TECH NEWS – Meanwhile, the “greens” unveiled the specs of the RTX 5070 Ti and RTX 5070, because we’ve been waiting for them.
DeepSeek has already lowered the value of Nvidia’s shares on the stock market, and the Bloomberg article made it worse: the Trump administration is planning to tighten the current GPU export to China even more. Nvidia’s GPUs were used by OpenAI when it unveiled ChatGPT, and over time the Biden administration has started to impose sanctions to make it harder for China to get Nvidia’s GPUs. In 2022, sales of the H100 and H200 GPUs to China were restricted, by which time Nvidia had exported a scaled-down version of both cards to the Asian country in 2023 (H800, A800), but before last year, sales of those cards were also banned.
Now the Trump administration is considering sanctions on Nvidia’s H80 GPUs. The H80 is a Chinese slaughtered version of the H100 GPU. Since the Blackwell cards are just starting to reach consumers, most MI models have been hardened to the H100 and H200 GPUs. There were rumors in Taiwan in September that Chinese companies had refused to render the H20 MI accelerator chips, and while the reason was not clear, it was already suspected that the U.S. government might have made the decision because of the Jensen Huang-led company. GPU sales could add up to $12 billion for Nvidia, even though it has 28% less power and 41% fewer cores than the H100.
The White House did not respond to Bloomberg’s request for comment, but Nvidia has already said it is willing to work with the new administration and its AI policies. The Biden administration has limited the number of countries that have free access to Nvidia’s products, and Nvidia has accused the administration of hindering the international development of AI.
Nvidia has confirmed the specifications of the RTX 5070 Ti and RTX 5070 in a PDF document. The RTX 5070 Ti is based on the GB203 GPU (like the 5080, but trimmed), has 16GB of GDDR7 VRAM, but is capable of 28Gbps instead of 30Gbps. Compared to the RTX 4070 Ti Super, it has 78% more bandwidth (896 GB/s), a 2452 MHz boost clock, 8960 CUDA cores, and 300W TGP (power consumption). In terms of performance, it will be twice as powerful as the RTX 4070 Ti and will have a suggested retail price of $750.
The RTX 5070 is based on the GB205 GPU on the PG147 reference board, has 6144 CUDA cores, 192-bit bus interface, 12GB of GDDR7 VRAM, 28Gbps memory (672GB/s bandwidth, a 33% increase over the RTX 4070/4070 Super Pair), 250W TBP (+14%), and will be available for $550 with a 2512MHz boost clock speed.
The RTX 5070 pricing is starting to resemble the rumored pricing of the AMD Radeon RX 9070…
Source: WCCFTech
Leave a Reply