Results

Project results

The following objectives of the project have been addressed by applications and use cases:

  • Energy efficiency: increase the energy efficiency by 2-5x compared with current available HPC solutions based on CPU or CPU+GPU thanks to:
    1. New energy-efficient processors and accelerators
    2. Integrated use of FPGAs
    3. Bi-phase cooling
    4. Mixed-precision and data compression techniques
    5. Direct communication between FPGAs
  • Sustained application performance: increase in performance by 2-10x compared with current solutions based on CPU or CPU+GPU thanks to:
    1. new AI accelerators and high-performance GPUs and FPGAs
    2. HW fast schedulers
    3. low-latency inter-/intra-node links
    4. better cooling allowing higher frequencies
    5. fast flow enabled at programming mode and run time levels
    6. Optimization of applications
  • Seamless integration of reconfigurable accelerators
    • TEXTAROSSA vendors will adopt tools and solutions of the project.
  • Integrated Development Platforms (IDVs)
  • Development of new IPs
    1. IDV-A – GPU
    2. IDV-E – Reconfigurable technology
  • Fine-tuned thermal policies integrated with an innovative cooling technology
    • Positioning in the HPC market of the integrated cooling technology.
  • Deployment of two heterogenous nodes featuring cutting edge hardware and refrigerated by the innovative two-phase cooling system
    1. IDV-E: ARM + FPGA (U280): 950W TDP
    2. IDV-A: Intel Xeon + NVIDIA H100: Open Sequana – 3.5 KW TDP
IDV-E with two-phase cooling
IDV-A with two-phase cooling
Seamless integration of reconfigurable accelerators
Goal: integration of FPGA-based accelerators
Means: VitisHLS extensions: APEIRON, OmpSs@FPGA, Fastflow, TAFFO
Achievements:

  • APEIRON: validated on IDV-E + RAIDER and FIPLib
  • OmpSs@FPGA: validated on IDV-E
  • Fastflow: integrates deployment of kernels developed with VitisHLS
  • TAFFO: developed support for VitisHLS to perform mixed precision computing in generated accelerators
Fine-tuned thermal policies integrated with an innovative cooling technology
Goal: validated hierarchical thermal control
Means: 2-phase cooling on IDVs, thermal control policies
Achievements:

  • TRL 7 validation of InQuattro 2-phase cooling technology
  • Commercial product from E4 based on X86+GPU+ 2-phase cooling
  • DVFS inner control loop + coolant flow rate outer control loop
  • Temperature within 1° of target set point for GPU
  • Temperature within 2° of target set point for 104 core

For further details please check the deliverable D6.3.