As the chip giant $NVIDIA (NVDA.US)$ confirms that CEO Jensen Huang will deliver a keynote speech at 10:30 AM Beijing time on January 7 next year, the "new graphics card" — the RTX 50 series graphics card based on the Blackwell architecture — is finally approaching its launch moment, much to the anticipation of global players.

Like most Consumer Electronics, along with the revealed configuration information of the 5060 graphics card this week, information concerning the RTX 50 series is also gradually being completed.

According to market consensus, Jensen Huang will announce the RTX 5090 and 5080 during the keynote speech, and may mention the 5070 Ti and 5070; however, these 70 series graphics cards may not be launched until February. Finally, targeting the "dessert" level market (around 2000-3000 yuan) will be the 5060 and 5060 Ti graphics cards, which are expected to be launched as early as March next year.

What news do the new graphics cards have?

As a "king of cards" level product, the RTX 5090 graphics card will be equipped with a record-breaking 32GB of GDDR7 memory, compared to the 4090 graphics card which uses 24 GB of GDDR6X memory.

According to a screenshot from fast media, the graphics card manufacturer Zotac's official website briefly "leaked in advance" the existence of several new NVIDIA graphics cards this week, and also added two configuration details to the filtering section that existing products had not reached.

Among the four graphics cards launched, besides the 5090 having an increase in video memory, the video memory of the 5070 Ti (16GB GDDR7) will also increase by 4GB compared to the 4070 Ti, while the other new graphics cards remain unchanged in video memory. This makes the 5070 Ti a new graphics card with both "benchmarking" and "volume" potential. The 5070 Ti will feature a 256-bit memory bus, achieving an impressive total memory bandwidth of 896 GB/s.

In terms of computational performance, since the previously announced datacenter chip B200 uses the Taiwan Semiconductor 4NP process, it is inferred that NVIDIA will also equip the Blackwell gaming graphics card with the same process, rather than the initially rumored Taiwan Semiconductor 3nm.

Reportedly, the GB202 used in the RTX 5090 will be the largest home gaming chip in recent years, with an area of 744 square millimeters, close to the size limit of the photomask (about 850 square millimeters), marking a 22% increase compared to the AD102 of the 4090. This also suggests that the newly released 5090 graphics card is likely to come at a higher price.

According to NVIDIA's usual practice, the GB202 used in the RTX 5090 is not the "full-version". Under the complete framework, the GB202 has 192 SM (streaming multiprocessors). Current intelligence shows that the GB202 for the 5090 will activate 170 SMs. The number of CUDA cores in the GPU will significantly increase by 33%, reaching 21,760.

Meanwhile, both the RTX 5080 and 5070 Ti will use the GB 203 chip, with the difference being that the latter's chip (GB 203-300) will have certain limitations. The CUDA counts for the two chips are 10,752 (an 11% increase over the previous generation) and 8,960 (+16%), respectively.

The RTX 5070 will use the GB 205 core, with the CUDA count dropping to 6,400, while the video memory and memory bus width will also further drop to 12GB GDDR7 and 192 bits.

Somewhat disappointingly, although the new generation of cards receives an upgrade in memory, this week's leaks reveal that NVIDIA still scheduled the RTX 5060 with 8GB of video memory. At a time when Apple's supply chain master Cook has completely phased out the 8GB configuration, Jensen Huang still intends to launch an 8GB graphics card in 2025.

Considering that Intel has equipped the "200 dollar graphics card" Arc B570 with 10GB of memory, whether "8GB of memory" is sufficient indeed requires a big question mark.

There are also market speculations that once the production capacity of GDDR7 memory increases, NVIDIA may upgrade to a higher memory version of the 5060 graphics card.

Be careful with the electricity bill.

The enhancement of Blackwell chips and memory has also brought about an increase in the NVIDIA RTX 50 series graphics cards that consumers might not be too pleased to see—dimensions and electrical consumption have also increased.

According to various leaks, the maximum theoretical load power consumption of the RTX 5090 can reach 600W, which is 150W higher than that of the 4090. The power consumption of the 5080, 5070 Ti, and 5070 graphics cards has also increased, but overall power consumption remains relatively controllable.

Another issue is that, given that the 4090 graphics card is already quite large, it is said that the 5090 will be even larger, potentially occupying "three to four slots" on the Main Board, while the 4090 occupies three slots on most Main Boards.

Editor/rice