The specifications of NVIDIA’s midrange Lovelace GPU have become clearer. The RTX 4070 will reportedly leverage 12GB of high-speed GDDR6X memory clocked at 21Gbps alongside the full-fat AD104 die. Featuring 7,680 FP32 cores across 60 SMs, 30 TPCs, and 5 GPCs, we’re possibly looking at at least 35-40 TFLOPs of single-precision performance. Since we’re talking about the fully enabled die, the L2 cache should also be untouched at 48MB.
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? |
The RTX 4070 will be the first RTX x70 card to feature GDDR6X memory, a technology NVIDIA has limited to the $500+ segments. However, at the same time, the bus width has also been reduced to 160-bit keeping the bandwidth largely the same. The TGP is also a very respectable 300W, 80W more than its predecessor, the RTX 3070. Keep in mind that we’re looking at RTX 3090 to 3090 Ti levels of performance in a $400-500 SKU here, so this is nothing to scoff at.
It’s unclear whether the RTX 4070 will launch this year or early next year but going by leaks and rumors, a CES 2023 release is highly probable.