NVIDIA builds an exotic RTX 4070 from a larger AD103 by disabling almost half of its shaders. In fact, it seems that some batches of GeForce RTX 4070 graphics cards are based on 5 nm ” AD103 ” silicon. A chip that is considerably larger than the ” AD104 ” that powers the original RTX 4070.
Where does this “discovery” come from?
The discovery was made by an owner of an RTX 4070 MSI Ventus 3X E 1 Go OC. He reported that the GPU-Z software was unable to correctly detect the characteristics of his card and became a little panicky. When the card was observed more closely by TechPowerUP (GPUZ’s editor), they found that the card was based on “AD103” silicon. What’s astonishing is that to use AD103 to make an RTX 4070, NVIDIA has to activate just 57% of the available shaders and only 36 MB of the 64 MB of L2 cache available on the die. But the castration goes further, since the memory bus must also be reduced to 192 bits instead of the 256 bits available, to handle the 12 GB of memory. On the other hand, the packaging size of the AD103 and AD104 is similar, making integration seamless for Nvidia’s card manufacturing partners. In real life, it’s unlikely that a typical user will be able to detect any difference between these exotic RTX 4070s and the classic models. On the other hand, it’s hard to understand what motivates Nvidia to use these much more complete chips and then degrade them to equip RTX 4070s. Several hypotheses can be put forward:
- Either Nvidia is having trouble selling its AD103s, which are “normally” intended for RTX 4080s.
- Or the AD103s used are models that have been rejected in quality tests for the 4080/90 and reused to produce 4070s.
The second hypothesis seems very credible: a few years ago, RTX 2060s were made from chips initially intended for RTX 2070 Super…
