Instead of holding the bus or retry loops like PCI, PCI-X splits the read into two independent bus transactions:

1. Read Request Phase (by the Requester):

   • Requester = initiator of the read.
   • Sends:
     • Address
     • Transaction type (read)
     • Byte count (how much data)
     • Requester ID (Bus:Device:Function number)

   👉 If the target (now called the Completer) can’t supply the data immediately, it responds with a Split Response.

   ✅ At this point:

   • The requester stops holding the bus.
   • The bus returns to idle, so other devices can use it.

2. Split Completion Phase (by the Completer):

   • Once the completer has the requested data ready:
     • It arbitrates for bus ownership.
     • Starts a Split Completion cycle, which looks a lot like a write.
   • The Requester recognizes its own ID in the completion and accepts the data.

   👉 No retries, no wasted polling — the completer just brings the data back when ready.

Key Efficiency Gains

• Bus is free while data is being fetched (no more “wait” hogging).
• Requester can do other work (even issue other requests).
• Only two bus transactions needed:
  • Request
  • Completion
    → Much more efficient than PCI’s repeated retries.

Naming Change

• PCI: Master ↔ Target
• PCI-X: Requester ↔ Completer
  (to reflect the split responsibility across two separate transactions).

Bus Utilization

• PCI efficiency: ~50–60%
• PCI-X efficiency: ~85%
  → This is a huge leap, especially in servers where multiple devices are competing for bandwidth.