Recommendations

[smi305]

Budget (including currency): Open but realistic

Country: USA

Intended Use: AI, Machine Learning, LLMs, Development Workstation / Server

Additional Details:

• Current system components:

  1. Motherboard: Gigabyte MZ01-CE0 (rev. 2.x)
  2. CPU: AMD EPYC 7702P - 2 GHz - 64 Cores - 128 Threads
  3. Primary Storage: Crucial P3 1TB PCIe Gen3 3D NAND NVMe M.2 SSD, up to 3500MB/s - CT1000P3SSD8 (1x)
  4. Memory: Samsung 64GB/4Gx4 DDR4-2666 ECC/REG Load Reduced CL19 Server Memory Model M386A8K40BM2-CTD7Q (8x)
  5. Graphics: PNY NVIDIA Quadro M6000 24 GB PCI Express Graphics Card (4x)
  6. Secondary Storage: Crucial BX500 2TB 3D NAND SATA 2.5-Inch Internal SSD, up to 540MB/s - CT2000BX500SSD1 (8x)
  7. Chassis: Thermaltake Tower 500 Vertical Mid-Tower Computer Chassis Supports E-ATX CA-1X1-00M1WN-00
  8. Power Supply: EVGA SuperNOVA 1600 G2 80+ GOLD, 1600W Fully Modular NVIDIA SLI and Crossfire Ready 10-Year Warranty Power Supply 120-G2-1600-X1
  9. Operating System: Ubuntu 22.04 LTS

 

• Cooling Considerations:

  • The current system experiences significant heat spikes in the M2, memory, GPUs, and CPU during heavy workloads.
  • Interest in implementing water cooling for all significant components to enhance thermal management and maintain performance during prolonged, intensive tasks.
  • I recently swapped the CPU cooler from this (Dynatron A38 Threadripper EPYC CPU Heatsink and Fan, sTRX4, TR4, SP3), which works perfectly but the noise levels drove me crazy so I switched to this (Enermax Liqtech TR4 II 240 Addressable RGB AIO CPU Liquid Cooler, AMD Threadripper sTRX4/TR4/SP3, Dual Chamber Pump, 5V 3-PIN ARGB, 500W+ TDP, 5-Yr Warranty (ELC-LTTRTO240-TBP))

 

Cooling Requirements:

• Water cooling solutions for:
  • CPU
  • 4x GPUs
  • 8x DDR4 memory modules
  • 1x M2 (if feasible, given space constraints with GPUs)

 

Components Needed for Water Cooling:

• Water Blocks: Suitable for CPU, GPUs, memory modules, and possibly the M2
• Radiators: Appropriate size and number to effectively dissipate heat
• Pumps: High-performance pump(s) to maintain coolant flow
• Fans: High airflow fans for radiator cooling
• Tubing: Durable tubing for efficient coolant circulation
• Fittings: Compatible fittings for secure and leak-free connections
• Coolant: High-quality coolant suitable for your system

 

Objective:

• Achieve efficient cooling to handle heavy workloads that run for extended periods (days to weeks) without thermal throttling or performance degradation.
• Use water cooling - prefer if we could use separate "loops" - one for CPU, one for memory, one for GPUs so that I ensure we get the coldest temps possible without completely overkilling it with Nitrogen or something nutty

 

Thoughts?

• I'm looking for a detailed list of all necessary components and any recommendations or advice on setting up this water cooling system to ensure optimal performance and reliability.

 

End Note

• I attached a screen shot of System Monitor in Ubuntu during average workloads I experience - granted that worst case all memory and CPU (for the most part) will be maxed out)

[OddOod]

The M6000 is nearly a decade old. You are clearly playing with big money, I'd recommend grabbing some pre-blocked 3080Tis (EVGA hydro is bomb) or, if VRAM is the goal, prebolcked 4090s. 
Alternatively, with this much heat, is it possible to rebuild it in a rack chassis?