Twelve gigabytes on a 192-bit bus is the constraint to plan around. The 5070 has more compute than its memory can keep fed for larger models, so it suits quantised 7B to 13B inference rather than anything that needs a large KV cache.

NVIDIA GeForce RTX 5070 12GB

Blackwell PCIe 2025 4nm Spec confidence: Official
FP8 (dense)
247
TFLOPS
FP32
31
TFLOPS
VRAM
12 GB
GDDR7
Bandwidth
672 GB/s
memory
TDP
250 W
123.6 TFLOPS/kW

Overview

The GeForce RTX 5070 is the flagship use of the GB205 die, with 48 of its 50 SMs enabled: 6,144 Blackwell CUDA cores, 192 fifth-generation Tensor cores, and 12 GB of GDDR7 on a 192-bit bus running at 672 GB/s inside a 250 W board. NVIDIA publishes its whole precision ladder in the RTX Blackwell whitepaper, so the 246.9 TFLOPS of dense FP8 and 493.9 TFLOPS of dense FP4 below are read straight from a vendor table rather than halved out of an AI TOPS headline. It is the cheapest card in the range with a die of its own rather than a cut of a bigger one, and the 31.1 billion transistor GB205 is notably efficient per watt. The 12 GB frame buffer is the honest limitation for AI use, and it is the same capacity the RTX 4070 shipped with two years earlier.

Performance

Peak theoretical throughput by precision type

PrecisionDense2:4 Sparse
FP64
64-bit floating point
0.5TFLOPS Structured sparsity is a tensor-core feature; this vector precision has no sparse form
FP32
32-bit floating point
31TFLOPS Structured sparsity is a tensor-core feature; this vector precision has no sparse form
TF32
TensorFloat-32
31TFLOPS 62TFLOPS
BF16
Brain Float 16
62TFLOPS 124TFLOPS
FP16
16-bit floating point
124TFLOPS 247TFLOPS
FP8
8-bit floating point
247TFLOPS 494TFLOPS
FP6
No verified data available The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision
FP4
4-bit floating point
494TFLOPS 988TFLOPS
INT8
8-bit integer
247TOPS 494TOPS

The GeForce RTX 5070 12GB in the GPU landscape

Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper

06001,2001,8002,4003,0000 W250 W500 W750 W1000 W1250 W1500 WInstinct MI355XGeForce RTX 5070 12GB

Higher and further left is better: more half-precision throughput for less power.

Specifications

Architecture

Blackwell

Form Factor

PCIe

Launch Year

2025

Process Node

4nm

Memory

12 GB GDDR7

Bandwidth

672 GB/s

TDP

250 W

Max Power

250 W

Transistors

31.1 billion

CUDA Cores

6,144

Spec Confidence

Official

Full Specifications

Compute Engine
CUDA Cores 6,144
Tensor Cores 192 (5th Gen)
Streaming Multiprocessors 48
Base Clock 2.33 GHz
Boost Clock 2.51 GHz
Chip Design
Transistors 31.1 billion
Die Size 263 mm²
Process Node 4nm
Memory
VRAM 12 GB
Memory Type GDDR7
Bandwidth 672 GB/s
Interface Width 192-bit
Memory Clock 28 GT/s
Interconnect & I/O
PCIe 5.0
Cache
L2 Cache 48 MB
Power & Thermal
TDP 250 W
Max Board Power 250 W
Cooling Active
Enterprise Features
Sparsity Yes
General
Form Factor PCIe
Architecture Blackwell
Process Node 4nm
Launch Year 2025

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
NVIDIA
Published
2025-05-12

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
8 of 9
Normalization
All values in TFLOPS

Rent this GPU

$0.11 / GPU-hr

Lowest of 1 live listing across 1 provider

See all rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Similar GPUs

Frequently Asked Questions

How many TFLOPS does the NVIDIA GeForce RTX 5070 have?

The NVIDIA GeForce RTX 5070 delivers 31 TFLOPS FP32, 124 TFLOPS FP16 and 247 TFLOPS FP8 at peak.

What is the power consumption of the NVIDIA GeForce RTX 5070?

The NVIDIA GeForce RTX 5070 has a TDP (Thermal Design Power) rating of 250 watts.

How much memory does the NVIDIA GeForce RTX 5070 have?

The NVIDIA GeForce RTX 5070 is equipped with 12 GB of VRAM with 672 GB/s of memory bandwidth.

What architecture is the NVIDIA GeForce RTX 5070 based on?

The NVIDIA GeForce RTX 5070 is based on the Blackwell architecture, launched in 2025.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs