YouTube Blogger Reuters 3080Ti Parameters, Huang Xiaoming Took out DGX A100 From the Oven to Preheat the Press Conference
Many sources indicate that Huang Renxun will release a new generation of Ampere GPUs at the GTC keynote on May 14, and he even made a funny video for this purpose. But recently, a YouTuber released comprehensive information about the new RTX 3080 Ti card.
Huang Xiaoming is doing something weird again. This time he put his new product into the oven.
On May 12, NVIDIA released a video on Youtube in which Huang Renxun took a DGX A100 out of the oven and said, "It has been baked for a while."
There is no doubt that this eye-catching operation is a warm-up for the upcoming Ampere GPU.
It is said that this DGX A100 system equipped with A100 GPU will be demonstrated at the GTC 2020 conference, and Tesla A100 will be the first Ampere architecture product.
Huang Xiaoming's keynote speech at NVIDIA GTC 2020 will officially begin at 9 pm Beijing time on May 14. The new graphics card series that everyone has been looking forward to is finally here.
YouTube blogger: Revealed detailed parameters of new products
But in fact, there have been a lot of leaks about the information about the new graphics card RTX 3000 series on the Internet. After all, this graphics card, which has been delayed many times, has consumed the patience of too many users.
On May 12, which was yesterday, Moore's Law is Dead, a well-known technology blogger on YouTube, revealed the main specifications, performance and comparison of RTX 3080 Ti.
According to his information, the entire series of graphics cards use RTX (graphics card ray tracing technology), HPC computing cards may be released in May, while gaming cards will have to wait until September.
The RTX 3080 Ti, based on the new Ampere architecture GA102 core, has undergone major updates from manufacturing process to core architecture to technical parameters.
The RTX 3080 Ti has 5376 built-in CUDAs, uses a 7nm process, has a core frequency of 2.2 GHz, a single-precision floating-point number of 21 TLFOPs, a video memory matched with 18Gbps GDDR6, a bit width of 384 bits, a total bandwidth of 863 GB/s, and power consumption maintained at 220~230 watts.
In addition to these powerful performance parameters, some information has been leaked before, such as 4 times the ray tracing performance and support for DLSS3.0.
In terms of performance, the Ampere GPU has an IPC performance improvement of 10-20%, a doubled L2 cache, and a higher operating frequency, reaching a baseline level of 1900MHz.
The appearance of RTX 3080 Ti is basically the same as RTX 2080 Ti, but it will change from dual-burner to triple-burner - with three fans.
It is said that this information comes from an NV insider. If the information is true, this time Huang Xiaoming has made a big move.
Up to 70% performance improvement compared to the previous generation
Compared to the previous generation RTX 2080 Ti, the 3080 Ti really surpasses it by a lot.
The 3080 Ti Boost frequency increased by 42%, the single-precision floating point increased by 50%, and the bandwidth increased by 40%.
In actual tests of multiple 4K games, the frame rate of 3080 Ti is 40% higher than that of RTX 2080 Ti, and can reach 50% after optimization. N cards specially optimized for games can achieve a performance improvement of 70%.
Let's make a detailed parameter comparison:
_ | RTX 2080 Ti | RTX 3080 Ti |
| Architecture | Turing | Ampere |
| Core Model | TU 102 | GA 102 |
| Process Technology | 12 nm | 7 nm |
| Single floating point performance | 13.4 TFLOPs | 21 TFLOPs |
| Stream Processor | 4352 | 5376 |
| Video memory type | GDDR6 | GDDR6 |
| Video memory capacity | 11 GB | 18 GB |
| Video memory width | 352Bit | 384bit |
| Core frequency | 1350-1545 MHz | 2200 MHz |
| Video memory frequency | 14000 MHz | 18000 MHz |
| Memory bandwidth | 616 GB/s | 864 GB/s |
| TDP Power Consumption | 250 W | 220-230 W |
| External power supply | 8 + 8 pin | 8 + 8 pin |
From these data, we can see that the RTX 3080 Ti has completely suppressed the RTX 2080 Ti. In this light, its strong performance is not surprising.
As for the authenticity of this leaked information, we can only wait until the NVIDIA GTC Keynote tomorrow night to find out. But considering the warm-up video that Huang Xiaoming has already released, he probably has some big moves to reveal this time.
Huang Xiaoming will be online at 9:00 pm Beijing time on May 14th. If you are interested, you can check the details on NVIDIA's official website:
https://www.nvidia.cn/gtc/keynote/
-- over--