![NVIDIA RTX 4070 Ti to Be as Fast as the RTX 3090 Ti with 7,680 Cores, 48MB L2 Cache, and 12GB GDDR6X Memory [Report]](https://i0.wp.com/michmutters.com/wp-content/uploads/2022/08/NVIDIA-RTX-4070-Ti-to-Be-as-Fast-as-the.jpg?fit=1200%2C387&ssl=1)
NVIDIA will be offering some serious gaming firepower with its upcoming RTX 40 series graphics cards. Codenamed Ada Lovelace, these GPUs will be the first to crack the 100 TFLOPs barrier, thanks to a brand new microarchitecture and process node. The top end AD102 die is set to feature a core count of 18,432 alongside 96MB of L2 cache and a 384-bit bus. The memory buffer will vary from 24GB to 48GB depending on the SKU. The TGP may exceed 600W+ when fully enabled (RTX 4090/4090 Ti).
| GPU | GA102 | AD102 | RTX 4090 | AD103 | RTX4080 | AD104 | RTX4070 |
|---|---|---|---|---|---|---|---|
| Arch | Ampere | Ada Lovelace | Ada Lovelace | Ada Lovelace | |||
| Process | Sam 8nm LPP | TSMC 5nm | TSMC 5nm | TSMC 5nm | |||
| CPG | 7 | 12 | eleven | 7 | 7 | 5 | 5 |
| TPC | 42 | 72 | 64 | 42 | 40 | 30 | 28 |
| SMS | 84 | 144 | 128 | 84 | 80 | 60 | 56 |
| shaders | 10,752 | 18,432 | 16,384 | 10,752 | 10,240 | 7,680 | 7,168 |
| PT | 37.6 | ~100 TFLOPs? | 90 TFLOPs? | ~50 TFLOPs | 47 TFLOPs? | ~35 TFLOPs | 33 TFLOPs? |
| Memory | 24GB GDDR6X | 24GB GDDR6X | 16GB GDDR6X | 10GB GDDR6 | |||
| L2 Cache | 6MB | 96MB | 72MB | 64MB | 48MB | ||
| Bus Width | 384-bit | 384-bit | 256-bit | 160-bit | |||
| TGP | 350W | 600W | 450W | 450W | 420W | 300W | 300W |
| Launch | Sep 2020 | September 22? | September 22? | September 22? |
Moving down the stack, we have the AD103 and AD104 which will power the RTX 4080 and 4070/4070 Ti, respectively. The former will feature 10,240 shaders alongside 16GB of GDDR6X memory across a 256-bit bus and ~60MB of L2 cache. It will have a TGP of 400W+. It should have a single-precision performance rating of 50-60 TFLOPs, roughly half as much as the RTX 4090.
Finally, we have the AD104 core which will form the backbone of the more mainstream RTX 4070 and 4070 Ti. The former will leverage a partially disabled die with 7,169 cores and a 160-bit bus paired with 10GB of GDDR6 memory and 40-44MB of L2 cache. The Ti will feature the fully enabled implementation of the AD103 die with 7,680 shaders, 48MB of L2 cache, and 12GB of GDDR6X memory.
According to Kopite7kimi, the RTX 4070 Ti will be as fast as the RTX 3090 Ti, the fastest GPU on the planet right now. It’ll have a TGP of 400W and enable top-notch 4K gaming, maybe even with ray-tracing enabled. It’s worth remembering that the RTX 4090/4090 Ti will be twice as fast as (if not more) the RTX 3090/3090 Ti with several-fold higher ray-tracing performance. The RTX 4080 should be ahead of the RTX 3080 by at least 60-70%.
the RTX 4090 is set to launch next month with the RTX 4080/4070 landing either during the holiday season or in the first quarter of 2023.