Oban

From XenonLibrary
(Redirected from Winchester CGPU)
Jump to navigation Jump to search
Oban
X888839-001.jpg
Part NumberX888839
X888841
IntroducedLate 2014
Designed ByIBM
Microsoft
TypeXCGPU
Used InWinchester
Clock Speed3.2GHz (CPU)
500MHz (GPU)
CacheL1: 32KB/32KB (CPU)
L2: 1MB (CPU)
Memory10MB eDRAM (GPU)
Instruction SetPowerPC RISC (CPU)
Cores3 (CPU)
Threads6 (CPU)
Process32nm
PackageFlip-chip BGA899
PredecessorVejle

Oban is the final XCGPU and is used on the Winchester motherboard. It is the successor to Vejle and is manufactured on the 32nm process. The chip combines the eDRAM into the CGPU die and thus the integrated heat spreader is no longer used.

Specifications

CPU

  • 3 two-way SMD-capable RISC cores clocked at 3.2GHz
  • SIMD: Two VMX128 units
  • 32KB L1 data cache
  • 32KB L1 instruction cache
  • 1MB L2 cache at 1.6 GHz with a 256-bit bus
  • 21.6GB/s FSB
  • 768 bits of IBM eFUSE One-Time-Program memory for fusesets
  • One-Time-Programmable ROM and 64KB SRAM for the 1BL and Config Ring
  • Big-endian architecture
  • A03 stepping
  • 32nm process

GPU

  • 500Mhz clock speed
  • 48 floating-point vector processors divided into 3 dynamically scheduled SIMD groups (16 each)
  • Unified shader architecture
  • 16 texture addressing units
  • 16 texture filtering units
  • 8 pixel rendering pipelines
  • Direct access to CPU L2 cache
  • A03 stepping
  • 32nm process

eDRAM

  • 10MB 256GB/s eDRAM
  • Contains logic for:
    • 4-sample anti-aliasing
    • Alpha compositing
    • Color
    • Z/stencil buffering
  • A03 stepping
  • 32nm process

Details

  • Part Numbers:
    • X888839
    • X888841

Prototypes

Engineering Sample, stepping is unknown.

Oban-ES.jpg