報告_OCP2026_Intel_ExxonMobil_浸沒式機櫃可靠度_20261015

PDF 原檔：報告_OCP2026_Intel_ExxonMobil_浸沒式機櫃可靠度_20261015_original.pdf

講者公開簡報（OCP 官方 Recorded Content 頁提供下載）。文字以 PyMuPDF 逐頁抽取，### p<n> 為頁碼；圖表內容見原檔。

圖片清單（已驗證 2026-10-19）

原始內容

p1

Samantha Yates Principal Engineer – Intel Corporation Betsy Cossette Immersion Cooling Market Development Manager - ExxonMobil Rack Scale Reliability Demonstration Testing of an Intel Datacenter Certified Immersion Solution Cooling Environments Immersion Project

p2

Intel Datacenter Certified Immersion Solution Tested Enveloped Intel 5th Gen Intel® Xeon® Platinum 8592+ Processor 5th Gen Intel® Xeon® BAC DUG Cool 26RU – 30kW: 2 CDU Configuration Rev 1.5: 16RU, 32RU, 48RU Hypertec Ciara Trident IC610DR-G6 ― Micron 64GB DDR5 4800MT/s RDIMM x 16 16GB, 32GB ExxonMobil EM DC 3235 Super EM DC 3220, EM DC 3250

p3

Intel Datacenter Immersion Certification High Risk Material Incompatibilities Identified? No Yes Certification Report Package Material Compatibility System Material Compatibility Reliability Demonstration Pass Warranty Rider Offered Fail Fluid will not be Certified Package Materials Package Reliability Testing Needed Hardware Characterization System Materials Replacement with Lower Risk Materials Needed Yes Material Compatibility Thermochemical Reliability

p4

Material Compatibility Results Component Result Resolution Intel® XEON® HIGH RISK PASSED: Package Reliability Testing Motherboard PASS Power Supply FAIL REPLACED: Wire Insulation Material CPU1 Heatsink PASS CPU2 Heatsink PASS Power Cable v1 PASS Power Cable v2 PASS Coin Cell PASS Chassis Screws PASS PCIe + Riser PASS Power Delivery Board PASS Chassis Mylar PASS PCH Heatsink PASS Network Heatsink PASS SSD PASS NIC Card PASS Micron RDIMM PASS Package Reliability Result Assembly Test Vehicle PASS Substrate High Speed IO PASS

p5

IRDT: Thermochemical FMEA Thermochemical Reliability • Fluid + Materials + Contaminants • Potential thermochemical interactions • Performance & Reliability Impact Rack-Scale Thermochemical FMEA • Component performance degradation due to secondary reactions with contamination • Film, deposit, or particle formation that results in hardware performance degradation • Fluid property changes that impact hardware performance Risk-Based Certification Plan • Thermal Performance Degradation Assessment • Signal Integrity Degradation Assessment • Memory Degradation • Performance Benchmark Degradation Assessment • Fluid Analysis • Contamination Baseline Data-Driven Certification • Demonstrated Reliability • Statistical Analysis of Performance Degradation • Fluid Stability • Contamination Quantification Fluid + Materials = Contamination Potential Thermochemical Reactions Potential Performance and Reliability Degradation

p6

Fluid Service Use Case: Ideal Batch Reactor Model 𝑃= [𝑅] 𝑒−𝑘𝑡 𝑅՜ 𝑘𝑐𝐼՜ 𝑘𝑖 [𝑃] [𝑃] = 1 −𝑒−𝑘𝑐𝑡[𝑆] Ideal Batch Reactor Assumptions: • System is closed so mass/volume is conserved and constant • System is well mixed and homogeneous Steady State 1st order reaction for a batch reactor: Intermediate Reactions: where k is the reaction rate and t is time. Steady state approximation for intermediate reactions: ts

p7

IRDT: Accelerating to Steady State ln 𝑘= ln 𝐴−𝐸𝑎 𝑅𝑇 Most reactions can be described with the Arrhenius relationship: Process Water Temperature Control Cabinet: • Precise control (± 1°C) of process water to ensure consistent boundary conditions at t0 and ts • Preheats process water to target a higher fluid operating temperature to accelerate achieving steady state U.S Patent No. 12,672,256

p8

IRDT: Data Collection Infrastructure Augmented Temperature Telemetry Tc TCs RDTs at Tank Top/Bottom RDTs on Tank In/Out Data Acquisition and Recording Augmented Networking 3 Switches: Network Traffic and BMC Access Custom Brackets

p9

IRDT: Validation Data Log Files Sensor Data Log Files 8681 3041 Lines of Data 1,090,163,098 419,004,192 Unique Sensors - 35 Thermal Sensors Power Sensors Logs CPU1_DIMMA1_Temp CPU1_DIMMB1_Temp CPU1_DIMMC1_Temp CPU1_DIMMD1_Temp CPU1_DIMME1_Temp CPU1_DIMMF1_Temp CPU1_DIMMG1_Temp CPU1_DIMMH1_Temp CPU1 Temperature CPU2_DIMMA1_Temp CPU2_DIMMB1_Temp CPU2_DIMMC1_Temp CPU2_DIMMD1_Temp CPU2_DIMME1_Temp CPU2_DIMMF1_Temp CPU2_DIMMG1_Temp CPU2_DIMMH1_Temp CPU2 Temperature PCH Temperature +12V +3.3V +3.3VSB +5V +5VSB Power Supply 1 PSU1 Power In PSU1 Power Out +VCCD_HV1 +VCCD_HV2 +VCCFAEHV_FIVRA1 +VCCFAEHV_FIVRA2 +VCCINFAON1 +VCCINFAON2 +VCORE1 +VCORE2 Sensor_Data (CSV) iLVSS (LOG) MUERRCHK (XML) Validation Stress Purpose Memory NUMA aware memory stress test that utilizes all CPU threads at 100%. I/O Auto-configured Storage and Network stress test. Server Management BMC Sensor test and monitoring SEL and logging events. Configuration and Event Monitoring System configuration saved pre-test and verified post-test. PCIe Configuration Check PCIe device information saved for post-test evaluation.

p10

IRDT: Test Flow t0 Performance Reliability Demonstration tS Performance Thermal Performance System Performance Signal Integrity Fluid Sampling Thermal Performance System Performance Signal Integrity Sensor Data Collection Validation Workload Error Log Collection Risk Assessment SI Assessment Performance Assessment MTBF MTTF GSAT Performance GSAT Performance Fluid Analysis Fluid Sampling Hardware Commissioning Thermal Assessment Fluid Analysis

p11

IRDT: Hardware Commissioning Fluid follows through the path of lease resistance. The open tank and chassis let too much fluid flow bypass the server components instead of cooling them. • A custom chassis hanger designed for the certified configuration to prevent fluid bypass.

p12

IRDT: Fluid Analysis ∆ Absorbance = Sample Absorbance284 nm – 1.8097 ts t0 Hardware Commissioning UV-VIS GC-MS DOPT Peaks ICP-MS ts t0 FTIR Fluid Analysis Test t0 ts GC-MS Plasticizers Plasticizers ICP-MS: Bromine 11.20 ± 0.03 ppb 25.09 ± 0.12 ppb Dielectric Constant Er = 2.11 Loss tangent = 0.0004 Er = 2.11 Loss tangent = 0.0004 FT-IR Spectroscopy Matched to Unused Matched to Unused UV-Vis Spectroscopy (Absorbance @ 284 nm) 3.394 3.654 Visual Color/Appearance Matched to unused Matched to unused

p13

IRDT: Dielectric Fluid Comparison Er = 2.09 Reference ts Air t0 Er = 2.09 Reference ts Air t0 Measurement Er Loss Tangent Frequency Air 1 0 10.3 GHz t0 2.11 0.0004 7.1 GHz ts 2.11 0.0004 7.1 GHz

p14

IRDT: Hardware Performance Analysis CPU Thermal Performance Test Deviation Result Linpack -1.0% PASS SPECrate -0.8% PASS GSAT +2.0% PASS Benchmarks Difference between t0 and ts to assess degradation CPU Configuration Bottom of Tank CPU 1 CPU 2 Top of Tank CPU1 Delta Trise average (ts-t0) = 1.6 oC CPU2 Delta Trise average (ts-t0) = 1.2 oC PCIe SSD Timing PCIe NIC Timing DDR UPM Mask = 4.4ps t0 Mean = 8.5ps ts Mean = 8.4ps Mask = 4.4ps t0 Mean = 8.3ps ts Mean = 8.3ps UPM = 0 = Low Risk

p15

Reliability Demonstration Statistics Mean time between failures (mtbf) during a continuous test can be calculated as the total test time t from all tested units divided by the number of failures k. 𝑚𝑡𝑏𝑓= 𝑡 𝑘 If the unit failure rate is assumed to be constant the 2 distribution can be used to calculate confidence intervals around a MTBF, and therefore demonstrated reliability can be extracted. The MTBF LCL for a time truncated test, such as IRDT, is described by 𝑀𝑇𝐵𝐹≤ 2𝑡 𝛸𝛼,2𝑘+2 2 In the case of a nonrepairable error, MTBF becomes mean time to failure MTTF. If the assumption that the failure rate is constant is maintained, the survival probability R(t) for a given service lifetime t can be estimated by the exponential distribution function 𝑅𝑡= 𝑒 −𝑡 𝑀𝑇𝐵𝐹 where the reliability confidence limit is defined by the MTBF lower confidence limit.

p16

Demonstrated Reliability of Hardware in EM DC3235 Unit Count Unit Hours MTBF (80% LCL, hours) MTBF (90% LCL, hours) MTBF (95% LCL, hours) Server 24 63,548 39,471 25,570 21,218 CPU 48 127,096 78,942 55,139 42,436 DIMM 384 1,016,832 631,533 441,114 339,488 Unit Service Lifetime Survival Probability (80% LCL) Survival Probability (90% LCL) Survival Probability (95% LCL) Server 1 yr 80% 71% 66% CPU 1 yr 89% 85% 81% DIMM 1 yr 99% 98% 97%

p17

IRDT Reliability Risk Assessment Test Area Tests Performed Result Package Material Compatibility Viscoelastic Properties: Storage Modulus HIGH RISK Viscoelastic Properties: Tan Delta HIGH RISK Coefficient of Thermal Expansion PASS Shear Adhesion PASS Package Reliability Assembly Test Vehicle Reliability PASS Substrate High Speed IO Reliability PASS System Material Compatibility Component Visual Inspection PASS Fluid Integrity: FT-IR PASS Fluid Integrity: GC-MS PASS Fluid Integrity: ICP-MS PASS iRDT: Immersion Reliability Demonstration Testing Benchmark Performance Degradation PASS Thermal Performance Degradation PASS Signal Integrity: Rank Margin (DDR) PASS Signal Integrity: Lane Margin (PCIe) PASS iLVSS PASS System Errors Info Only Fluid Analysis: Contamination Steady State PASS Fluid Analysis: Contamination Analysis PASS Fluid Analysis: Degradation Analysis PASS Characterization Xeon® N/A System Components PASS Demonstrated Reliability MTBF Info Only ADDL Immersion Certification PASS

p18

• All board components need to be compatible with the fluid ― and the immersion solution needs to be tested to ensure the materials are compatible with each other. • The industry needs to standardize methodologies to measure the reliability of hardware in an immersion solution with statistical confidence limits ― and align on how to share the data. • Contact any partner for additional information on the Intel Datacenter Certified Immersion solution Call To Action

p19

Thank You!