Just Now! The Official Price of the World's Fastest GPU RTX 5090 Is Over 14,000 Yuan, and the Price of the Castrated Version of 5090 D in China Is Over 16,000 Yuan
After much anticipation, NVIDIA's RTX 50 series has finally been officially announced.
In the just concluded CES 2025 keynote speech, Huang Renxun unveiled the RTX 5090 in a new leather jacket. Without further ado, let's take a look at the performance and price:
* The fastest GeForce RTX GPU ever
* 92 billion transistors, providing 3,352 TOPS of computing power
* 2x the performance of a GeForce RTX 4090 GPU
* For desktop users, the GeForce RTX 5090 GPU with 3,352 AI TOPS and the GeForce RTX 5080 GPU with 1,801 AI TOPS will be available on January 30 for $1,999 (approximately RMB 14,651) and $999 (approximately RMB 7,322) respectively.
The configuration was exposed in advance, and the price really increased
In mid-September 2024, news that "GeForce RTX 4090 and RTX 4090D will be discontinued in October" spread like wildfire. This "lonely and invincible" performance beast has always been at the top of the high-end graphics card market. Now it is revealed that it is about to be discontinued. On the one hand, it has aroused the market's call for "stockpiling", and on the other hand, it also makes people speculate.The move is intended to pave the way for the upcoming RTX 50 series.
It has long been rumored that the RTX 50 series will be released in early 2025, and as the time node approaches, the outline of this major update is gradually becoming clearer. During Christmas 2024, a photo of the RTX 5090 PCB was suddenly exposed and quickly sparked heated discussions.Its extra-large size also makes the previously rumored 32GB of video memory more credible.
At that time, Huang Renxun's speech at CES 2025 had been announced to be scheduled for January 6th local time, so people also set it as the debut day of RTX 5090.
The day before Huang Renxun's speech, a photo of the RTX 5090 box was accidentally leaked, adding fuel to the heat of its debut. Judging from the photos released by foreign media VideoCardz,Previous rumors that the RTX 5090 will be equipped with 32GB of GDDR7 video memory were also confirmed for the first time.
In addition to performance, the price is what everyone is most concerned about. With the significant improvement in performance, the price increase seems reasonable. According to forecasts, the official launch price of RTX 4090 is US$1,599, while RTX 5090 is expected to reach US$1,999-2,499, equivalent to RMB 14,633-18,294.
Just now, the suspense was officially revealed.
The minimum price threshold for the RTX 50 series using NVIDIA Blackwell architecture is $549 (about RMB 4,023). Interestingly, Huang Renxun specifically mentioned the initial price of RTX 4090 before announcing the price.It's designed to highlight that the RTX 5070 can achieve performance comparable to the RTX 4090 for $549.
The highly anticipated top-of-the-line RTX 5090 will be available for an initial price of $1,999.NVIDIA officials also directly positioned it as the "World's Fastest GPU".
In terms of performance, the GeForce RTX 5090 is based on the NVIDIA Blackwell architecture.It has 92 billion transistors and provides more than 3,352 TOPS of computing power per second.The DLSS 4 it uses introduced multi-frame generation technology (Multi Frame Generation) for the first time, working in conjunction with the DLSS technology suite to increase performance to 8 times that of traditional rendering technology.
DLSS 4 also introduces the graphics industry's first real-time application of the Transformer model architecture. Based on the Transformer DLSS Ray Reconstruction and Super Resolution models,Using 2 times more parameters and 4 times more computation,Provides greater stability and detail rendering in gaming scenes.
At the same time, RTX 5090 also adds FP4 support, which can reduce the model size.Run generative AI models 2x faster with a smaller memory footprint than the previous generation.
*FP4 uses less than half the memory compared to FP16 (the default method used by most models)
For example, the FLUX.1 [dev] model generates an image in 15 seconds using a GeForce RTX 4090 with FP16, and in just 5 seconds using a GeForce RTX 5090 with FP4.
In addition, the TRX 50 series also features the ninth-generation NVIDIA encoder for advanced video editing and live streaming, and is equipped with NVIDIA DLSS 4 and up to 32GB of VRAM for handling large 3D projects.
Specifically, the RTX 5090 is equipped with a GB202 GPU and has 21,760 built-in CUDA cores, making it the first GeForce GPU with more than 20,000 cores. The sub-flagship RTX 5080 has 1,0752 CUDA cores, 16GB GDDR7 video memory with a 256-bit bit width, and a TDP power consumption of 360W.
The official announcement of the RTX 50 series immediately sparked heated discussions. Nvidia scientist Jim Fan highlighted the key points for everyone:New card uses neural network to generate more than 90% pixels for your game?Traditional ray tracing algorithms can only render about 10%, which is equivalent to a "rough sketch", and then the generative model fills in the rest of the details.
RTX 5090 D has appeared on the Chinese official website
There is no doubt that the new generation performance ceiling RTX 5090 is still out of the Chinese market.
I believe everyone knows that due to US export controls, NVIDIA has customized a "castrated" version of the RTX 4090 D for the Chinese market. Although the RTX 4090 D strives to retain the performance characteristics of the RTX 4090 as close as possible to those of the RTX 4090 without violating regulations, in order to provide high-performance gaming experience and AI computing capabilities,But in AI reasoning tests, its performance is still about 5% lower.But the prices of the two are basically similar.
Specifically, compared to the RTX 4090, the CUDA cores of the RTX 4090 D have been reduced from 16,384 to 14,592, and the Tensor cores have been reduced from 512 to 456, both reduced by about 11%. This also causes its performance in graphics rendering, physics simulation, and AI reasoning to be slightly inferior to the original RTX 4090.
In addition, the RTX 4090 D is equipped with the same 24GB GDDR6X video memory as the RTX 4090, with a bit width of 384 bits. In terms of total power consumption (TGP), the RTX 4090 D is 425W, slightly lower than the 450W of the RTX 4090. As for the frequency setting, the base frequency of the RTX 4090 D is increased to 2,280 MHz, higher than the 2,235 MHz of the RTX 4090, and the acceleration frequency of both is 2,520 MHz.
In general, the RTX 4090 D basically maintains a similar memory configuration and slightly improved frequency as the RTX 4090, but has reduced core count and performance.
The performance of RTX 5090 is 2 times better than that of RTX 4090. If nothing unexpected happens, NVIDIA will still customize an exclusive RTX 5090 D for the Chinese market.We also found the performance difference between RTX 5090 D and RTX 4090 D on its Chinese official website.As shown in the following figure:
In addition, the official website also provides some comparisons of which specific configurations have been "castrated", which is of general concern to everyone. As shown in the figure below, the configuration of RTX 5090 D is basically the same as that of RTX 5090.It mainly limits the processing power of the graphics card when performing AI tasks.Its AI TOPS (artificial intelligence trillion operations per second) decreased from 3,352 to 2,375.
It is worth mentioning that the official website has not yet indicated the specific price of the RTX 5090 D. Compared with the $1,999 price of the RTX 5090, whether this "castrated version" of the RTX 5090 D will remain basically unchanged like last time, we will wait and see.
