NVIDIA will likely hold onto the performance crown with its next-gen GeForce RTX 40 series graphics cards set to launch later this year. The GeForce RTX 4090, 4090 Ti, and the 4080 Ti will leverage the massive AD102 die packing as many as 18,432 cores, 96MB of L2 cache, and a 384-bit bus interface. Of course, all these ALUs will need at least 600W of power when fully enabled and running at their peak clocks.
| GPU | TU102 | GA102 | AD102 | AD103 | AD104 |
|---|---|---|---|---|---|
| Arch | Turing | Ampere | Ada Lovelace | Ada Lovelace | Ada Lovelace |
| Process | TSMC 12nm | Sam 8nm LPP | TSMC 5nm | TSMC 5nm | TSMC 5nm |
| GPC | 6 | 7 | 12 | 7 | 5 |
| TPC | 36 | 42 | 72 | 42 | 30 |
| SMs | 72 | 84 | 144 | 84 | 60 |
| Shaders | 4,608 | 10,752 | 18,432 | 10,752 | 7,680 |
| TP | 16.1 | 37.6 | ~100 TFLOPs? | ~50 TFLOPs | ~35 TFLOPs |
| Memory | 11GB GDDR6 | 24GB GDDR6X | 24GB GDDR6X | 16GB GDDR6 | 16GB GDDR6 |
| L2 Cache | 6MB | 6MB | 96MB | 64MB | 48MB |
| Bus Width | 384-bit | 384-bit | 384-bit | 256-bit | 192-bit |
| TGP | 250W | 350W | 600W? | 350W? | 250W? |
| Launch | Sep 2018 | Sep 2020 | Aug-Sep 2022 | Q4 2022 | Q4 2022 |
The AD102 die will get a haircut before going into the RTX 4090, dropping the core count from 18,432 to 16,384. This means that a few of the SMs, TPCs, and GPCs along with the L2 cache will also be axed. As for the clocks, Kopite7kimi states that we can expect core boost clocks well over 2.8GHz. This should result in a staggering single-precision throughput of almost 100 TFLOPs.
| GPU Name | RX 7700 XT (Navi 33) | RX 7800 XT (Navi 32) | RX 7900 XT (Navi 31) | RX 7950 XT (Navi 30?) |
|---|---|---|---|---|
| Process Node | 6nm | 5nm + 6nm | 5nm + 6nm | 5nm + 6nm |
| Die Config | 6nm x1 | 5nm x1, 6nm x4 | 5nm x1, 6nm x5 | 5nm x1, 6nm x6 |
| Shader Engines | 2 | 4 | 6 | 6 |
| WGPs | 16 | 32 | 48 | 48 |
| CUs | 32 | 64 | 96 | 96 |
| Cores | 4,096 | 8,192 | 10,752? | 12,288 |
| Memory Bus | 128-bit | 256-bit | 320-bit? | 384-bit? |
| Memory Type | GDDR6 | GDDR6 | GDDR6 | GDDR6 |
| Memory Capacity | 8GB | 16GB | 20GB? | 24GB? |
| Infinity Cache | 64MB | 128MB | 192MB | 192MB? |
| TGP | 200W | 250W? | 300? | 350W? |
| Launch | Q4 2022 | Q1 2023 | Q4 2022 | 2023 |
Moving on to the Radeon RX 7900 XT and the Navi 31 core powering it, we’re looking at a core count of 12,288 for the former and not a lot more for the latter. That’s a delta of 4,000 cores between the 7900 XT and the RTX 4090, and even wider if you consider the RTX 4090 Ti. AMD is rumored to be working on a larger Navi 3X core featuring up to 16,384 cores but even then the fully enabled AD102 die vis-Ã -vis the RTX 4090 Ti has a few thousand core advantage. There’s also no guarantee that it’ll actually enter volume production. Add to that the use of faster GDDR6X memory on the higher-end 40 series cards and Lovelace gets a bandwidth advantage.
There’s also the matter of the release dates. From what we know, AMD will kickstart the RDNA 3 family with the release of the Radeon RX 7700 XT in October, followed by the RX 7900 XT in November or December. This will give the top-tier RTX 4080 and 4090 to rule the enthusiast gaming market for a good one or two quarters. AMD may come up with superior GPUs in terms of performance per dollar and performance per watt but as usual, NVIDIA seems ontrack to retain the performance crown once again.
