According to the latest news, NVIDIA’s next-generation consumer-grade flagship graphics card, the GeForce RTX 4090, may be configured with 16,128 CUDA cores and twice the performance of the RTX 3090, and is expected to be released in mid-July.
Although NVIDIA only launched the updated flagship GPU, the GeForce RTX 3090 Ti at the end of March 2022, according to past cycle practices, it should launch a new architecture consumer GPU this year. So after the GTC 2022 enterprise-level Hopper architecture was unveiled, rumors about the consumer-level Ada Lovelace architecture were rampant.
OK, let’s do a new summary.
RTX 4090, AD102-300, 16128FP32, 21Gbps 24G GDDR6X, 450W, ~2×3090.
I am disappointed with RDNA3.
That’s all.– kopite7kimi (@ kopite7kimi) May 16, 2022
The latest and relatively credible rumor comes from a well-known whistleblower kopite7kimihe pointed out that the RTX 4090 will use the GPU chip model AD102-300, which contains 16128 CUDA cores specializing in single-precision floating-point operations (FP32), which is equivalent to 126 sets of Streaming Multiprocessors (SM), with a rate of 21 Gbps 24GB GDDR6X Memory, the overall performance should be up to 2 times that of the RTX 3090.
Let me add that the rumored complete AD102 chip has 144 SMs and consumes up to 600W. It is now estimated that it may be applied to the subsequent higher-end RTX 4090 Ti. After the RTX 4090 has been adjusted in the number of SMs, the power consumption of the entire card reaches 450W, which is equivalent to the RTX 3090 Ti launched earlier this year.
Mid July.
– kopite7kimi (@ kopite7kimi) May 16, 2022
As for the launch time of new products,kopite7kimi Also a clear reply falls in mid-July.
Current estimates GeForce RTX 40-series GPUs The first wave of styles will include RTX 4090 / RTX 4080 / RTX 4070, but remember that the specs rumors are unconfirmed and change very fast, we will try to update the information with high confidence.
NVIDIA GeForce RTX 40 Series GPU Rumored Specs
| product | GeForce RTX 4090 | GeForce RTX 4080 | GeForce RTX 4070 |
| Architecture | Ada | Ada | Ada |
| Process | TSMC N4 (4nm) | TSMC N4 (4nm) | TSMC N4 (4nm) |
| Chip model | AD102-300 | AD103 | AD104-400 |
| PCB model | PG137/139-SKU330 | to be confirmed | PG141-SKU341 |
| Streaming Multiprocessors (SM) | 126 | to be confirmed | to be confirmed |
| CUDA Cores | 16128 | to be confirmed | to be confirmed |
| memory form | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6 |
| memory speed | 21 Gbps | to be confirmed | 18 Gbps |
| TDP | 450W | to be confirmed | 300W |
| launch/market time | July 2022 | July 2022 | July 2022 |
