The NVIDIA RTX 40 would have twice the performance of the RTX 30

NVIDIA’s next-generation Ada Lovelace GPUs will deliver at least twice the performance of Ampere GPUs. According to @kopite7kimi, the RTX 4090 will have no difficulty outperforming the RTX 3090., even with normal speeds and a partially disabled AD102 Chip. This means that the top of the range next-gen GeForce graphics cards will be about 2x faster than the RTX 3090 Ti.

The RTX 4090 Ti, if it ever arrives, will take advantage of the full AD102 CHIP with higher speeds.

The AD102 will feature 12 GPCs, a 70% and 50% increase over the RTX 3090/3090 Ti (GA102) and GA100/H100, respectively. It is rumored that FP32+INT32 or the core count per SM will grow by 50% over Ampere, resulting in a 33% increase in thread count. ROP count by GPC is also scheduled to double, while the L1 cache will be expanded to 192 KB (+50%). Finally, the L2 cache is set to get an upgrade, going from just 6MB on the RTX 3090/3090 Ti to 96MB on the RTX 4090/4090 Ti.

I found that some friends didn’t understand what I said.
I mean a low-frequency AD102 without a full-fat config.

– kopite7kimi (@ kopite7kimi) May 13, 2022

Just looking at the total increase of FP32+INT32 and cache, you can see how NVIDIA will be able to achieve twice the performance increase of Ampere. With a more polished compute node ( TSMC N4 ), microarchitectural improvements, and higher speeds, the RTX 4090 could offer the biggest generational leap in performance we’ve seen in quite some time.