NVIDIA considered different configurations for the GeForce RTX 4070. It is logical, after all, the RTX X070 models are always among the most important of their generation, since they are located in what we can consider the entry into the high range. We all know that he finally settled on a configuration with 5,888 shaders and 12 GB of GDDR6X memory at 21 Gbps with a 192-bit bus.

However, the green giant originally considered a very different model that would have a much more powerful GPU, but this would be penalized by both bandwidth and amount of graphics memory. In terms of shader configuration, it was supposed to be like the GeForce RTX 4070 SUPER, but its memory and bus configuration was clearly lower, because it was supposed to have 160-bit bus and only with 10GB of memory GDDR6X with a speed of 21 Gbps.

In the end, NVIDIA opted for a configuration with 12 GB of GDDR6X memory, probably because it allowed it to offer a more balanced performance at the target resolution of this graphics card, 1440p, and because by reducing the number of shaders, it could leave enough room for a later release of the TI version without stepping on it too much with the standard model. The game was successful, no doubt about it.

Specifications of the original GeForce RTX 4070 and the final version

The original model was supposed to have:

• 7168 shaders at 1920MHz-2520MHz.
• 224 texturing units.
• 80 raster units.
• 224 tensor kernels of the fourth generation.
• 56 cores of the third generation RT.
• 160-bit bus.
• 10 GB of GDDR6X memory at 21 Gbps. Bandwidth 420.2 GB/s.
• L2 cache: 48 MB.

With these specs, its performance would be lower than the GeForce RTX 4070 SUPER due to less bandwidth and also less graphics memory, which would undoubtedly make an important difference in certain games by enabling ray tracing and generation. . However, in scenarios where the 10 GB was sufficient, the performance of this model would be quite enough close to GeForce RTX 4070 SUPER.

The version officially launched by NVIDIA has:

• 5,888 shaders at 1,920 MHz-2,475 MHz, normal and turbo mode.
• 184 texturing units.
• 184 tensor kernels of the fourth generation.
• 64 raster units.
• 46 cores of the third generation RT.
• 192-bit bus.
• 12 GB of GDDR6X memory at 21 GHz Bandwidth: 504.2 GB/s.
• L2 cache: 36 MB.

In raw performance, the model that officially hit the market was clearly below the previous configuration, but is better balanced in bandwidth and total amount of graphics memory. In scenarios where 10GB of graphics memory doesn’t limit the original version, it would 5 to 10% more powerful than the model that reached the market.

Personally, I think NVIDIA should have reserved this 160-bit configuration and 10 GB of GDDR6 memory for GeForce RTX 4060 Tia model that, if it had come with that extra 2 GB and a wider memory bus, would have positioned itself much better on the market.