The 5070 Ti is expected to start at 849 euros (roughly $932) through various retail outlets. Meanwhile, the standard Geforce RTX 5070 will also be available as a Founders Edition model, but its exact release date is yet to be finalized. Nvidia has indicated that the 5070 will represent an entry point into the Blackwell GPU series at 649 euros (about $712), making it accessible for users who seek high-performance graphics at a somewhat lower cost.
Looking ahead, Nvidia is preparing to expand the Blackwell series further with the anticipated Geforce RTX 5060 Ti and 5060. These models are predicted to arrive between March and May of 2025, filling out the lower end of the product range. Market observers expect that these future cards will continue to incorporate the advanced architectural benefits seen in the higher-tier products, such as refined ray tracing cores and efficient power usage.
| Geforce RTX 5090 | Geforce RTX 5080 | Geforce RTX 5070 Ti | Geforce RTX 5070 | |
|---|---|---|---|---|
| GPU codename | GB202 | GB203 | GB203 | GB205 |
| GPU architecture | Nvidia Blackwell | Nvidia Blackwell | Nvidia Blackwell | Nvidia Blackwell |
| GPCs | 11 | 7 | 6 | 5 |
| TPCs | 85 | 42 | 35 | 24 |
| SMS | 170 | 84 | 70 | 48 |
| CUDA Cores / SM | 128 | 128 | 128 | 128 |
| CUDA Cores / GPU | 21,760 | 10,752 | 8,960 | 6,144 |
| Tensor Cores / SM | 4 (5th Gen) | 4 (5th Gen) | 4 (5th Gen) | 4 (5th Gen) |
| Tensor Cores / GPU | 680 (5th Gen) | 336 (5th Gen) | 280 (5th Gen) | 192 (5th Gen) |
| RT Cores | 170 (4th Gen) | 84 (4th Gen) | 70 (4th Gen) | 48 (4th Gen) |
| boost clock | 2,407 MHz | 2,617 MHz | 2,452 MHz | 2,512 MHz |
| Peak FP32 TFLOPS | 104.8 | 56.3 | 43.9 | 30.9 |
| Peak FP16 TFLOPS | 104.8 | 56.3 | 43.9 | 30.9 |
| Peak BF16 TFLOPS | 104.8 | 56.3 | 43.9 | 30.9 |
| Peak INT32 TOPS | 104.8 | 56.3 | 43.9 | 30.9 |
| RT TFLOPS | 317.5 | 170.6 | 133.2 | 93.6 |
| Peak FP4 Tensor TFLOPS | 1,676/3,352 | 900.4/1.8012 | 703/1.406 | 493.9/987.8 |
| Peak FP8 Tensor TFLOPS (FP16) | 838/1,676 | 450.2/900.4 | 351.5/703 | 246.9/493.9 |
| Peak FP8 Tensor TFLOPS (FP32) | 419/838 | 225.1/450.2 | 175.8/351.5 | 123.5/246.9 |
| Peak FP16 Tensor TFLOPS (FP16) | 419/838 | 225.1/450.2 | 175.8/351.5 | 123.5/246.9 |
| Peak FP16 Tensor TFLOPS (FP32) | 209.5/419 | 112.6/225.1 | 87.9/175.8 | 61.7/123.5 |
| Peak BF16 Tensor TFLOPS (FP32) | 209.5/419 | 112.6/225.1 | 87.9/175.8 | 61.7/123.5 |
| Peak TF32 Tensor TFLOPS | 104.8/209.5 | 56.3/112.6 | 43.9/87.9 | 30.9/61.7 |
| Peak INT8 Tensor TOPS | 838/1,676 | 450.2/900.4 | 351.5/703 | 246.9/493.9 |
| Frame Buffer Memory Size | 32 GiB GDDR7 | 16 GiB GDDR7 | 16 GiB GDDR7 | 12 GiB GDDR7 |
| Memory Interface | 512-bit | 256-bit | 256-bit | 192-bit |
| Memory Clock (Data Rate) | 28 Gbps | 30 Gbps | 28 Gbps | 28 Gbps |
| Memory Bandwidth | 1,792 GiB/s | 960 GiB/s | 896 GiB/s | 672 GiB/s |
| ROPs | 176 | 112 | 96 | 80 |
| Pixel Fill Rate (Gigapixels/s) | 423.6 | 293.1 | 235.4 | 201 |
| texture units | 680 | 336 | 280 | 192 |
| Texel fill rate (Gigatexels/s) | 1636.8 | 879.3 | 686.6 | 482.3 |
| L1 Data Cache/Shared Memory | 21,760 KiB | 10,752 KiB | 8,960 KiB | 6,144 KiB |
| L2 Cache Size | 98,304 KiB | 65,536 KiB | 49,152 KiB | 49,152 KiB |
| Register File Size | 43,520 KiB | 21,504 KiB | 17,920 KiB | 12,288 KiB |
| video engines | 3× NVENC (9th Gen) 2× NVDEC (6th Gen) | 2× NVENC (9th Gen) 2× NVDEC (6th Gen) | 2× NVENC (9th Gen) 1× NVDEC (6th Gen) | 1× NVENC (9th Gen) 1× NVDEC (6th Gen) |
| TGP (Total Graphics Power) | 575 watts | 360 watts | 300 watts | 250 watts |
| number of transistors | 92.2 billion | 45.6 billion | 45.6 billion | 31.1 billion |
| chip area (die size) | 750 mm² | 378 mm² | 378 mm² | 263 mm² |
| manufacturing process | TSMC 4nm (4N) | TSMC 4nm (4N) | TSMC 4nm (4N) | TSMC 4nm (4N) |
| PCI Express interface | Gen 5 x16 | Gen 5 x16 | Gen 5 x16 | Gen 5 x16 |
You can check out the white paper (.PDF) here.
