The leaks about Nvidia’s next top graphics card model are increasing. Accordingly, the jump in performance from the current GeForce RTX 4090 to the upcoming RTX 5090 could be as big as from the RTX 3090 Ti to the RTX 4090. This is not a given – there have been significantly smaller gaps in the past.
Advertisement
“panzerlied” in the Chinese Chiphell forum has already provided quite concrete information, which the Twitter/X member “Kopite7kimi” corroborates with the website Videocardz. Correct leaks have come from both in the past.
More of everything
Accordingly, the GeForce RTX 5090 should scale around 50 percent higher than the RTX 4090 – probably meaning 50 percent more shader cores. The memory transfer rate should increase by 52 percent, the cache capacity by 78 percent and the GPU clock frequency by 15 percent.
The bottom line is that this would lead to a 70 percent increase in performance. However, this is probably a best case scenario if the additional computing cores scale well.
The information refers to the comparison of GeForce RTX 5090 against GeForce RTX 4090 and not to the underlying graphics chips. Nvidia does not use a fully functional AD102 GPU in the RTX 4090. 16 of the 144 existing shader multiprocessors are disabled; that corresponds to 11 percent.
The information suggests more than 24,000 shader cores with a boost clock of 2.9 GHz in the case of the GeForce RTX 5090. The level 2 cache of the RTX 4090 is 72 MB, 78 percent more would be 128 MB. However, Nvidia could also introduce a third cache level, similar to AMD’s Infinity Cache in the Radeon RX 6000 and RX 7000 series.
50 percent more memory transfer rate would correspond to 1.5 TByte/s. According to previous rumors, the memory interface should be 512 bits wide, which means 32 GB of 24 Gbit/s RAM of the type GDDR6X or its successor GDDR7 would be conceivable. If reduced to 448 bits, 24 GB of 28 Gbit/s GDDR7 would be possible.
Codename Blackwell
Nvidia is reportedly launching the next GeForce graphics cards and server GPU accelerators under the common code name Blackwell. The GeForce GPUs will be branded as GB20x, the accelerators as GB10x. At least when it comes to servers, Nvidia should switch to a multi-chip approach. Accordingly, a GB100 could consist of two GB102s.
Those interested will have to wait even longer for the GeForce RTX 5090 and other 5000 series graphics cards: They are not scheduled to appear until 2025.
(mma)
To the home page
#GeForce #RTX #Reportedly #percent #faster #RTX