Navi 48 leak changes the midrange power math
That draw comes with 50 TFLOPs, 64 compute units, a 256-bit bus, 693 GB/s of memory bandwidth, and a 240 mm2 die.
| Navi 48 | Navi 44 | |
|---|---|---|
| WGP | 32 | 16 |
| CUs | 64 | 32 |
| Memory bus | 256-bit | 128-bit |
| Memory bandwidth | 693 GB/s | 288 GB/s |
| Effective bandwidth | 2,770 GB/s | 515 GB/s |
| Die size | 240 mm2 | 130 mm2 |
| Compute | 50 TFLOPs | |
| Power | 215 W |
That draw comes with 50 TFLOPs, 64 compute units, a 256-bit bus, 693 GB/s of memory bandwidth, and a 240 mm2 die.
Against the RTX 4070 Ti SUPER, the Navi 48 sample sits at 50 TFLOPs and 215 W, while the card sits at 40 TFLOPs and 285 W.
That margin does not prove a shipping card will keep the compute, clocks, memory behavior, price, or lower-tier role.
Missing from the sheet are cache sizes, clock rates, process node, memory type, capacity, official TDP, and an exact launch date.
Navi 44 gets 32 compute units and a 130 mm2 die, but the sheet does not give its power or compute.
The 130 mm2 die points to a budget part, not a leap.
The sheet is a power-budget claim, not a performance verdict.
After notebookcheck.biz. We did not report this. The pictures, if any, are theirs.
