Golden Cove Processor

What is Golden Cove Processor?

Golden Cove processors come with an architecture that is fabricated on the 10 nm Enhanced SuperFin node or Intel 7 process node. The cores of these CPUs are designed to support hyperthreading, thereby allowing running two threads on a single core.

Technically, these processors come with a better front-end and back-end as well as other improved features such as simple decoders, Branch Target Buffers, Reorder Buffers, and execution ports.

Understanding Golden Cove Processor

Golden Cove is the codename of the microarchitecture of the processors developed and manufactured by Intel and released on November 4, 2021.

Typically, the architecture of these processors is the successor to four different architectures, such as:

• The 10 nm 1S and 2S Sunny Cove servers
• The 14 nm 4S and 8S Skylake servers
• The 10 nm mobile Willow Cove processors
• The 14 nm Cypress Cove desktop processors

Built on the Intel 7 process node, this particular microarchitecture is typically used in the P-core or high-performance cores of the 12th generation Intel Core Alder Lake processors.

Apart from that, the microarchitecture is capable enough to power up the 4th generation Xeon Scalable Sapphire Rapids server processors as well.

In addition to that, the Golden Cove microarchitecture also allows splitting the execution ports using a wider decoder.

This allows for handling and executing a larger number of operations at the same time.

This feature eventually allows for higher Instructions per Cycle (IPC) and Instruction Level Parallelism (ILP) to take full advantage of a given workflow.

Typically, a wider decode will need a lot more power to operate, but in the case of the Golden Cove architecture, Intel claims that it will not need as much power due to its 4K micro-op cache and a more improved front-end.

These design features allow the decoding engine to spend 80% of the time power gated.

Typically, the maximum clock rate attained by the cores of this CPU ranges between 1 GHz and 5.5 GHz.

Architectural Changes

Some of the architectural changes made in the Golden Cove with respect to the architecture of its predecessor, Willow Cove, are as follows:

• Improved and stronger Instructions per Cycle (IPC) by about 19%
• Improved AI workload with Advanced Matrix Extensions (AMX)
• Enhanced 5G network performance
• Better security features
• Additional Arithmetic Logic Unit (ALU) and Load Effective Address (LEA) instruction in the execution engine to make the total up to 5.

And, with respect to its other predecessor, Sunny Cove, the major architectural changes are as follows:

• A 6-wide partial instruction decoder in place of the 4-wide decoders with the ability to fetch up to 32 bytes instead of 16 instructions per cycle
• Removed complex decoder in spite of its wider 6-wide microarchitecture
• Increased micro op cache size from 2.25K to 4K entries
• Twelve execution ports in place of ten ports
• A bigger out-of-order instruction window
• Increased size of the Reorder Buffer from 352 to 512 entries
• A bigger vector or floating-point register file of 332 entries instead of 224
• 192 load queues in place of 128
• 114 store queues in place of 72
• 1.25 MiB Level 2 cache per core for consumer CPUs and 2 MiB L2 cache size per core for the server processors
• Larger number of instruction extensions support
• Exclusive floating-point adders

Front-end and Back-end

The front end of the architecture also comes with some significant improvements and additional features as follows:

• Two additional simple decoders to bring the total up to six simple decoders
• x2.5 Branch Target Buffer (BTB) at 12K entries
• 2x pages in 4K
• Additional 256 and 32 pages of 2 MB and 4 MB, respectively

As for the back end, the features include the following:

• Improved Reorder Buffer (ROB) 512
• Two additional execution ports to bring the total up to 12

Instruction Set and Extension Support

There are several new and old extensions of x86-64 Instruction Set Architecture (ISA) supported by the Golden Cove processors as follows:

• CLDEMOTE or Cache Line Demote
• PTWRITE instructions to write data to a processor trace packet
• User-wait instructions such as TPAUSE or Time Pause, UMONITOR or User Level Set Up Monitor Address, and UMWAIT or User Level Monitor Wait
• Architectural LBRs or Last Branch Records
• HLAT or Hypervisor-managed Linear Address Translation
• SERIALIZE Instruction Execution
• EHFI or Enhanced Hardware Feedback Interface
• HRESET or History Reset
• AVX or Advanced Vector Extensions, along with AVX2
• AVX-VNNI or Advanced Vector Extensions Vector Neural Network Instructions
• AVX-512 with Advanced Vector Extensions 512 Half-precision Floating Point Format or AVX 512-FP16
• AES-NI or Advanced Encryption Standard New Instructions
• CLMUL or Carry-less Multiplication
• RDRAND or Read Random
• SHA or Secure Hashing Algorithm
• TXT or Text File extension
• MMX or MultiMedia eXtensions
• SSE or Streaming SIMD Extensions, along with all its variants such as SSE2, SSE3, SSSE3, SSE4, SSE4.1, and SSE4.2
• FMA3 or Fused Multiply Add
• TSX or Transactional Synchronization Extensions
• VT-x and VT-d or Virtualization extensions

The server Sapphire Rapids processors, however, support TSX with Transactional Synchronization Extensions Suspend Load Address Tracking, or TSXLDTRK.

Cache Memory

The Golden Cove processors come with three levels of cache memory in them as follows:

• The Level 1 cache memory measures 80 KB per core, with 32 KB for instructions and 48 KB for data
• The Level 2 cache   memory measures 1.25 MB per core for the client processors and 2 MB per core for the server processors
• The Level 3 cache   memory measures 3 MB per core.

With all these features in them, the Golden Cove processors are true and worthy competitors of the Zen 3 and Zen 4 processors from the house of AMD.

In spite of the smaller 10 nm Enhanced SuperFin node, these Intel processors have more Out-of-Order Execution (OoOE) capabilities due to their extended back and front-ends.

This allows them to offer real-world performance while maintaining power efficiency all throughout.

Golden Cove vs Skylake

• Golden Cove is comparatively a more modern technology, being released on November 4, 2021, as opposed to the Skylake processors that were released on August 5, 2015.
• The maximum operating speed of the Golden Cove processors ranges between 1 GHz and 5.5 GHz. On the other hand, the cores of the Skylake processors can reach a maximum operating speed of up to 4.5 GHz.
• The Level 1 cache memory in the Golden Cove processors measures 80 KB per core, with 32 KB reserved for instructions and 48 KB for data. On the other hand, the Level 1 cache memory of the Skylake processors measures 64 KB for each core.
• The Level 2 cache memory in the Golden Cove processors measures 1.25 MB and 2 MB per core in the client and server processors, respectively. On the other hand, the Level 2 cache memory of the Skylake processors measures 256 KB for each core and 1 MB per core in the Skylake-X processors.
• The Level 3 cache memory in the Golden Cove processors measures 3 MB per core, but in comparison, the Level 3 cache memory of the Skylake processors measures up to 2 MB for each core and 1.375 MB per core in the Skylake-X CPUs.
• There is no Level 4 cache memory in the Golden Cove processors, but in comparison, in some specific models of the Skylake processors, there is a Level 4 cache memory of 128 MB of Embedded Dynamic Random Access Memory or eDRAM.
• The technology on which the Golden Cove processors are built is referred to as the 10 nm Enhanced SuperFin or ESF node. On the other hand, the Skylake processors are built on the 14 nm manufacturing process using bulk silicon 3D tri-gate transistors.
• The predecessors of Golden Cove processors are Sunny Cove, Willow Cove, Cypress Cove and Skylake. On the other hand, the predecessor of the Skylake processors is the Broadwell processors.
• The successor to the Golden Cove processors is Raptor Cove. However, the successors to the Skylake processors are the Kaby Lake, Cascade Lake, and Palm Cove processors.

Read Also:  What is Clock Doubling? (Explained)

Conclusion

The Golden Cove processors, the successors to Willow Cove, come with the features that make them a true competitor of the Zen 3 and Zen 4-based CPUs from AMD.

The 10 nm Enhanced SuperFin node, the extended front and back ends, enhanced OoOE and other features offer better and higher performance.