Bd derived cores is just crap vs a modern core like zen.
Well, that is a comparison of products of largely different eras.
The general idea of a FDX product is that the density and perf/watt would catch up.
28nm 9T -> 22FDX 8T => ~0.7x Logic scaling / ~0.8x Register/SRAM scaling
^-- this is enough area to complete the ALUs in the AGUs. To get four EX units, two are complex(MUL/DIV) and two are Agen focused. This is relatively minor to what they did with the FPU between BD and SR. As the integer scheduler is already quad-ported in the cores. Arithmetic(EXs) and logic(AG Logic Us) units share a reservation station. Since, there is two cores and no sharing on the important bits, 8 EXs is equivalent to 2-cores of Zen's EX.
22FDX 8T -> 12FDX 7.5T => ~0.6x Logic scaling / ~0.55x Register/SRAM
^-- this shrink allows for the architecture to use Zen's tiles/macros/datapaths. This comes with the added advantages...
So, physical design comes from Zen which is 14LPP 9T and is ported and optimized to 12FDX 7.5T. The architecture baseline comes from 28nm and is ported to 12FDX 7.5T.
Enhanced Integer IPC. The FPU is up in the air if it will Zen-derived; 2x FMAC via bridged 2 MUL and 2 ADD or XV-derived; 2x FMACs via 2x Fused MACS. FDX has single-well and mixed-well SRAM/PRFs/CAMs, etc. The FDX design can be quite a bit more dense than the FinFET physical design.
Basically, a 12FDX CPU can be at worst a speedy and efficient Excavator-esque module and at best a module with two Zen-esque cores with a single Zen FPU. The latter is actual much closer to the original Cluster-based Multithreaded design for the 2005-slide.
An Excavator core has two macro-op decoders with two cycles to get four macro-ops.
Whereas a Zen core has four macro-op decoders which operates at same principle as the Bulldozer's module four macro-ops. Two threads get 4 macro-ops every two cycles.
Excavator-derived core rigged with Zen IP, still fits the portfolio of Zen-lite.
22FDX => 50% lower power w/ same design
12FDX => 50% lower power w/ same design
Redesign/New design => 60% lower power
Rerouted and replaced => 40% lower power
Improved/Redesigned power delivery => 30% lower power
so on, so forth.
1. The product will have the same paper specifications(core count + compute unit count) as Bristol Ridge and Stoney Ridge.
2. The product will be on a newer node, specifically 12FDX.
3. The product will have higher power efficiency targets than Zen/Zen+, however lower total performance targets than Zen/Zen+.
4. The product will cost less than its counterparts, example;
Renoir(1xx mm2 die) => Less than or equal to $160 // 12FDX BR(1xx mm2 die) => Less than or equal to $60(closer to $40-$50)
Raven2(1xx mm2 die)/Renoir2(0xx mm2 die) => Greater than or equal to $60 // 12FDX SR(0xx mm2 die) => Less than or equal to $30(closer to $15-$20)
5. Do to GlobalFoundries more aggressive ramping, the production of the lineup is no longer tied to a single foundry location. If AMD wants to target a specific region, they can target the region with the fab. Examples being Chengdu => APAC/Greater China, Malta => Americas, Dresden => EMEA.
FDX => Cost-effective performance // Cost-effective -> Economical in terms of the goods or services received for the money spent.
FinFET => Performance and density at any cost. // at all costs (or at any cost) -> In any way possible, regardless of risk or expense.
Just wanted to post this.
There was also an extremely close relationship between 12FDX and 7nm;
- Development on planar and finFET technologies: 22nm 14nm 12FDX 7nm at GF Fab 8 / Fab 10
- Pre-clean chamber and Producer and Precision CVD products for customer 7nm and 12FDX technology. <-(customer is GlobalFoundries)
It is also closely related to the FinFET nodes overall;
- 14LPP/12LP/12FDX Device Director
- Working on 7nm,14LPP,12FDX
- •GLOBALFOUNDRIES 14LPP Mobility Solution ... •GLOBALFOUNDRIES 7nm FinFET Mobility Solution ... •GLOBALFOUNDRIES 12FDX SOI Technology
- Experienced Tech node: 7nm, 12FDX, 14nm.
12FDX European project is known,
12FDX American project is known,
12FDX Chinese project is I don't know; GloFo's SVP of Asia business development, Americo Lemos => Looking ahead, Lemos said that GloFo expects to roll out its 12nm FD-SOI (12FDX) process as early as 2019. => 19 February 2019, 12:11, DigiTimes
All three should hopefully be producing 12FDX by the end of 2019.
- If late last year was v0.2 qualification PDK for 12FDX
and Q1 2016 was v0.2 for 22FDX, Q4 2016 for v1.0
Then, v1.0 should be Q3/2H 2019.