NVIDIA has changed the specifications of its Lovelace GPUs various times. Yesterday, we got lower total board power figures for the RTX 4070 and the RTX 4080.
As a recap, last month TSE (Time Spy Extreme) benchmarks for the RTX 4070 and the RTX 4080 were leaked. The former had synthetic performance on par with the RTX 3080. That was already impressive, however, a week ago the RTX 4070 was revised with newer and better specifications.
RTX 4070 New vs Old
- 7680 vs 7168 Cuda cores (7% increase)
- 12GB VRAM vs 10GB VRAM (20% increase)
- 192 bit-bus vs 160 bit-bus (20% increase)
- 21Gbps vs 18Gbps (16% increase)
- 504GBps vs 360GBps (40% increase)
This newer 4070 was comparable to the leaked 4070 Ti (specification wise). Many believe the RTX 4070 Ti will launch next year with GDDR7 memory, although that’s just a speculation. This 4070 was benchmarked and challenged the RTX 3090 Ti from Ampere.
The new 4070 and the original 4080 had their total board powers changed yesterday. Kopite stated 320W and 285W as the total board power for the 4080 and 4070 respectively. This drastically tipped the balance in favor of Lovelace as shown in the charts below. The efficiency crown may just go to NVIDIA this time (despite higher power consumption).
Kopite today released the base, boost and max clocks for this RTX 4070. The increments are massive!
— kopite7kimi (@kopite7kimi) August 10, 2022
This is really impressive because now custom GPUs from Lovelace will easily break the 3GHz barrier. In comparison to the RTX 3070, the its next-gen successor features a 700-900MHz increase in clocks. Expect TFLOP compute power to go past 40TFLOPS which is 2 times more than the 3070 or similar to the RTX 3090 Ti.