oldpuck
Bob-omb
- Pronouns
- he/they
RE: WTF SAMSUNG 7BLT???
I'm not gonna conjecture about node likelihood in this post, just laying out some info. We somewhat fall victim to node branding here again. Here is a quick table that might help
As you can see from the table, there is no "official" 8nm node. 8nm (8LPP, more specifically) is an advancement of the "10nm" class nodde that Samsung had already developed.
7nm is the next official class down. Moving to Samsungs 7nm class nodes would be a significant leap, not "just" a further refinement of 10nm.
What about 7LPH? It doesn't exist, at least as far as the public is concerned. Whether it is the name of some customized 7nm class node of Samsung's is conjecture outside my scope here.
But is it any good? Well, it's hard to know because we don't have a lot of data to compare. But if we conjecture that it's roughly equivalent to the TSMC N7, and compare desktop RTX 30 to the server version, we can ballpark that it's a 30% performance/power improvement. But that is a lot of conjecturing.
Would that be good enough? Thraktor's power numbers have been predictive of power numbers from various graphics card efficiency tests, so I tend to still believe them. The implication seemed to be that Nintendo needed a full 50% increase in performance/power
That assumes that Nvidia can't find additional savings for this specific use case, and that Nintendo chooses to keep power draw roughly the same. These are not hard and fast truths.
I'm not gonna conjecture about node likelihood in this post, just laying out some info. We somewhat fall victim to node branding here again. Here is a quick table that might help
| "Official" node names/node classes | TSMC Node Names | SEC Node names |
| 10nm | 10FF | 10LPE |
| 10LPP | ||
| 8LPP (Orin/Desktop Ampere ) | ||
| 8LPA | ||
| 7nm | N7 (Server Ampere, Xbox Series, Steam Deck | 7LPP |
| N6 (PS5 "Slim", Steam Deck OLED) | 6LPP | |
| 5nm | N5 (Most RDNA 3) | 5LPE |
| 4N (Lovelace) | 5LPP | |
| 4LPE | ||
| 4LPP | ||
| 3nm | N3 (Blackwell? iPhone 15) | 3GAE |
As you can see from the table, there is no "official" 8nm node. 8nm (8LPP, more specifically) is an advancement of the "10nm" class nodde that Samsung had already developed.
7nm is the next official class down. Moving to Samsungs 7nm class nodes would be a significant leap, not "just" a further refinement of 10nm.
What about 7LPH? It doesn't exist, at least as far as the public is concerned. Whether it is the name of some customized 7nm class node of Samsung's is conjecture outside my scope here.
But is it any good? Well, it's hard to know because we don't have a lot of data to compare. But if we conjecture that it's roughly equivalent to the TSMC N7, and compare desktop RTX 30 to the server version, we can ballpark that it's a 30% performance/power improvement. But that is a lot of conjecturing.
Would that be good enough? Thraktor's power numbers have been predictive of power numbers from various graphics card efficiency tests, so I tend to still believe them. The implication seemed to be that Nintendo needed a full 50% increase in performance/power
That assumes that Nvidia can't find additional savings for this specific use case, and that Nintendo chooses to keep power draw roughly the same. These are not hard and fast truths.
