NVIDIAAda Lovelace2022

RTX 6000 Ada

Current median: $1.57/hr across 7 providers (observed), 0.00 (0.0%) vs yesterday. Typical range: $0.74–$3.54/hr (P10–P90). 30-day trend unlocks after 30 days of captures

TDP 300W · 182.2 TFLOPS · 48GB VRAM · 960 GB/s bandwidth

The RTX 6000 Ada has a 300W TDP, 182.2 TFLOPS of FP16 compute, 48GB of GDDR6 VRAM, 960 GB/s memory bandwidth.

Flagship Ada professional GPU. 48GB GDDR6 with NVLink — two cards give 96GB unified. Excellent for 70B+ model inference.

VRAM

48GB

GDDR6

FP16

182.2

TFLOPS

Bandwidth

960.0

GB/s

TDP

300W

power

Best for:Large model inferenceMulti-GPU professional workloadsHigh-VRAM fine-tuning

RTX 6000 Ada currently rents for a median of $1.57/hr across 7 providers (observed on-demand prices). RTX 6000 Ada has 48GB of memory. Compare its specs and listed cloud rental prices below. The workload must fit in VRAM with room for runtime memory; a listed price does not confirm stock.

RTX 6000 Ada Overview

The RTX 6000 Ada is NVIDIA’s flagship Ada Lovelace professional card, combining 48GB of ECC GDDR6 with 182.2 TFLOPS of FP16/BF16 tensor performance in a 300W workstation-friendly design. It offers a professional reliability and driver profile that consumer AD102 cards do not.

Its 960 GB/s memory bandwidth is healthy for GDDR6 and the 48GB capacity accommodates substantially larger workloads than 24GB cards. NVLink provides a path for paired-card configurations, but bandwidth remains well below HBM data-center platforms, so multi-GPU language-model workloads should be assessed for communication overhead rather than assumed to scale linearly.

This GPU is a strong fit for high-VRAM fine-tuning, professional visualization plus AI, and workstation-scale LLM inference. It is less attractive than HBM hardware for bandwidth-bound large-model serving and lacks the dedicated FP8 positioning of Hopper data-center products.

Memory

VRAM48 GB

Memory TypeGDDR6

Bandwidth960 GB/s

NVLink BW112 GB/s

Compute Performance

FP3291.1 TFLOPS

FP16182.2 TFLOPS

BF16182.2 TFLOPS

INT8364 TOPS

Hardware Specifications

Chip

ArchitectureAD102

GenerationAda Lovelace

Process NodeTSMC 4N

Transistors76.3B

Die Size608 mm²

Release DateDecember 19, 2022

Launch MSRP$6,800

Processors

CUDA / Shader Cores18,176

Tensor Cores568

RT Cores142

Clocks

Base Clock1,455 MHz

Boost Clock2,505 MHz

Memory

VRAM48 GB

Memory TypeGDDR6

Memory Bus384-bit

Bandwidth960 GB/s

L2 Cache96 MB

NVLink BW112 GB/s

Power

TDP300 W

InterconnectNVLink 4.0 / PCIe 4.0

Relative Performance

FP16 Compute2%

VRAM Capacity17%

Mem Bandwidth6%

Relative to highest-spec GPU in database

Limitations

GDDR6 memory bandwidth far below HBM alternatives

Professional but older Ampere architecture — no FP8 support

Higher cost than consumer Ada equivalents for same compute

Live Cloud PricingOn-demand hourly rates

Live9 providers · 25% price spread

Provider	On-Demand	Spot	Region	Availability
BESTSwissGPU	$1.13/hr	—	CH	unknown
FluidStack	$3.60/hr	—	EU/US	unknown
Runcrate	$1.04/hr	—	Kansas City, KS, United States, Des Moines, IA, United States, Las Vegas, NV, United States, US Midwest, Seattle, WA, United States	unknown
Modal	$3.03/hr	—	us-east	unknown
DigitalOcean	$1.57/hr	—	US	unknown
TensorDock	$3.50/hr	—	Various	unknown
RunPod	$0.69/hr	—	US	unknown
QuantaCloud	$0.78/hr	—	US/Midwest	unknown
Verda	$1.41/hr	—	EU-North	unknown

Cheapest / month

$825

$1.13/hr × 730h

Median / month

$1146

at median price

Spot savings

N/A

vs on-demand (best)

Job Cost Estimator

At median $1.57/hr across 7 providers (observed on-demand). Spot pricing typically 40–70% lower.

7B LLM fine-tune

8× GPU, 24 h · Full fine-tune, single node

$301

spot ~$136

70B LLM fine-tune

64× GPU, 72 h · Multi-node, 3-day run

$7.2k

spot ~$3.3k

Inference server

1× GPU, 730 h · 1 month continuous

$1.1k

spot ~$516

Image gen batch

4× GPU, 8 h · Batch generation job

$50

spot ~$23

Estimates only. Actual cost depends on provider, region, availability, and utilization. Use the GPU cost calculator for custom inputs.

Compare RTX 6000 Ada vs…

View all GPU comparisons

Use Case Guidance

Large model inference

Multi-GPU professional workloads

High-VRAM fine-tuning

LLM Model Size Guidance

Max model (FP16)~24Bparameters at FP16 precision

Max model (INT8)~48Bparameters at INT8 precision

Max model (INT4)~96Bparameters at INT4/GGUF

Estimates only. Actual capacity depends on context length, KV cache, and framework overhead.

LLM APIs Running on This GPU Class

Providers that serve frontier LLM inference on Ada Lovelace-class hardware.

OpenAIGPT-4o, o3, o4-mini

AnthropicClaude 3.5 Sonnet, Claude 3 Opus

GoogleGemini 2.0 Flash, Gemini 1.5 Pro

Together AILlama 3.3, Qwen, Mistral

Fireworks AILlama 3.1, DeepSeek, Mixtral

Browse all 42 LLM models

RTX 6000 Ada vs Alternatives — Spec Comparison

Spec	RTX 6000 Ada this	L40S	L40	RTX 5880 Ada
VRAM	48GB GDDR6▲	48GB GDDR6▲	48GB GDDR6▲	48GB GDDR6▲
Memory Bandwidth	960 GB/s▲	864 GB/s	864 GB/s	960 GB/s▲
FP16 TFLOPS	182.2	183▲	181	181
BF16 TFLOPS	182.2	183▲	181	181
FP8 TFLOPS	—	—	—	—
INT8 TOPS	364	366▲	362	362
TDP	300W	350W▲	300W	285W
Process Node	TSMC 4N▲	TSMC 4N▲	TSMC 4N▲	TSMC 4N▲
Architecture	AD102▲	AD102▲	AD102▲	AD102▲
Release Year	2022	2023	2022	2024▲
Max model (FP16)	~24B params▲	~24B params▲	~24B params▲	~24B params▲
Max model (INT4)	~96B params▲	~96B params▲	~96B params▲	~96B params▲

▲ indicates best value in row · FP16/BF16 TFLOPS at full precision · Max model estimates at 2 bytes/param (FP16) and 0.5 bytes/param (INT4)Full side-by-side comparison