1.79TB/s
Memory bandwidth
3,352TOPS
AI compute (FP4)
575W
TDP
Architecture

Blackwell for the consumer frontier.

The RTX 5090 brings 5th-gen Tensor Cores, native FP4, and 1.79 TB/s GDDR7 to a PCIe card — the most capable consumer GPU for AI inference ever made.

5th-gen Tensor Cores + FP4

Native FP4 precision delivers 3,352 AI TOPS — 2× the AI throughput of RTX 4090. First consumer GPU with FP4 support.

32 GB GDDR7 at 1.79 TB/s

78% more bandwidth than RTX 4090. Fits 13B at FP16 and 32B at Q4 on a single card.

PCIe Gen5 form factor

Drops into any PCIe Gen5 system — no SXM motherboard required. Accessible Blackwell for any server environment.

DLSS 4 + Multi Frame Generation

4th-gen RT Cores and hardware AV1 encode make the RTX 5090 the best consumer GPU for AI video and image generation.

Technical specs

NVIDIA RTX 5090 specifications.

SpecificationValueGreat for
GPU architecture
NVIDIA Blackwell
5th-gen Tensor Cores with native FP4.
GPU memory
32 GB GDDR7
13B at FP16 native. 32B at Q4 on one card.
Memory bandwidth
1.79 TB/s
78% faster than RTX 4090. Near H100 PCIe.
FP32 compute
104.8 TFLOPS
Strong inference and general-purpose compute.
AI compute (FP4 sparse)
3,352 TOPS
Highest AI throughput of any consumer GPU.
CUDA cores
21,760
Massive parallel compute for training.
Host interface
PCIe Gen5 x16
No SXM board needed. Fits any Gen5 server.
Power
575W TDP
Peak Blackwell perf. Needs 1000W PSU.
Pricing

Three ways to run RTX 5090.

Dedicated single-tenant — plus multi-node clusters.

Coming soon
DedicatedHourly · Single-tenant
$0.59 /GPU-hr

Full RTX 5090 reserved exclusively for you. Zero noisy-neighbour risk, 99.99% SLA.

Join waitlist →
DedicatedMonthly · Single-tenant
TBC /month

Reserved RTX 5090 at a flat monthly rate. Full single-tenant isolation, predictable cost exclusively for you. 99.99% SLA, zero noisy-neighbour risk.

Launching soon
Multi-node Cluster
From 8 GPUs

Scale inference across multiple RTX 5090 nodes with InfiniBand interconnect.

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

What the RTX 5090 is built for.

Sub-30B LLM inference

1.79 TB/s bandwidth and 32 GB VRAM make the 5090 the fastest consumer GPU for token generation on 7B–13B models at FP16.

  • 13B at FP16 natively
  • 32B at Q4 on one card
  • ~1.8× faster than RTX 4090

Fine-tuning & LoRA

32 GB gives you headroom for LoRA and QLoRA fine-tuning of 13B models without sharding — at a fraction of H100 cost.

  • LoRA / QLoRA native
  • Blackwell FP8 Tensor Cores
  • Hourly billing — no waste

AI image & video generation

DLSS 4 Multi Frame Generation, 4th-gen RT Cores, and hardware AV1 encode make the RTX 5090 the best consumer GPU for FLUX, SDXL, and video AI.

  • FLUX / SDXL natively
  • AV1 hardware encode
  • DLSS 4 Multi Frame Gen
FAQ

NVIDIA RTX 5090, answered.

For anything else, reach help@packet.ai.

What is the NVIDIA RTX 5090?

Blackwell flagship consumer GPU: 32 GB GDDR7, 1.79 TB/s, 3,352 AI TOPS. The most powerful consumer GPU ever built.

When will RTX 5090 be available on packet.ai?

Coming soon. Join the waitlist to be notified the moment capacity opens.

How much does RTX 5090 cost?

Dedicated from $0.59/GPU-hour, single-tenant. Monthly pricing TBC at launch. See pricing →

What models fit in RTX 5090?

13B at FP16 natively; 32B at Q4 on a single card. For 70B+, use H100 or H200.

How does RTX 5090 compare to H100?

H100 has more memory (80 GB vs 32 GB), ECC, NVLink, and 24/7 datacenter reliability. RTX 5090 wins on cost-per-token for sub-30B inference and generative AI workloads.

Does RTX 5090 support NVLink?

No. PCIe Gen5 only. For NVLink multi-GPU, use H100 SXM or B200.

RTX 5090 — coming soon.

The most powerful consumer GPU ever built. Join the waitlist for early access on packet.ai.

Join waitlist →Talk to a human

No commitment · we’ll notify you by email

NVIDIA RTX 5090from $0.59/GPU-hr Coming soon
Join waitlist →