NVIDIA GeForce RTX 4070 SUPER 12GB Specs, Benchmarks & Pricing

• Release Date: January 17, 2024
• MSRP: $599 USD
• GPU Architecture: ada
• Hardware-Accelerated GEMM Operations:
  FP64 FP32 TF32 BF16 FP16 FP8 FP4 INT8 INT4
• CUDA Compute Capability : 8.9

Specifications for NVIDIA GeForce RTX 4070 SUPER

Gaming Benchmarks for NVIDIA GeForce RTX 4070 SUPER

Rankings based on 805,324 real-world benchmark results from OpenBenchmarking.org

References

• https://images.nvidia.com/aem-dam/Solutions/geforce/ada/nvidia-ada-gpu-architecture.pdf
• https://images.nvidia.com/aem-dam/Solutions/Data-Center/l4/nvidia-ada-gpu-architecture-whitepaper-v2.1.pdf
• https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4070-family/
• https://www.nvidia.com/en-us/geforce/news/geforce-rtx-4080-4070-ti-4070-super-gpu/
• https://www.waredb.com/processor/nvidia-geforce-rtx-4070-super
• https://www.techpowerup.com/gpu-specs/geforce-rtx-4070-super.c4214
• https://videocardz.net/nvidia-geforce-rtx-4070-super
• https://www.newegg.com/nvidia-founders-edition-900-1g141-2534-000-geforce-rtx-4070-super-12gb-graphics-card/p/1FT-0004-008P2
• https://www.asus.com/us/motherboards-components/graphics-cards/dual/dual-rtx4070s-o12g/
• https://www.asus.com/us/motherboards-components/graphics-cards/tuf-gaming/tuf-rtx4070s-o12g-gaming/
• https://us.msi.com/Graphics-Card/GeForce-RTX-4070-SUPER-12G-VENTUS-3X-OC/Specification
• https://us.msi.com/Graphics-Card/GeForce-RTX-4070-SUPER-12G-GAMING-X-SLIM/Specification
• https://www.gigabyte.com/Graphics-Card/GV-N407SGAMING-OC-12GD
• https://www.gigabyte.com/Graphics-Card/GV-N407SWF3OC-12GD-rev-10
• https://www.amazon.com/PNY-Graphics-SFF-Ready-DisplayPort-supports/dp/B0CS197RJM
• https://www.newegg.com/pny-geforce-rtx-4070-super-vcg4070s12tfxxpb1-o/p/N82E16814133864
• https://www.zotac.com/us/product/graphics_card/zotac-gaming-geforce-rtx-4070-super-twin-edge-oc-12gb-gddr6x
• https://www.palit.com/palit/vgapro.php?id=4997&lang=en
• https://www.palit.com/palit/vgapro.php?id=5077&lang=en
• https://en.colorful.cn/en/home/product?mid=102&id=008fb798-f96a-49d8-ab36-d6c723b7a909
• https://en.colorful.cn/en/home/product?mid=102&id=aa12009a-02f0-45a2-9bbe-1211583e910f

Notes

1. MSRP of $599 USD sourced from NVIDIA's official announcement. A Founders Edition was available at this price, matching the original RTX 4070's launch price while offering significantly better performance.
2. fp16TFLOPS of 283.84 is FP16 Tensor Core throughput (FP16 accumulate) with 2:4 structured sparsity, 8 × the FP32 vector figure. NVIDIA's RTX 4070 family page gives 568 AI TOPS (FP8 with sparsity), and GeForce Ada runs FP16-accumulate Tensor math at half the FP8 rate: 568 / 2 = 284. Dense FP16 Tensor is 141.92 TFLOPS.
3. int8TOPS of 283.86 represents non-sparse (dense) INT8 performance, derived from sparse INT8 of 567.71 TOPS ÷ 2 per NVIDIA Ada architecture whitepaper. With 2:4 structured sparsity, INT8 performance is 567.71 TOPS. Dense value used for consistent cross-vendor comparison.
4. fp8TFLOPS of 283.86 represents non-sparse (dense) FP8 performance. NVIDIA's Ada Lovelace 4th-generation Tensor Cores process FP8 and INT8 at identical per-clock throughput (both are 8-bit formats) per the NVIDIA Ada GPU Architecture whitepaper; this equal-rate relationship is confirmed by official NVIDIA per-SKU datasheets (e.g. RTX 4090, L40, L40S, L4 all publish identical FP8 and INT8 dense/sparse figures). Dense FP8 = dense INT8 = 283.86 TFLOPS. Dense value used for consistent cross-vendor comparison.
5. Ada Lovelace has no FP4 hardware path (FP4 tensor core support was introduced in Blackwell) — fp4TFLOPS is left null and is a confirmed hardware negative, not a research gap, so it is not added to unverifiedSpecs.
6. Power based on what NVIDIA refers to as "TGP (Total Graphics Power)".
7. The RTX 4070 SUPER uses the AD104-350 variant of the AD104 chip with 56 of 60 streaming multiprocessors enabled.
8. Memory bandwidth calculated as 192-bit interface × 21 Gbps = 504 GB/s.
9. Process node is TSMC 4N (5nm-class) with 35.8 billion transistors.
10. The RTX 4070 SUPER offers significantly better value than the original RTX 4070, with 21% more CUDA cores and 12GB VRAM (vs 12GB) at the same MSRP and power envelope.
11. NVIDIA Part Number (NVPN) 900-1G141-2534-000 sourced from Best Buy and Newegg listings for the Founders Edition.
12. Product SKU PG141 SKU 334 and Board ID PG141 sourced from VideoCardz.net technical specifications.
13. PCI Device ID 2783 and GPU variant AD104-350 sourced from VideoCardz.com and linux-hardware.org.
14. Third-party product identifiers sourced from official manufacturer websites (ASUS, MSI, Gigabyte, Zotac, Palit, Colorful) and retail listings (Amazon, Newegg, Micro Center).
15. EVGA does not manufacture RTX 4070 SUPER cards as they exited the GPU market in September 2022 before the RTX 40 series launch.
16. Gainward RTX 4070 SUPER models (Panther OC, Ghost) exist but specific part numbers were not available from search results.
17. fp32TFLOPS of 35.48 is TF32 Tensor Core throughput without sparsity. It equals FP32 vector (CUDA core) throughput, computed as 7,168 CUDA cores × 2 × 2.475 GHz boost = 35.48 TFLOPS. On GeForce Ada, dense TF32 runs at the FP32 vector rate, as Tables 3 to 6 of NVIDIA's RTX Blackwell whitepaper show (https://images.nvidia.com/aem-dam/Solutions/geforce/blackwell/nvidia-rtx-blackwell-gpu-architecture.pdf).