This past week has been pretty dull on the life front and, at the same time, very active on the digital front. Since WHO has called Coronavirus a Pandemic, most of the people in the affected 125 countries are trapped inside the houses. Many significant events scheduled in March/April are canceled. These include the iPhone 9 and iPad reveal event by Apple, Games developers conference, and finally, the E3. Google decided to make its conference a digital event earlier, and now Xbox has decided that it will move towards the digital-only event as E3 is getting delayed.
On the other hand, rumor-mil is going strong. Xbox series X specifications leaked a few days back, PS5 to reportedly have 11 TFLOPS of FP32 compute power.
E3 has always been an important moment for Team Xbox. Given this decision, this year we'll celebrate the next generation of gaming with the @Xbox community and all who love to play via an Xbox digital event. Details on timing and more in the coming weeks https://t.co/xckMKBPf9h
— Phil Spencer (@XboxP3) March 11, 2020
Nvidia had to cancel its developer’s conference amidst COVID-19. It was reported that Nvidia was going to announce the critical specifics of its upcoming architecture for both its consumer and HPC markets. A few days back, we reported that specifications and early benchmarks of three mysterious GPUs from Nvidia were leaked online. Looking at the leaked specs, one could say that these graphics cards were aimed at the HPC market. Although, there is a teeny-tiny possibility that one of these GPUs could be the TITAN GPU aimed at the pro-consumer market.
Now, another leak has surfaced explicitly detailing that the leaked specifications are of a GeForce Graphics card. It should be noted that the HPC GPUs leaked earlier, and the GeForce GPUs may feature different architectures. We have seen that both Ampere and Hopper names have been associated with upcoming architecture from Nvidia. We are still unclear regarding the assertion that both of these architectures will power GPUs from different segments of the market.
The leak from two different Twitter accounts suggests that the flagship GeForce GPU allegedly based on Ampere architecture called GA 102 will be manufactured by Samsung. Debunking earlier rumors which pointed out towards 7nm EUV architecture from Samsung, the leak suggests that the GPU will be manufactured by 10nm node from Samsung. The upcoming mobile Tegra GPUs will also be fabricated under the same node. The new 10nm architecture from Samsung is said to be closer to the 8LPP technology, but it is not EUV based, making sure that it is not the 7nm architecture.
We know that Nvidia has decided to divide the fabrication of its GPU between TSMC’s foundry and Samsung’s foundry. Wccftech reports that Nvidia has already announced that its life-long foundry partner TSMC will handle most of the 7nm orders while only a handful of the 7nm orders will be given to Samsung. The leak reinforces that since the leaked GPUs are said to be manufactured under the 10nm architecture. If TSMC were to get the bulk of 7nm orders, it means the HPC parts leaked earlier will be manufactured by TSMC’s 7nm process. The fact that most of the orders will be 7nm hints towards more and more GPU demand in the AI and data science (HPC) market.
Now that we know the architecture (Ampere?) to be used in the GeForce GPUs is going to be manufactured under Samsung’s 10nm node, let us talk about the specifications of the GPUs. According to KittyCorgi on Twitter, the whole GeForce lineup will be powered by different configurations of 5 Ampere GPUs. The flagship GPU, called GA 102, will be the successor of TU 102 found in TITAN RTX and GeForce RTX 2080Ti. Other GPUs include GA 103, GA 104, GA 106, and GA 107.
Out of these only GA 103, GPU has no predecessor, which could mean that the xx80 SKUs may have their own GPU rather than the cut-down version of GPUs used in the xx80ti or TITAN graphics cards. SLI support through NVLINK will only be available to connect two flagship GA 102 GPUs. No other GPU will feature SLI capabilities, which is the right decision since most games do not have support for SLI.
1 Surprisingly, the next gen chip will be based on Samsung 10nm node 2 Not very solid about what GA102 really got, may struggle to be 40% above 2080ti, but bound to be under that number if compared against full bore TU102 3 SLi will be available to GA102 ONLY 4 RTX for everyone pic.twitter.com/b5YAKmOmgf
— Yuko Yoshida (@KittyYYuko) March 11, 2020
Another next-gen feature would be Ray Tracing for every GPU, even the low-end versions too. Ray Tracing or RTX support was the controversial aspect of the Turing release. Though Ray Tracing, termed as the “Holy Grail” of graphics by Nvidia, shaped the gaming industry, the implementation was not great. Since Nvidia wants Real-time Ray Tracing in all of the Ampere GPUs, it means there will be structural enhancements in the architecture and improvements in the design of RT cores so that even the low powered GPUs can handle such a “divine” task.
On the rasterization front, the leak suggests that the FP32 units will be doubled in the Ampere graphics cards, but doubled units do not mean doubled rasterization performance. The biggest critique on the Turing GPUs other than Ray Tracing was a small generational leap over the Pascal graphics cards. Ampere GPUs would try to rectify that by increasing the rasterization performance.
Lastly, the GPU will feature the new PCIe 4.0 interface, which is the new industry standard. Ampere architecture will further optimize the Tensor cores, too, so that the latest graphics cards can utilize the DLSS technology to the full.