Intel Core Ultra 5 225T Processor
Exhibit Type: Desktop & Edge Hardware Specification & Architectural Dossier
•
Microarchitecture: Arrow Lake-S (TSMC 3nm N3B / Foveros 3D)
•
Socket: Socket LGA 1851 (FC-LGA18W)
•
Museum Inventory: 1 Unit (Installed in HP ProDesk 4 Mini G1i AI Black • S-Spec: SRVFA)
•
Source Authority: TechPowerUp, Intel & PassMark
•
Microarchitecture: Arrow Lake-S (TSMC 3nm N3B / Foveros 3D)
•
Socket: Socket LGA 1851 (FC-LGA18W)
•
Museum Inventory: 1 Unit (Installed in HP ProDesk 4 Mini G1i AI Black • S-Spec: SRVFA)
•
Source Authority: TechPowerUp, Intel & PassMark
1. Technical Overview & Key Specifications
| Category | Property | Specification |
|---|---|---|
| Processor Identity | Model Name | Intel® Core™ Ultra 5 225T (SKU: AT8076806772 • PassMark ID: 6550) |
| Microarchitecture | Arrow Lake-S (Intel Core Ultra 200 Series) | |
| Market Segment | Desktop / Energy-Efficient AI PC / Mini PC Workstation | |
| Launch Date | January 13, 2025 (CES 2025 Launch • PassMark charts Q1’25) | |
| Part Numbers / S-Spec | SRVFA (OEM / Tray: AT8076806772 • Verified Museum Inventory: 1 unit) |
|
| Physical & Manufacturing | CPU Socket | Socket LGA 1851 (FC-LGA18W package, 1851-pin) |
| Packaging Technology | Disaggregated 3D Foveros Packaging (Active Modular Tile Architecture) | |
| Fabrication Nodes | Compute Tile: TSMC 3 nm (N3B) • GPU Tile: TSMC 5 nm (N5P) • SoC & I/O: TSMC 6 nm (N6) • Base Tile: Intel 16 (22 nm) | |
| Transistor Count | 17.8 Billion transistors (Combined multi-tile complex) | |
| Die / Package Area | 45.0 × 37.5 mm package (243 mm² aggregate package die envelope) | |
| Thermal Junction Max | 105°C (tJMax) | |
| Cores & Performance | Physical Cores / Threads | 10 Cores / 10 Threads (Hybrid: 6 P-Cores + 4 E-Cores • Hyper-Threading omitted for IPC efficiency) |
| Performance Cores (Lion Cove) | 6 P-Cores • 2.50 GHz Base • Up to 4.90 GHz Max Turbo Frequency | |
| Efficient Cores (Skymont) | 4 E-Cores • 1.90 GHz Base • Up to 4.40 GHz Max Turbo Frequency | |
| Base Clock / Multiplier | 100 MHz BCLK / 25.0× (Multiplier Locked) | |
| Workload Orchestration | Intel® Thread Director (Hardware telemetry feeding OS scheduler) | |
| Multi-CPU Scalability | Single Socket (1S Only) | |
| Thermal & Electrical | Processor Base Power (PL1) | 35 W (T-Series Ultra-Low Power Profile) |
| Maximum Turbo Power (PL2) | 114 W (Configurable up to 65W base in performance mini-PC profiles) | |
| Nominal Voltage Range | Dynamic VID (0.70 V – 1.25 V) | |
| Cache Hierarchy | Level 0 (L0) Cache | P-Core L0-D: 48 KB • P-Core L0-I: 64 KB per core |
| Level 1 (L1) Cache | P-Core: 192 KB total (48K Data + 64K Inst per core) • E-Core: 96 KB total (32K Data + 64K Inst per core) | |
| Level 2 (L2) Cache | P-Core: 3 MB dedicated per core (18 MB) • E-Core: 4 MB shared per 4-core module • 22 MB Total L2 | |
| Level 3 (L3) Cache | 20 MB Intel® Smart Cache (shared pool across all 10 cores on bi-directional ring bus) | |
| Memory Controller | Memory Architecture | Dual-Channel DDR5 interface (2 discrete 64-bit sub-channels) |
| Maximum Memory Speed | Up to DDR5-6400 MT/s | |
| Maximum Memory Bandwidth | 102.4 GB/s (At native DDR5-6400 rating) | |
| Max Capacity / ECC | Up to 256 GB • Non-ECC Unbuffered (UDIMM / SODIMM) | |
| Museum Host Installed RAM | 32 GB (2 × 16 GB) DDR5-5600 SODIMM (Nanya NT16GF64F88AX3S-Q5 + Micron CT16G56C46S5.C8H • CL46 @ 1.1V) | |
| AI & Neural Processing (NPU) | NPU Hardware Architecture | Intel® AI Boost NPU 3 (Dedicated neural coprocessor on SoC Tile) |
| Dedicated NPU Throughput | 13 TOPS (INT8) dedicated inference compute | |
| Total Platform AI Compute | Up to 23 TOPS combined (NPU 13 TOPS + CPU Cores 6 TOPS + Xe GPU 4 TOPS) | |
| AI Software Frameworks | Intel OpenVINO™ Toolkit, DirectML, ONNX Runtime, WebNN | |
| Expansion & Graphics | PCI Express Interface | PCIe 5.0 (20 Lanes: 1×16 + 1×4 or 2×8 + 1×4) + PCIe 4.0 (4 Lanes direct NVMe) |
| Integrated Graphics Solution | Intel® Arc™ Xe-LPG Graphics (16 EUs / 2 Xe-Cores • 300 MHz base / 1800 MHz boost) | |
| Display & Media Engine | DisplayPort 2.1 (UHBR20), HDMI 2.1 • AV1, HEVC, AVC, VP9 hardware encode/decode | |
| PassMark Benchmarks | Average CPU Mark (Multithread) | 25,900 (Rank #1 in Mateo’s Museum CPU Tier List • Median: 26,190 • Local Host Test: 25,911) |
| Single Thread Rating | 4,269 MOps/Sec (131st fastest overall across 6,006 CPUs in PassMark database) | |
| Desktop CPU Category Rank | Rank #220 out of 1,588 Desktop CPUs in PassMark database |
2. Silicon Topology & Arrow Lake Disaggregated Architecture
Modular Compute Scale
17.8B
Transistors across 3D Foveros disaggregated tiles
Physical Execution Engines
10 Cores / 10 Threads
6 Lion Cove P-Cores + 4 Skymont E-Cores (up to 4.90 GHz)
Cache & Edge Neural Compute
20MB L3 / 13 TOPS
Intel Smart Cache ring + Intel AI Boost NPU 3
Internal Silicon Die Topology & Foveros 3D Modular Architecture
+---------------------------------------------------------------------------------------------------+
| INTEL CORE ULTRA 5 225T DISAGGREGATED SILICON TOPOLOGY (ARROW LAKE-S) |
| |
| +---------------------------------------------------------------------------------------------+ |
| | COMPUTE TILE (TSMC 3nm N3B PROCESS) | |
| | | |
| | +-------------+ +-------------+ +-------------+ +-------------+ +-----------------+ | |
| | | P-CORE 0 | | P-CORE 1 | | P-CORE 2 | | P-CORE 3 | | E-CORE CLUSTER | | |
| | | Lion Cove | | Lion Cove | | Lion Cove | | Lion Cove | | (4x Skymont) | | |
| | | 48K+64K L1 | | 48K+64K L1 | | 48K+64K L1 | | 48K+64K L1 | | 32K+64K L1 ea. | | |
| | | 3MB L2 | | 3MB L2 | | 3MB L2 | | 3MB L2 | | 4MB L2 Shared | | |
| | +------+------+ +------+------+ +------+------+ +------+------+ +--------+--------+ | |
| | | | | | | | |
| | === ring stop ======== ring stop ====== ring stop ====== ring stop ======== ring stop ===+ | |
| | || | | | | | || | |
| | || 2.5MB L3 2.5MB L3 2.5MB L3 2.5MB L3 2.5MB L3 || | |
| | || || | |
| | || BI-DIRECTIONAL ULTRA-LOW-LATENCY RING INTERCONNECT (20 MB L3 POOL) || | |
| | || || | |
| | || 2.5MB L3 2.5MB L3 2.5MB L3 2.5MB L3 2.5MB L3 || | |
| | || | | | | | || | |
| | === ring stop ======== ring stop ====== ring stop ====== ring stop ======== ring stop ===+ | |
| | | | | | | | |
| | +------+------+ +------+------+ +------+------+ +------+------+ +--------+--------+ | |
| | | P-CORE 4 | | P-CORE 5 | | D2D FABRIC | | D2D FABRIC | | POWER MGMT UNIT | | |
| | | Lion Cove | | Lion Cove | | INTERCONNECT| | CONTROLLER | | HARDWARE SCHED | | |
| | | 48K+64K L1 | | 48K+64K L1 | | (Die-to-Die)| | (Foveros 3D)| | Thread Director | | |
| | | 3MB L2 | | 3MB L2 | | | | | | | | |
| | +-------------+ +-------------+ +-------------+ +-------------+ +-----------------+ | |
| +---------------------------------------------------------------------------------------------+ |
| || (High-Density Foveros Micro-Bumps) |
| +-----------------------------+ +-------------------------------+ +-------------------------+ |
| | GRAPHICS TILE (TSMC 5nm) | | SoC TILE (TSMC 6nm N6) | | I/O TILE (TSMC 6nm N6) | |
| | - Intel Arc Xe-LPG (16 EUs) | | - Intel AI Boost NPU 3 (13T) | | - PCIe 5.0 (20 Lanes) | |
| | - 2 Xe-Cores @ up to 1.8GHz | | - Dual-Channel DDR5 Memory Ctl| | - PCIe 4.0 (4 Lanes NVMe)| |
| | - DP 2.1 / HDMI 2.1 Engine | | - Security & Power Engine | | - DMI 4.0 x8 to PCH Q870| |
| +-----------------------------+ +-------------------------------+ +-------------------------+ |
| |=================== PASSIVE BASE TILE (INTEL 16 / 22nm FINFET INTERPOSER) ===================| |
+---------------------------------------------------------------------------------------------------+
3. Dual-Channel DDR5 Memory & High-Speed Tile Interconnect
Integrated Dual-Channel DDR5
- Memory Architecture: Dual-channel memory controller integrated directly on the dedicated SoC Tile.
- Rated Speed: Officially rated up to DDR5-6400 MT/s, delivering up to 102.4 GB/s theoretical bandwidth.
- Museum Host Deployment: Equipped with 32 GB (2 × 16 GB) DDR5-5600 SODIMM modules operating at native 1.1V low-power envelope.
- Workload Scaling: High-bandwidth DDR5 architecture feeds both compute threads and Xe-LPG graphics/NPU pipelines simultaneously.
Foveros 3D Die-to-Die Interconnect
- Interconnect Architecture: High-density 36-micron pitch micro-bumps bonding active silicon tiles to the passive Intel 16 base interposer.
- Sub-Nanosecond Latency: Die-to-die (D2D) fabric interconnect ensures near-monolithic communication latency between Compute, SoC, and Graphics tiles.
- Direct PCIe Root Complex: 20 native PCIe 5.0 lanes plus 4 PCIe 4.0 lanes driven directly from CPU silicon to host M.2 slots and expansion buses.
- High-Speed PCH Link: DMI 4.0 x8 interconnect linking the processor to the Intel Q870 Arrow Point chipset at 15.75 GB/s.
4. Advanced Technologies, Virtualization & Platform Security
| Technology Category | Feature / Instruction Set | Functional Capability |
|---|---|---|
| Hardware Virtualization | Intel® VT-x with EPT | Hardware virtualization assistance with Extended Page Tables (EPT) and Mode-based Execute Control for nested virtualization |
| Intel® VT-d (IOMMU) | Directed I/O virtualization enabling full PCI passthrough (NVMe SSDs, network NICs, iGPU) to guest VMs in Proxmox VE 9.2.4 | |
| Intel® Thread Director | Microcontroller-level real-time thread scheduling providing OS runtime guidance to place latency-sensitive tasks on P-cores and background tasks on E-cores | |
| AI & Cryptographic Acceleration | Intel® AI Boost NPU 3 | 13 TOPS dedicated neural coprocessor offloading continuous AI workloads, voice recognition, computer vision, and local LLM embeddings at zero CPU overhead |
| Intel® AES-NI & SHA | Hardware-accelerated AES encryption achieving 18,790 MB/s sustained throughput for line-rate WireGuard VPNs and encrypted ZFS storage pools | |
| AVX2 & FMA3 / DLBoost | Advanced Vector Extensions 2 (256-bit SIMD), Fused Multiply-Add 3, and Intel Deep Learning Boost (VNNI) for high-performance tensor computing | |
| Platform Security & Power | Intel® TXT & XD Bit | Trusted Execution Technology and Execute Disable Bit memory buffer protection providing hardware root-of-trust verification |
| Intel® Turbo Boost 2.0 | Dynamic opportunistic clock scaling boosting P-cores up to 4.90 GHz and E-cores up to 4.40 GHz within thermal and electrical headrooms | |
| Energy-Efficient 35W Profile | T-Series low-power envelope allowing 24/7 homelab operation with minimal power draw and whisper-quiet acoustic footprint |
5. PassMark Benchmark Dossier & Synthetic Performance (PerformanceTest V10)
Multithread Rating (CPU Mark)
25,900
Median: 26,190 across 81 samples (Rank #1 in Mateo’s Museum Collection)
Single Thread Rating
4,269
MOps/Sec • 131st fastest overall CPU in PassMark database
Data Encryption (AES-NI)
18,790
MBytes/Sec sustained hardware cryptographic rate
PassMark PerformanceTest V10 Detailed Test Suite Results
| Test Suite Routine | Average Result Metric | Operational Context |
|---|---|---|
| Integer Math | 60,344 MOps/Sec | General OS kernel execution, database processing, and hypervisor overhead |
| Floating Point Math | 83,747 MOps/Sec | Complex numeric simulation, 3D rendering, and financial computation |
| Find Prime Numbers | 300 Million Primes/Sec | Pure integer execution pipeline and branch prediction efficiency |
| Random String Sorting | 29,810 Thousand Strings/Sec | DDR5 memory controller bandwidth and L3 cache indexing responsiveness |
| Data Encryption | 18,790 MBytes/Sec | Hardware AES-NI encrypted NVMe storage & high-throughput WireGuard/IPsec tunnels |
| Data Compression | 242,836 KBytes/Sec | ZFS LZ4/ZSTD compression and Proxmox backup server (PBS) data deduplication |
| Physics Simulation | 2,026 Frames/Sec | Multi-threaded real-time kinematics and vector transformation throughput |
| Extended Instructions | 21,129 Million Matrices/Sec | Intel AVX2 and FMA3 256-bit matrix vectorization throughput |
| Single Thread Performance | 4,269 MOps/Sec | Single-core Lion Cove responsiveness at up to 4.90 GHz peak turbo clock |
PassMark Relative Hierarchy: Arrow Lake & Compact AI PC Stack
| Processor Model | Cores / TDP | Average CPU Mark | Relative Variance |
|---|---|---|---|
| Intel Core Ultra 5 225F | 10C (6P+4E) / 10T • 65W | 31,013 | +19.7% |
| Intel Core Ultra 5 235T | 14C (6P+8E) / 14T • 35W | 30,805 | +18.9% |
| Intel Core Ultra 5 225 | 10C (6P+4E) / 10T • 65W | 30,479 | +17.7% |
| Intel Core Ultra 5 245T | 14C (6P+8E) / 14T • 35W | 30,132 | +16.3% |
| Intel Core Ultra 5 225T (Exhibit) | 10C (6P+4E) / 10T • 35W | 25,900 | Baseline (0.0%) |
| AMD Ryzen AI 7 PRO 350 | 8C (4P+4c) / 16T • 28W | 24,001 | -7.3% |
| Apple M4 10 Core (Post #2425) | 10C (4P+6E) / 10T • SoC | 23,785 | -8.2% |
| AMD Ryzen AI 7 PRO 360 | 8C (4P+4c) / 16T • 28W | 22,120 | -14.6% |
| AMD Ryzen AI 5 340 | 6C (2P+4c) / 12T • 28W | 19,576 | -24.4% |
6. Platform Synergy: Pairing with HP ProDesk 4 Mini G1i AI Black
Seamless Architectural Integration
The Intel Core Ultra 5 225T is the silicon heart of the HP ProDesk 4 Mini G1i AI Black cataloged in Exhibit #2851. Built on Intel’s Arrow Point Q870 chipset with enterprise vPro support, the compact desktop form factor delivers exceptional compute and edge AI density:
| Hardware Interface | Intel Core Ultra 5 225T | HP ProDesk 4 Mini G1i AI Black (Post #2851) |
|---|---|---|
| CPU Socket & Chipset | Socket LGA 1851 (FC-LGA18W) | Intel® Arrow Point Q870 Chipset with Intel vPro® Enterprise |
| Memory Architecture | Dual-Channel DDR5 (up to 6400 MT/s) | 32 GB DDR5-5600 SDRAM (Dual SODIMM: Nanya + Micron • CL46) |
| High-Speed Storage | PCIe 4.0 ×4 direct CPU NVMe lanes | Dual M.2 2280 PCIe Gen4 x4 NVMe SSD slots (512GB Kingston + 2TB upgrade) |
| Display & Media Engine | Intel Arc Xe-LPG Graphics (DP 2.1 / HDMI 2.1) | Dual DisplayPort™ 2.1 + HDMI 2.1 ports driving multi-monitor 4K/8K displays |
| Edge AI Acceleration | Intel AI Boost NPU 3 (13 TOPS) | Dedicated AI PC form factor for low-power containerized neural models |
| Power & Thermodynamics | 35W Base TDP / 114W PL2 maximum turbo | 90W External power adapter (up to 89% efficiency) in 17.7 × 18.1 × 3.42 cm chassis |
7. Curator’s Assessment & Homelab Applicability
- Museum Hardware Provenance: Mateo’s Museum archive holds 1 verified physical unit of the Intel Core Ultra 5 225T processor (OEM Part:
AT8076806772, S-Spec:SRVFA). The processor is in active operational deployment inside the HP ProDesk 4 Mini G1i AI Black (Model:BY6X9ET, S/N:1HF6181F4F), dual-booting Proxmox VE 9.2.4 and Windows 11 Pro. - Top-Tier Computational Authority: With a multithread PassMark CPU Mark of 25,900 (and local baseline of 25,911), the Core Ultra 5 225T holds the #1 highest ranked compute score across all processors in Mateo’s Museum collection, outscoring both the Apple M4 10-core (23,785) and dual-socket Xeon enterprise servers.
- Edge AI Inference Engine: The onboard 13 TOPS Intel AI Boost NPU 3 introduces dedicated neural acceleration to the museum homelab. Under Linux / Proxmox containerization, the NPU offloads continuous embedding generation, local whisper transcription, and computer vision tasks without consuming CPU or GPU execution cycles.
- Low-Power 35W Hypervisor Density: Operating at a base TDP of just 35 W, the 225T achieves staggering compute-per-watt efficiency. It allows the HP ProDesk 4 Mini G1i to run dense microVM workloads, Docker container suites, and storage services 24/7 with minimal power draw and low thermal output.
- Superb Single-Thread Responsiveness: Achieving a Single Thread Rating of 4,269 MOps/Sec at up to 4.90 GHz on its Lion Cove P-cores, the processor provides blazing database index lookups, instantaneous web server responsiveness, and low latency for interactive homelab services.
- Robust Modern I/O Virtualization (VT-d): Full support for Intel VT-d, PCIe 5.0, and PCIe 4.0 root complexes allows pristine hardware passthrough in Proxmox VE 9.2.4, making it an ideal ultra-compact virtualization powerhouse for homelab environments.