NVIDIA’s next-gen Lovelace flagship, the RTX 4090 will reportedly feature 72MB of L2 cache. This is a steep cut from the 96MB available on the fully enabled AD102 die. According to Kopite7kimi, the memory subsystem of the upcoming Ada family is still family but “if everything goes well”, then the RTX 4090 will pack 72MB of last-level cache.
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? |
From previous rumors, we know that the RTX 4090 will be based on the AD102 die with 16,384 FP32 cores enabled and a bus width of 384-bit. The memory buffer will be once again limited to 24GB with faster GDDR6X 21Gbps memory chips, instead of 19.5Gbps. The RTX 4090 is likely to boast a single-performance rating of 90 TFLOPs. It will have a TGP of 450W, and launch sometime next month.
The RTX 4090 Ti (if it ever releases) should leverage the full-fat AD102 graphics core. we’re looking at roughly 18,000 cores paired with 48GB of GDDR6X memory running at 24Gbps. The L2 cache should also be 96MB wide with a TGP of 600-800W. The RTX 4090 Ti will likely launch sometime in mid-to-late 2023, following the RTX 4080 Ti and the 4070 Ti.