NVIDIA has its next-generation GeForce RTX 40 series gaming graphics cards, based on the Ada Lovelace GPU architecture, getting ready for a 2022 release. According to DigiTimes’ most recent report, partners including Taiwan factories that work with NVIDIA are preparing for a big GPU upgrade next year in the shape of the GeForce RTX 40 series.
We’ve already heard that NVIDIA may use TSMC’s 5nm production node for its next-generation gaming GPUs dubbed Ada Lovelace from trustworthy leakers, but this time the information comes directly from the Taiwanese factory where these GPUs would be manufactured. While the DigiTimes piece is behind a paywall, RetiredEnginner (@chiakokhua) exposed a portion of the material on Twitter.
The NVIDIA Ada Lovelace GPUs will power the next-generation GeForce RTX 40 graphics cards, which will compete with AMD’s Radeon RX 7000 series graphics cards based on RDNA 3. There is still considerable speculation about NVIDIA’s usage of MCM. The Hopper GPU, which is primarily geared at the Datacenter & AI market, is rumoured to be nearing completion and will have an MCM CoWoS architecture. NVIDIA will not use an MCM architecture on its Ada Lovelace GPUs, instead opting for the standard monolithic design. Architecturally, the Ada Lovelace GPUs are projected to introduce some major advances.
NVIDIA GeForce RTX 4090 Graphics Card – Ada Lovelace Powered AD102 Flagship GPU
Previous speculations suggested that NVIDIA will use TSMC’s N5 (5nm) manufacturing node for its Ada Lovelace GPUs. This includes the AD102 SKU, which will be completely monolithic. In his most recent tweet, which discusses particular GPU combinations, the AD102 GPU is stated to have a clock speed of up to 2.5 GHz (2.3 GHz average boost). The exact tweet specifies that the GPU frequency for Ada Lovelace ‘AD102’ might be 2.3 GHz or higher, so let’s use it as a baseline, along with previously disclosed data, to figure out where the performance should fall.
Based on preliminary specs (which may change), the NVIDIA AD102 “ADA GPU” looks to contain 18432 CUDA Cores packed within 144 SM units. This is nearly double the number of cores found in Ampere, which was already a significant improvement over Turing. A clock speed of 2.3-2.5 GHz would provide up to 85-92 TFLOPs of computational capability (FP32). This is more than double the FP32 performance of the current RTX 3090, which has 36 TFLOPs of FP32 computation capability.
The 150 percent performance increase appears to be significant, but keep in mind that NVIDIA has provided a significant increase in FP32 numbers this generation with Ampere. The Turing TU102 GPU (RTX 2080 Ti) delivers 13 TFLOPs, whereas the Ampere GA102 GPU (RTX 3090) gives 36 TFLOPs. That’s a more than 150 percent boost in FP32 Flops, yet the RTX 3090’s real-world gaming performance gain averaged approximately 50-60% quicker than the RTX 2080 Ti. So, one thing to keep in mind is that these days, flops do not imply GPU gaming performance. Furthermore, we don’t know if 2.3-2.5 GHz is the average boost or the peak boost, with the former implying that AD102 has considerably greater computing capacity.
Aside from that, the leaker claims that the NVIDIA GeForce RTX 40 flagship will keep the same 384-bit bus interface as the RTX 3090. What’s intriguing is that the leaker cites G6X, implying that NVIDIA would wait until after Ada Lovelace and use the higher pin-speeds of G6X of 21 Gbps for its next-generation cards before migrating to a newer standard (e.g. GDDR7). The card will have 24 GB of memory, which means we may expect either single-sided 16GB DRAM or dual-sided 8GB DRAM modules.
NVIDIA CUDA GPU (RUMORED) Preliminary:
| GPU | TU102 | GA102 | AD102 |
| Architecture | Turing | Ampere | Ada Lovelace |
| Process | TSMC 12nm NFF | Samsung 8nm | 5nm |
| Graphics Processing Clusters (GPC) | 6 | 7 | 12 |
| Texture Processing Clusters (TPC) | 36 | 42 | 72 |
| Streaming Multiprocessors (SM) | 72 | 84 | 144 |
| CUDA Cores | 4608 | 10752 | 18432 |
| Theoretical TFLOPs | 16.1 | 37.6 | ~90 TFLOPs? |
| Memory Type | GDDR6 | GDDR6X | GDDR6X |
| Memory Bus | 384-bit | 384-bit | 384-bit |
| Memory Capacity | 11 GB (2080 Ti) | 24 GB (3090) | 24 GB (4090?) |
| Flagship SKU | RTX 2080 Ti | RTX 3090 | RTX 4090? |
| TGP | 250W | 350W | 450-650W? |
| Release | Sep. 2018 | Sept. 20 | 2022 (TBC) |
The NVIDIA Ada Lovelace GPUs will power the next-generation GeForce RTX 40 graphics cards, which will compete with AMD’s Radeon RX 7000 series graphics cards based on RDNA 3. There is still considerable speculation about NVIDIA’s usage of MCM.
The Hopper GPU, which is primarily geared at the Datacenter & AI market, is rumoured to be nearing completion and will use an MCM design. NVIDIA will not use an MCM architecture on its Ada Lovelace GPUs, instead opting for the standard monolithic design.
