We have reached the most rumored time of the year, as “Copite7kimi“One of the most reliable leakers of details and information about NVIDIA hardware and devices, shared some new details about the type of graphics processing unit coming from the green team, although there are some changes or contradictions in the narrative from his previous leak.
Kopite7kimi initially claimed that the flagship gaming GPU of the upcoming series would feature a 512-bit memory bus, and noted that these details had been confirmed by several sources according to him. However, after a few months, he changed his words and said that the memory bus would not It is 512 bits, but only 384 bits.
It was also hinted last week that the RTX 50 series GPUs as well as the RTX 40 series may share a very similar memory bus capability, suggesting there are no significant changes to the flagship GPU known as the GB202.
But Kopite7kimi has now backtracked again and put the 512-bit option back in the interface without going into more details, and it is unclear whether NVIDIA is currently exploring different memory buses, or if the leaker is just making educated guesses, but given its track record Previously, one might be inclined to believe the first option.
More importantly, Kopite now claims that the new series of GPUs will stick to 28 Gbps GDDR7 memory, exceeding the GDDR6X memory specification by a large margin, and will still have NVIDIA, which will allow the company to make major future upgrades, as this type of memory is announced. Memory typically has speeds of 32+ Gbps.
The first batches of RTX 50 units may not have 3GB of GDDR7 memory, but we know that they are in development now and should be launched next year, and NVIDIA may launch mid-range and budget GPUs as well.
Finally, the leaker also mentioned that the GB203 GPU may be half as powerful as the GB202, while neither confirming nor denying that the GB202 will use a multi-chip package. The current flagship AD102 model features 80% more cores than the AD103, and this has led to a push for GPUs. The RTX 4090(D) takes graphics to a completely different performance level, so if NVIDIA takes a similar approach with the Blackwell architecture, we could see a big performance gap again between the RTX 5090 and 5080 series, especially with the rumors circulating about a 512-bit bus.
At worst, Nvidia will stick to using the same memory bus width in its next-generation GPUs, but this will still result in a significant increase in bandwidth thanks to faster GDDR7 memory. However, if the company actually decides to use a 512-bit memory bus , this will mean that the graphics processor will allow more memory capacity and provide approximately 80% higher bandwidth compared to the AD102 processor.
What is GB203 GPU?
The GB203 is an upcoming graphics processor (GPU) from Nvidia based on the new Ada Lovelace chip, which is rumored to be part of the RTX 50 series, and is believed to be the successor to the current RTX 4090 Ti graphics processor, and so far there is no official information about its release date.
