NVIDIA announced its RTX 40 series GPUs earlier this month to a moderate to lukewarm welcome. The RTX 4090 looks like a proper flagship with the RTX 4080 16GB holding its own, but its 12GB sibling is a bit of an abomination. Featuring the AD103 die, it packs 7,680 shaders (less than the RTX 3080 but more than the 3070 Ti) alongside a 192-bit bus and 48MB of L2 cache.
| GPU | GA102 | AD102 | RTX 4090 | AD103 | RTX 4080 16GB | AD104 | RTX 4080 12GB |
|---|---|---|---|---|---|---|---|
| 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 |
| Memory | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | ||||
| L2 Cache | 6MB | 96MB | 72MB | 64MB | 64MB | 48MB | 48MB |
| Bus Width | 384-bit | 384-bit | 256-bit | 192-bit | |||
| TGP | 450W | 450W | 320W | 285W | |||
| Price | $1,999 | $1,599 | $1,199 | $899 |
The RTX 4080 16GB leverages the larger AD104 die with 9,728 FP32 cores across a 256-bit bus and 72MB of L2 cache. It comes with a price tag of $1,199, $200 more than its 12GB sibling. That’s twice as much as the $100 delta between the ($499) RTX 3070, ($599) RTX 3070 Ti, and the ($699) RTX 3080.
This is the first time two memory variants of the RTX x80 are based on separate dies with different bus widths, board power, and cache. Furthermore, the price delta (but not the performance) is the same as that between the RTX 3070 and the RTX 3080. Quite frankly, I expect the prices of the upper midrange to drop through this cycle as AMD reveals its cards at the end of the year.
