NVIDIA GeForce RTX 3070 Ti vs NVIDIA GeForce RTX 4070

Every spec side by side, from each maker's published figures.

Only show differences

Better figure, where one is clearly better

Share comparison

GeForce RTX 3070 Ti compared with GeForce RTX 4070

	NVIDIAGeForce RTX 3070 Ti	NVIDIAGeForce RTX 4070
Overview
Released
Released	June 2021	April 2023
Performance
Chip maker
Chip maker	NVIDIA	NVIDIA
Architecture
Architecture	Ampere	Ada Lovelace
Shader cores
Shader cores	6,144	5,888
Boost clock
Boost clock	1,770 MHz	2,475 MHz (better)
Architecture
Architecture	NVIDIA reference specifications. The photo shows PNY's stock-clock XLR8 Gaming REVEL EPIC-X RGB Triple Fan; its cooler differs from the Founders Edition geometry below.	AD104 GPU; 46 SMs. NVIDIA reference specifications; photo illustrates an ASUS DUAL partner cooler without establishing its memory type or factory overclock.
Manufacturing process
Manufacturing process	–	TSMC 4N NVIDIA custom process; 35.8 billion transistors; 294.5 mm² die
Ray tracing
Ray tracing	Second-generation RT Cores	46 third-generation RT Cores; 67 RT TFLOPS
AI cores
AI cores	Third-generation Tensor Cores	184 fourth-generation Tensor Cores; 466 AI TOPS
Upscaling and frame generation
Upscaling and frame generation	NVIDIA DLSS upscaling; no DLSS Frame Generation	DLSS 3.5 with Super Resolution, DLAA, Ray Reconstruction and Frame Generation
Base clock
Base clock	NVIDIA lists 1.58 GHz rounded; stock-clock PNY REVEL specifies 1575 MHz base and 1770 MHz boost	1.92 GHz; stock PNY and Gainward cards confirm 2,475 MHz boost, while NVIDIA rounds it to 2.48 GHz
Cache
Cache	–	36 MB L2
Memory
Video memory
Video memory	8 GB GDDR6X	12 GB GDDR6X or 12 GB GDDR6 (better)
Memory bus
Memory bus	256-bit (better)	192-bit
Memory bandwidth
Memory bandwidth	608 GB/s	504 GB/s or 480 GB/s
Memory detail
Memory detail	19 Gbps GDDR6X; 608 GB/s bandwidth verified by PNY's stock-clock REVEL specification	GDDR6X at 21 Gbps and 504 GB/s (stock PNY); later GDDR6 at 20 Gbps and 480 GB/s (stock Gainward). Both have 12 GB on a 192-bit bus
Connections and power
Board power
Board power	290 W	200 W (better)
Interface
Interface	PCIe 4.0	PCIe 4.0
Display outputs
Display outputs	3 × DisplayPort 1.4a, 1 × HDMI 2.1; up to 4 displays; HDCP 2.3	3 × DisplayPort 1.4a, 1 × HDMI 2.1a; up to 4 displays; HDCP 2.3
Maximum resolution
Maximum resolution	7680 × 4320; HDMI supports 4K 120 Hz HDR or 8K 60 Hz HDR; DisplayPort supports 4K 240 Hz with DSC	4K at 240 Hz or 8K at 60 Hz with DSC, HDR
Power connector
Power connector	2 × PCIe 8-pin cables through the included adapter to 1 × 12-pin (Founders Edition)	1 × 300 W or greater PCIe Gen 5 cable, or 2 × PCIe 8-pin cables with the adapter in the box; some partner cards use 1 × 8-pin
Recommended power supply
Recommended power supply	750 W, based on NVIDIA's Intel Core i9-10900K test system	650 W
Slot width
Slot width	2-slot (Founders Edition)	2-slot (original GDDR6X Founders Edition)
Size (reference card)
Size (reference card)	267 × 112 mm, 2-slot (Founders Edition)	244 × 112 mm, 2-slot (original GDDR6X Founders Edition)
Video encode and decode
Video encode and decode	Seventh-generation NVENC and fifth-generation NVDEC; no hardware AV1 encoding	1 × eighth-generation NVENC, 1 × fifth-generation NVDEC; AV1 encode and decode