The specifications of NVIDIA’s upcoming GeForce RTX 5070 and 5070 Ti graphics card have materialized. The data comes from reputed tipster @kopite7kimi, implying that fourth Blackwell gaming GPUs will be unveiled at CES next month. The numbers are similar to the RTX 4070 Super/RTX 4070 Ti Super, and the primary differences boil down to the faster GDDR7X memory. The L2 cache may be consolidated as well, improving hit-rates at higher resolutions.
NVIDIA GeForce RTX 5070 Ti Specifications
The GeForce RTX 5070 Ti is expected to feature 8,960 FP32 cores across 70 SMs paired with 16 GB GDDR7 memory across a 256-bit bus. The memory will be clocked at 28 Gbps, producing an external bandwidth of 896 GB/s. For reference, the RTX 4070 Ti Super consists of 8,448 shaders (66 SMs) and 16 GB of GDDR6 21 Gbps memory. It has a peak memory bandwidth of 672 GB/s and TBP of 285W, compared to 300W for the 5070 Ti.
The RTX 5070 Ti should be clocked a notch higher than the RTX 4070 Ti Super, perhaps closer to the 3 GHz mark. The RTX 4070 Ti Super has a boost clock of 2,610 MHz, and 48 MB of L2 cache.
| NVIDIA RTX 5090 | NVIDIA RTX 5080 | NVIDIA RTX 5070 Ti | NVIDIA RTX 4070 Ti Super | NVIDIA RTX 5070 | NVIDIA RTX 4070 Super | |
|---|---|---|---|---|---|---|
| GPU | GB202-300 | GB203-400 | GB203-300 | AD103-275 | GB205-300 | AD104-350 |
| SMs | 170 | 84 | 70 | 66 | 48 | 56 |
| CUDA Cores | 21,760 | 10,752 | 8,960 | 8,448 | 6,144 | 7,168 |
| Memory | 32 GB @ 28 Gbps | 16 GB @ 32 Gbps | 16 GB @ 28 Gbps | 16 GB @ 21 Gbps | 12 GB @ 28 Gbps | 12 GB @ 21 Gbps |
| Memory Bus | 512-bit | 256-bit | 256-bit | 256-bit | 192-bit | 192-bit |
| Bandwidth | 1792 GB/s | 1024 GB/s | 896 GB/s | 672 GB/s | 672 GB/s | 504 GB/s |
| TBP | 600W | 400W | 300W | 285W | 250W | 220W |
| Launch | Q1 2025 | Q1 2025 | Q1 2025 | Jan 2024 | Q1 2025 | Jan 2024 |
NVIDIA GeForce RTX 5070 Specifications
The GeForce RTX 5070 looks a little underwhelming on paper, packing fewer cores than the RTX 4070 Super (6,144 vs. 7,168). The memory bus is unchanged at 192-bit, as is the memory size at 12 GB. Courtesy of the upgrade to GDDR7 standard, we’re looking at memory clocks of 28 Gbps (versus 21 Gbps on the 4070 Super). This translates to a peak bandwidth of 672 GB/s, versus 504 GB/s on the former.
The GeForce RTX 5070 should also feature higher core clocks of 2.7 GHz or higher which explains why the TBP has been raised from 220W to 250W. The RTX 5070 and 5070 Ti will likely be announced in early Jan, followed by a Feb 2025 retail launch. Previously, the specifications of the RTX 5080 16 GB and RTX 5090 32 GB had also leaked out.
