Next Generation NVIDIA GeForce RTX Lovelace GPU Rumors to Bring 18432 CUDA Samples for Sample Performance

Next Generation NVIDIA GeForce RTX Lovelace GPU Rumors to Bring 18432 CUDA Samples for Sample Performance

by
GeForce RTX 3090
We barely had a chance to familiarize ourselves with Ampere, with NVIDIA launching its GeForce RTX 30 series relatively recently. Nevertheless, the water cooler chatter has already turned to what comes next. According to rumors, there is a 5-nanometer “Lovelace” GPU architecture that threatens to eventually displace Ampere, and it is speculated that it could come with up to 18,432 CUDA cores.

Let’s put this in perspective, shall we? The current generation GeForce RTX 3090 is the fastest consumer card NVIDIA has ever introduced. It is built around a GA102 Ampere GPU with 72 power multiprocessors and 10,496 CUDA cores, along with 328 texture map units, 112 version output units, 328 Tensor cores, and 82 dedicated radiation detection cores.

You can look at our GeForce RTX 3090 review to see what numbers it can set up in games, but it’s simply the ‘fastest consumer GPU currently on the market, without any’, as we wrote. Imagine doubling the number of CUDA cores. For extra reference, here is the current range of NVIDIA …
  • GeForce RTX 3090: 72 SMs, 10,496 CUDA cores
  • GeForce RTX 3080: 68 SMs, 8,704 CUDA cores
  • GeForce RTX 3070: 46 SMs, 5,888 CUDA cores
  • GeForce RTX 3060 Ti: 38 SMs, 4,864 CUDA cores
Returning to the rumored Lovelace GPU, it appeared for the first time last week in a alleged roadmap by prominent gamer and Twitter user @ kopite7kimi. According to the analyst, NVIDIA initially planned to launch a GPU called Hopper before Lovelace, but things have apparently changed and Lovelace will debut first. Hopper will come out later and could possibly be strictly transferred to the high performance computer market (HPC).

The same player has now suggested that Lovelace could use an 18,432 CUDA core. Look…

NVIDIA Lovelace CUDA Cores

It sounds more like speculation than inside information, but given the leaker’s record, it can speculate based on inside information. Unfortunately @ kopite7kimi does not elaborate on the alleged 12 * 6 structure, but it is interesting nonetheless.

This amounts to an increase in graphics processing groups (GPCs) to 12, compared to a full-fat GA102 Ampere GPU with 7 GPCs. If accurate, Lovelace could potentially accommodate 72 texture processing groups (TPCs) and 144 streaming multiprocessers, providing a whopping 18,432 CUDA cores.

This would be an absolute sample if the information is accurate. And who really knows this at this early stage. Remember that Turing was first launched in September 2018, while Ampere came out two years later. It is therefore an architecture that we will only see in the latter part of 2022.

Source