AMD's rumoured to tap TSMC and Samsung when they move to 5nm

Smart move from a business perspective. Get even better pricing and since AMD already is cheaper than Intel it can either lower cost further or maintain same prices for better margins, which wouldn't be bad as they need all the money they can get in the GPU space.

this just begs the question of quality/parity between the CPUs since they will be using different manufacturers.
 
Dual sourcing is regularly used by NVidia & AMD to be fair, there's never any issue of quality of parity because they choose the foundry on a per-product basis, the only primary exceptions of dual sourcing being used for the same IC is with Apples SoCs. However other parts like Vega's interposer were still multi-sourced. For Pascal, NVidia used Samsung for their lower end cards & TSMC for the high end. AMD planned to dual source on 7nm but then GloFlo's 7nm vanished, while Samsung doesn't have a first gen 7nm and is instead waiting for EUV to mature, so 7nm being single sourced has been more of an exception than a rule.

Dual sourcing is actually pretty expensive business especially given many foundrys competing nodes arn't directly compatible/portable with each other, some times even with a fully complete design it could take many many months of reconfiguring it to suit a different node, but it's kinda necessary when the stakes & costs are this high, planning to single source is kinda suicide as any little delay could ruin & knock off your whole product line for years, just look at Intel.
 
Last edited:
Dual sourcing is regularly used by NVidia & AMD to be fair, there's never any issue of quality of parity because they choose the foundry on a per-product basis, the only primary exceptions of dual sourcing being used for the same IC is with Apples SoCs. However other parts like Vega's interposer were still multi-sourced. For Pascal, NVidia used Samsung for their lower end cards & TSMC for the high end. AMD planned to dual source on 7nm but then GloFlo's 7nm vanished, while Samsung doesn't have a first gen 7nm and is instead waiting for EUV to mature, so 7nm being single sourced has been more of an exception than a rule.

Dual sourcing is actually pretty expensive business especially given many foundrys competing nodes arn't directly compatible/portable with each other, some times even with a fully complete design it could take many many months of reconfiguring it to suit a different node, but it's kinda necessary when the stakes & costs are this high, planning to single source is kinda suicide as any little delay could ruin & knock off your whole product line for years, just look at Intel.

Yeah, AMD is likely to be using different nodes on a per product basis. AMD wouldn't waste the money for making a single product on two nodes unless it was an absolute necessity.

The only case AMD has done this in recent history is dual sourcing GlobalFoundries and Samsung, as their 14nm nodes are basically the same thing. As GF Licensed Samsung's node for 14nm.

AMD is right to keep their options open with 5nm, as Globalfoundries has shown everyone that you shouldn't rely on a single source for a new node. Thankfully TSMC and Samsung have the money behind them to keep them in the Leading Edge silicon race, but it is sad to see only two companies at the top these days. Intel doesn't really count, seeing as they only really produce parts for themselves.
 
Back
Top