NVIDIA inside GPU codenames usually go one thing like this: first a G (for “graphics” or “GeForce”), then the primary letter of the structure title, after which three digits. The first digit signifies the architectural revision, the second digit marks the chip revision, after which the final digit suggests the particular chip’s anticipated market place, with increased digits utilized to lower-end GPUs.
So, going by historic codenames, we might usually count on to see GB102, GB104, GB106, and so forth. However, in response to the newest leak out of China in addition to new data from common NVIDIA leaker kopite7kimi, we cannot be seeing any of these. Instead, NVIDIA’s next-generation graphics structure goes to begin with GB202 after which proceed by way of GB203, GB205, GB206, and eventually GB207.
There are a lot of oddities right here. First, and smallest amongst them, is the continued presence of an “x03” GPU. First launched within the Ampere sequence a lot later than the remainder of the GPUs, “x03” signifies a graphics processor between the “x02” vary—that is the “Titan” or “RTX 4090” tier—and the standard “x04” class, which was usually the high-end GPU upon which the “x80” card could be primarily based. As an instance, the GTX 1080 was primarily based on GP104, and the RTX 2080 was primarily based on TU104.
With Ampere, NVIDIA moved the “104” bracket down a tier, so the GA104 GPU powered the RTX 3070 and its derivatives, whereas the RTX 3080 was primarily based on a considerably cut-down GA102. Most not too long ago, with Ada Lovelace, NVIDIA wedged the AD103 GPU in between the 102 and 104 chips, in order that the RTX 4090, RTX 4080, and RTX 4070 are all primarily based on totally different silicon—a primary for GeForce playing cards.
Automatically translated from Chinese by Google.
According to recognized NVIDIA insider Panzerlied, NVIDIA’s next-generation elements will as a substitute skip the “x04” tier altogether in favor of a heretofore-unseen “x05” tier. The attainable implications of this are fairly fascinating, however with solely this factoid to go on, it is manner too early to make any judgements about what it really means. Our cash’s on NVIDIA focusing its efforts on making a smaller, inexpensive GPU to serve the midrange.
The different curiosity on this data is that Blackwell will apparently ship in its second-revision kind. It will not be the primary time; the GeForce GTX 900 sequence primarily comprised GM20x GPUs, being the second-generation Maxwell structure. What’s uncommon is that we have not heard something about any plans to ship a first-generation Blackwell structure in any market section.
Over at Videocardz, WhyCry took the leak from Chiphell to kopite7kimi to substantiate its veracity. That’s the place the “GB20x” nomenclature got here from, apparently. That fellow’s leaks have been fallacious earlier than, however they’ve additionally been very correct a number of occasions up to now, so he clearly has an angle on insider data at NVIDIA. We in all probability will not hear something immediately from the inexperienced group about Blackwell throughout the subsequent yr, so do not maintain your breath.