The specifications of NVIDIA’s next-gen RTX 40 series flagship are becoming clearer with each passing day. Well-reputed leaker Kopite7kimi has shared the fine-grained details of the AD102 die set to power the RTX 4090 and 4090 Ti. The top-end Lovelace die will apparently feature close to 80 billion transistors with a single-precision compute performance of over 100 TFLOPs.
Remember this. 😁Most people made the right choice. More than 75B.
— kopite7kimi (@kopite7kimi) September 13, 2022
According to Kimi, both the RTX 4090 and 4090 Ti will be able to touch or even cross the 100 TFLOP single-performance mark. The former should attain the same with a mild overclock while the latter ought to break it at stock frequencies itself.
| GPU | GA102 | AD102 | RTX 4090 | AD103 | RTX 4080 | RTX 4070 Ti (AD104) | RTX 4070 |
|---|---|---|---|---|---|---|---|
| Arch | Ampere | Ada Lovelace | Ada Lovelace | Ada Lovelace | |||
| Process | Sam 8nm LPP | TSMC 5nm | TSMC 5nm | TSMC 5nm | |||
| GPC | 7 | 12 | 11 | 7 | 7 | 5 | 5 |
| TPC | 42 | 72 | 64 | 42 | 40 | 30 | 30 |
| SMs | 84 | 144 | 128 | 84 | 80 | 60 | 60 |
| Shaders | 10,752 | 18,432 | 16,384 | 10,752 | 9,728 | 7,680 | 7,680 |
| TP | 37.6 | ~100 TFLOPs? | 83 TFLOPs | ~50 TFLOPs | 47 TFLOPs? | ~35 TFLOPs | 35 TFLOPs? |
| Memory | 24GB GDDR6X | 48GB GDDR6X | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | ||
| L2 Cache | 6MB | 96MB | 72MB | 64MB | 48MB | ||
| Bus Width | 384-bit | 384-bit | 256-bit | 192-bit | |||
| TGP | 350W | 600W | 450W | 450W | 285-340W | 300W | 285W |
| Launch | Sep 2020 | Sept 22? | Sept 22? | Q1 2023? |
NVIDIA is all set to announce its next-gen GeForce RTX 40 series graphics cards during its GTC 2022 keynote on the 20th of September. Company CEO Jensen Huang will unveil the RTX 4090 and possibly the RTX 4080 next week with a launch allegedly planned for October and November, respectively.
The top-end RTX 4090 will be the first to launch between early and mid-October, with an announcement slated for the following Monday. Based on the AD102 core, it will feature a partially disabled die with 16,384 FP32 cores and 24GB of GDDR6X memory across a 384-bit bus and 72MB of L2 cache.
The RTX 4080, on the other hand, will come in two variants: One with 12GB of GDDR6X memory and the other with 16GB. The former will utilize a 10-layer PCB and the latter will have a 12-layer PCB. Both will leverage the AD103 die alongside a 256-bit bus and 64MB of L2 cache. (Provigil)
The 12GB SKU is expected to have a TGP of 285W and the 16GB part will raise it all the way to 340W. As such, the performance delta between them could very well be over 20%. This is notably larger than the delta between the two variants of the RTX 3080, putting the latter on a separate price/performance slab.
The RTX 4070 will be based on the AD104 core with 7,680 shaders across 60 SMs paired with 12GB of GDDR6X memory via a 160 or 192-bit bus. It will come with 48MB of L2 cache and a TGP of up to 300W. The RTX 4060 is planned for an early 2023 release/unveil. It’ll offer a more conservative uplift than its brethren but should be one of the more popular parts, nonetheless. You can read more about the RTX 40 series GPUs below:
- NVIDIA RTX 4060 Allegedly 70% Faster than the RTX 3060 in TimeSpy Extreme, Limited to PCIe x8 Lanes
- NVIDIA RTX 4090 >2x Faster than the RTX 3090 in Leaked Bench: 3GHz+ Boost Clocks, 21Gbps GDDR6X Memory
- NVIDIA RTX 4090 Ti May be 2.3x Faster than the RTX 3090 Ti, 2.6x Faster than the RTX 3090 in TimeSpy Extreme [Rumor]
- NVIDIA is Set to Announce the RTX 4080/4090 on the 20th of September at Voyager HQ
