The Lovelace architecture, named after the mathematician who is considered one of the founders of the creation of computers, will be the one to replace the Ampere With another possible upgrade in the production process
Launch of the Ampere generation of NVIDIA Last year did not work as planned, at least in the eyes of consumers - the hardware did show us an impressive leap thanks to the transition to a production process of סמסונג At 8 nm compared to an older 12 nm production process (from TSMC) in the generationTuring, But all this did not really help many of us when the low availability led to inflated prices and many disappointments on the part of those who planned to integrate the new age video cards into their systems and did not get a real opportunity to do so.
In the current situation, moving to an even more advanced manufacturing process where the chip output is probably even more modest sounds like madness, right? Well, by some leaks Twitter Popular these are the plans of Jen-san Huang and his people anyway.
According to preliminary information received winks and approvals from several channels, the next generation architecture of NVIDIA Named Lovelace, the name of a 19th-century British mathematician best remembered for her contribution to the development of the most advanced mechanical computer of its time - it will contain up to 18,432 active CUDA units in its largest AD102 configuration, based on modern 5-nm lithography Only a limited number of processing chips are leading (and very small, compared to a graphics core where a video card is) to the mobile world.
This is an increase of more than 70 percent to the amount of CUDA cores active in products based on the GA102 core today, so naturally one can expect another very big jump in possible performance - online speculation experts already offer calculations that describe maximum performance of up to 66TFLOPS on paper , Compared to the theoretical 38TFLOPS in the Ampere generation and 16 theoretical TFLOPS in theTuring. If Lovelace products are indeed launched during 2021 - it will be possible to declare a doubling of the maximum performance by 2 times in just over a year, in fact.
It is tempting to imagine the launch of the GeForce RTX 4000 generation with more monstrous performance than anything we have been selling somewhere in the middle of the current year, yet it is hard to ignore the same inventory and supply issues that make a headache forNVIDIA And for us since September 2020 and today. It is not inconceivable that the Lovelace generation will target the world ofTesla The prestige where the possibility of meeting the existing demand is much more reasonable, at least for a start - while the gaming market will settle for refreshed SUPER models in 2021 and advance to an era of 5 nanometers only in 2022.
On the other hand, AMD's reported plans to incorporate 5 nm lithography from TSMC into its RDNA 3 generation cannot be ignored, perhaps through the use of display separation for small chipsets that will significantly improve productivity similar to its desktop processor strategy - so increased technological competition can Drag the NVIDIA To a fast and somewhat surprising race.
We hope to receive more updates about the exciting future plans of our favorite hardware keys in the near future.