Famous and avid leaker Kopite details Nvidia’s RTX 5070 seriesand the Ada Lovelace’s overall spec bump is a mixed bag, at least on paper. The leaker has a good record and has previously leaked RTX 5090 and RTX 5080 specs. The RTX 5070 series includes the RTX 5070 Ti and the base RTX 5070. rumor There is news that Nvidia will launch the RTX 5090, RTX 5080 and RTX 5070 series for the first time on Blackwell International Consumer Electronics Show next month.
According to data leaked by Kopite, the RTX 5070 Ti is equipped with a GB203-300-A1 chip, which is similar to the RTX 5080 and has 8960 CUDA cores or 70 SMs; 16% more than the RTX 4070 Ti. Through the 256-bit interface, the RTX 5070 Ti is equipped with 16GB GDDR7 VRAM and is rumored to operate at 28Gbps The total bandwidth is 896 GB/s. This puts the RTX 5070 Ti very close to the RTX 5080, despite a 20% difference in core count.
For context, there’s a clear spec gap between the RTX 4080 and 4070 Ti, prompting Nvidia to price them nearly $400 apart. It’s reasonable to expect that Blackwell’s delta won’t be that large, but the ball is in Nvidia’s court. The RTX 5070 Ti is expected to deliver 300W of power, 15W more than its predecessor.
Next, the RTX 5070 is said to be powered by a GB205-300-A1 chip. This is an improvement compared to the RTX 4070, which uses an AD104-250, XX104-class GPU, and is higher than the XX105/205-class GPU. The RTX 5070 uses a smaller chip, resulting in a significant reduction in the number of cores to 6144 CUDA cores, but it is still 4% more than the RTX 4070. bandwidth. The TDP is slightly higher at 250W, which is 25% higher than the RTX 4070.
| GPU name | RTX 5070 Titanium | RTX 4070 Titanium | RTX 5070 | RTX 4070 |
|---|---|---|---|---|
| die | GB203-300-A1 | AD104-400-A1 | GB205-300-A1 | AD104-250-A1 |
| CUDA color | 8960 | 7680 | 6144 | 5888 |
| bus width | 256 bit | 192 bits | 192 bits | 192 bits |
| memory | 16 GB | 12GB | 12GB | 12GB |
| TDP | 300W | 285W | 250W | 200W |
The RTX 5080’s significantly reduced specs compared to the RTX 5090 set a bland tone for the rest of the Blackwell lineup. While the RTX 4090 has 68% more cores than the RTX 4080, that gap has increased to 102% from generation to generation, at least according to unconfirmed leaks.
Samsung is expected Mass production of GDDR7 24Gb (3GB) memory will begin early next year. A potential RTX 50 Super update could feature these newer mods to increase VRAM capacity by 50%, but that’s just speculation. In theory, Nvidia could launch a 48GB RTX 5090 SUPER, but since both Intel and AMD have exited the high-end market, it doesn’t have to. Users who need higher VRAM capacity (perhaps for artificial intelligence) will have to pay a high price and choose Nvidia’s Blackwell data center accelerator or future Blackwell workstation GPUs.