Informant panzerlied shared data about what users and enthusiasts can expect from the next generation NVIDIA RTX 50 (Blackwell) graphics cards. Instead of specific characteristics of the chip or video cards, the insider spoke about improvements in percentages.
According to new information, the following changes are expected in NVIDIA RTX 5090 video cards compared to the RTX 4090:
50% increase in scale (probably cores) 52% increase in memory bandwidth. 78% increase in cache (probably L2 cache) 15% increase in frequency (probably GPU acceleration) 1.7x improvement (probably performance)
Panzerlied later clarified that the data is based on the specifications of the RTX 4090, and not the AD102 chip. Considering that the RTX 4090 has 21 Gbps of memory, and the new product will receive approximately 32 Gbps (an increase of 52.4%), this could mean that the RTX 50 will receive GDDR7 technology. According to rumors, the successor to the AD102 will also receive a 512-bit memory bus, although it may not necessarily be used in the RTX 5090.
It seems unlikely that NVIDIA could use the fastest GDDR7 memory on its card right away, so RTX 5090 configurations such as 512-bit/24Gbps or 448-bit/28Gbps could also be considered.
Assuming that the other claims are also based on the RTX 4090 as a benchmark, a 15% increase in frequency will push the clock speed up to 2.9 GHz, and an actual workload will likely result in a clock speed of 3.0 GHz or higher. Additionally, the 78% cache increase suggests that the GB202 GPU will get 128MB of L2 cache.
There is more than a year left before the possible launch of NVIDIA RTX 5090 video cards, but according to currently known data they would look something like this:
NVIDIA introduced GEFORCE RTX 4090 video cards for $1599 and RTX 4080 for $899 – on sale from October 12