The GeForce RTX 50 series GPU, based on Nvidiathe last Blackwell Architecture, is one of the most anticipated PC components to be launched or announced this year. Many of us were waiting for Jensen to mention or even give a teaser about the upcoming GeForce RTX 50 series during Computex 2024. Unfortunately, the NVIDIA CEO didn't say anything about the RTX 50 series GPUs. He didn't even mention anything about gaming. Everything revolved around AI. Fortunately, famous leaker Koppite7kimi recently tweeted about the possible specifications and configuration of the upcoming RTX 50 series GPUs.
Possible NVIDIA GeForce RTX 50 Series GPU Lineup Leaked
According to kopite7kimi, the RTX 50 series lineup includes five GPU chips: GB202, GB203, GB205, GB206, and GB207. It should be noted that there is no “GB204”, a successor to the AD104. The AD104 is the chip used by the GeForce RTX 4070 Super and RTX 4070 Ti. However, instead of a GB204, there is a GB205. NVIDIA may be trying to segment their GPUs differently this time around and hopefully they will avoid making the same mistake they made with the RTX 40 series.
The flagship GeForce RTX 5090 (GB202) gaming GPU
Currently, NVIDIA The flagship GPU is the RTX 4090, which uses an AD102 chip. Assuming NVIDIA doesn't change the naming scheme, its successor will be the RTX 5090 with a GB202 GPU chip. Assuming NVIDIA will use the same configuration as before and based on the leaked GPU specifications, the GB202 chip will have 24,576 CUDA cores. That's a 33.33% increase in CUDA core count over the full AD102, which has 18,432 cores.
At this point, it's unclear whether the RTX 5090 will get the full GB202 with 24,576 CUDA cores or perhaps a slightly scaled-down version, like the RTX 4090. The AD102 chip used in the RTX 4090 is not the full AD102 as it only has 16,384 cores.
Aside from the GPU chip, the RTX 5090 will use a faster GDDR7 VRAM with a potential memory bus width of 512 bits. It could also have a total VRAM of 32GB or 64GB. However, I'm leaning towards around 32GB of VRAM only. NVIDIA is known to be quite “conservative” when it comes to VRAM capacity with its GeForce RTX graphics cards.
The GeForce RTX 5080 (GB203) high-end gaming GPU
Next is the GB203 chip, which we may see in the RTX 5080, RTX 5080 Ti/Super or even an RTX 5070 Ti Super if NVIDIA plans to follow such naming schemes.
The GB203 will only have 10,752 CUDA cores, only a 5% increase over a full AD103 with 10,240 CUDA cores. He GeForce RTX 4080 Super Use a complete AD103 chip with 10,240 CUDA cores. Assuming GDDR7 and the new Blackwell architecture will offer a significant performance boost, the RTX 5080 may be on par or slightly faster than the RTX 4090 in raw performance.
According to the leak, the GB203 will also use GDDR7 memory with a 256-bit memory bus width and potentially 16GB or 32GB of VRAM capacity.
NVIDIA GB205 and GB206 could be mid-level GPUs
As mentioned above, if the leaks are true, NVIDIA does not have a direct successor to the AD104: the GB204. Instead, NVIDIA will release a GB205, skipping the GB204 naming or numbering scheme. There is also the GB206, which is close to the GB205 in terms of configuration and number of cores.
The GB205 Blackwell chip could be used in the RTX 5070 series and the GB206 could be assigned to the RTX 5060 series. I am using the word “series” because I expect the 5070 to have a Ti and Super variant in the future. The same can be said for the 5060 since we have a Ti variant of the RTX 4060.
Surprisingly, the GB205 will only have 6,400 CUDA cores, a 17% reduction from the AD104. Meanwhile, the GB206 will have a similar core count as the AD106, which is 4608 cores. I am worried about this configuration as we may end up with a mediocre RTX 5070 and RTX 5060 series graphics card.
On the plus side, the GB205 and GB206 will use the newer and faster GDDR7 memory with 192-bit and 128-bit memory interfaces, respectively. The GB205 could have 12GB to 24GB VRAM, while the GB206 could have only 8GB or 16GB VRAM. I suspect gamers would be frustrated and angry if NVIDIA insisted on giving their gaming GPUs only 8GB of VRAM, especially for the mid-range and mainstream segment.
8 GB of VRAM is no longer enough nowadaysespecially with modern triple-A titles that require a lot of graphics and a more significant amount of VRAM.
NVIDIA GB207 for entry-level gaming GPU
Finally, for the entry-level segment, NVIDIA has the GB207 chip. And unlike the other GPUs above, the GB207 won't have GDDR7 memory; instead, a GDDR6 memory is expected to be its companion. I don't know if it's just GDDR6 or GDDR6X. Maybe NVIDIA will use GDDR6 since it is an entry level GPU.
The GB207 may find its way into a possible RTX 5050 graphics card. Although I hope NVIDIA doesn't call it RTX 5050, as in 50-50, because that would be really funny. And I'm sure the community would be making puns if that were the case.
NVIDIA RTX 50 Series Blackwell GPU Release Date
NVIDIA CEO Jensen Huang did not mention anything about the successor to the RTX 40 series during Computex 2024. Most, if not all, of the things he discussed during his presentation were about AI. Speaking of AI, the RTX 50 series is expected to have AI features as well. Who knows what kind of AI we'll get with the next gaming GPUs?
However, there were reports earlier that the RTX 5090 and RTX 5080 would launch first. Well, it's not a surprise since NVIDIA usually releases the flagship GPUs first, and then comes to the mid-range and then the entry level.
Word on the street is that NVIDIA may announce or reveal the RTX 50 series GPUs in Q4 2024, with the graphics card's possible launch date in Q1 2025. However, there are also reports that NVIDIA could announce the RTX 50 series GPUs during CES 2025. However, we expect to learn more about the upcoming RTX 50 series graphics card later this year.
Sources: kopite7kimithrough Technological ignition
Leave feedback about this