The Chiphell leaker who initially disclosed information about NVIDIA’s upcoming consumer GPU lineup has now released additional rumors regarding the capabilities of the new architecture.
Just to recap, Panzerlied was the individual who first unveiled NVIDIA’s intention to omit the XXX04-class GPU in their upcoming gaming product series. Following this revelation, we contacted Kopite7kimi, a highly regarded NVIDIA insider, who verified these rumors. Additionally, Kopite7kimi informed us that the forthcoming NVIDIA RTX 50 series would embrace the GB2XX naming convention.
Today, Kopite7kimi has unveiled fresh information regarding the Blackwell series, which is the internal codename for NVIDIA’s upcoming generation of GPUs. It is now expected that Blackwell will encompass both data-center and gaming series. However, distinct naming conventions will be used for these two GPU series, with GB1XX assigned to high-performance computing (HPC) and GB2XX designated for gaming GPUs.
Meanwhile, Panzerlied has offered insights into what graphics enthusiasts can anticipate from NVIDIA’s upcoming generation. Rather than disclosing exact numerical values, Panzerlied is providing percentage improvements across different aspects of the Blackwell family.
When comparing NVIDIA RTX 5090 to RTX 4090:
– There’s a 50% increase in size, likely referring to the number of cores.
– The memory bandwidth sees a 52% boost.
– Cache, presumably the L2 cache, experiences a substantial 78% increase.
– Frequency, possibly related to GPU boost, shows a 15% uptick.
– Overall performance appears to improve by 1.7x.
In a later clarification within the discussion, Panzerlied specified that these claims pertain to the RTX 4090 specifications, not the AD102. If we consider that the RTX 4090, which currently features 21 Gbps memory, might undergo an upgrade to 32 Gbps (a 52.4% increase), this could imply the potential use of GDDR7 technology in the RTX 50 series. It’s worth noting that the successor to the AD102 is also rumored to incorporate a 512-bit memory bus, although it may not necessarily be employed in the RTX 5090 specifically; it might be reserved for a TITAN/RTX workstation variant or a future 5090Ti model.
It appears unlikely that NVIDIA would immediately adopt the fastest GDDR7 memory for their card, so alternative RTX 5090 configurations like 512-bit/24 Gbps or 448-bit/28 Gbps could also be under consideration.
If we assume that the other claims are also based on the RTX 4090 as a reference point, a 15% increase in frequency would result in a boost clock of 2.9 GHz, with real-world workloads likely achieving clocks of 3.0 GHz or even higher. Moreover, a 78% increase in cache suggests that the GB202 GPU would likely feature 128MB of L2 cache.
Source & Image: Chiphell