Start Building
In stock · Provisions in ~5 min

NVIDIA Blackwell, 96 GB GDDR7, PCIe Gen5

NVIDIA RTX 6000 ProMost memory of any PCIe GPU.

The NVIDIA RTX 6000 Pro is a Blackwell-generation workstation GPU with 96 GB of GDDR7 memory and 1.79 TB/s of bandwidth, the most memory of any PCIe GPU. It runs 30B–70B models natively without quantisation, making it ideal for development, fine-tuning, and cost-efficient inference. Available on packet.ai from $0.66/GPU-hour.

From $0.66/GPU-hr≈ 71% below H100 cost

Dynamic $0.66/hr · Monthly $299/mo

96GB
GDDR7 memory
1.79TB/s
Memory bandwidth
96TFLOPS
FP32 compute
300W
TDP
Architecture

Blackwell compute, PCIe form factor.

The RTX 6000 Pro brings next-generation Tensor Cores and 96 GB of GDDR7 memory to a standard PCIe card.

Blackwell Tensor Cores

Next-gen Tensor Cores with FP4 and FP8 support. Blackwell inference in PCIe form factor for the first time.

96 GB GDDR7 at 1.79 TB/s

96 GB GDDR7 runs 30B models at FP16 and 70B at 4-bit on a single PCIe card, no NVLink required.

PCIe Gen5 form factor

Drops into any Gen5 server without SXM motherboards, offering the widest deployment flexibility.

AV1 + DLSS 4

Hardware AV1 encode and DLSS 4 make the RTX 6000 Pro uniquely capable for AI video inference.

Technical specs

NVIDIA RTX 6000 Pro specifications.

SpecificationValueGreat for
GPU architecture
NVIDIA Blackwell
Blackwell Tensor Cores with FP4/FP8 for next-gen inference.
GPU memory
96 GB GDDR7
30B at FP16 and 70B at 4-bit, no sharding needed.
Memory bandwidth
1.79 TB/s
High bandwidth for large model token generation.
FP32 compute
96 TFLOPS
Strong general-purpose and inference compute.
Host interface
PCIe Gen5 x16
Next-gen PCIe bandwidth in standard server form factor.
Power
300W TDP
High throughput within a 300 W envelope.
Pricing

Three ways to run RTX 6000 Pro.

Hourly or monthly, shared or dedicated, plus multi-node clusters.

DynamicMonthly · Shared
$299 /month
~37% off hourly rate

Same shared RTX 6000 Pro at a flat monthly rate. Predictable billing, cancel anytime.

Deploy Monthly →
Multi-node Cluster
From 8 GPUs

Scale inference across multiple RTX 6000 Pro nodes with InfiniBand interconnect.

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

What the RTX 6000 Pro is built for.

Large-model inference

96 GB runs 30B at FP16 and 70B at 4-bit without NVLink, at a fraction of H100 cost.

  • 96 GB, most of any PCIe GPU
  • 30B at FP16 natively
  • 70B at 4-bit on one card

Fine-tuning & research

Blackwell FP8 Tensor Cores and 96 GB enable full fine-tuning of 30B+ models without multi-GPU setups.

  • LoRA & QLoRA native
  • 30B+ fine-tuning
  • Single-card simplicity

Video AI pipelines

Hardware AV1 encode and DLSS 4 make it the best PCIe GPU for AI video generation and rendering.

  • AV1 hardware encode
  • DLSS 4 support
  • Real-time throughput
FAQ

RTX 6000 Pro, answered.

For anything else, reach help@packet.ai.

What is the NVIDIA RTX 6000 Pro?

Blackwell workstation GPU: 96 GB GDDR7, 1.79 TB/s, most memory of any single PCIe GPU.

How much does it cost?

$0.66/GPU-hour dynamic, $1.25/hr dedicated, or $299/month flat rate.

What models fit in RTX 6000 Pro?

30B at FP16 natively, 70B at 4-bit. For larger models, use H200 or B200.

Does it have NVLink?

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

Can RTX 6000 Pro be used for training?

Yes. Blackwell Tensor Cores with FP8 make it ideal for LoRA, QLoRA, and full fine-tuning of 30B–70B models on a single card.

How fast can I deploy?

SSH-ready in under 5 minutes on Dynamic or Dedicated hourly. Multi-node clusters provision in under 1 hour.

Run the RTX 6000 Pro. 96 GB Blackwell from $0.66/hr.

The most memory of any PCIe GPU. $0.66/hr dynamic, or $299/mo flat.

On-demand · hourly billing · US & EU regions

NVIDIA RTX 6000 Profrom $0.66/GPU-hr In stock
Deploy RTX 6000 Pro →