World's First 10x TITAN RTX, 2080 Ti Liquid Cooled GPU Server. Benchmarks. Deep learning/AI, render

10x TITAN RTX GPU liquid cooled server vs. air cooled server.

In this video we are going to compare the performance, temperature and noise level of 8-10 GPU air-cooled server vs custom liquid-cooled server by BIZON.
We run deep learning benchmarks, 3D rendering benchmarks (Luxmark).

BIZON Z9000 liquid-cooled server
More details https://bizon-tech.com/bizon-z9000.html
World's first 8-10 NVIDIA TITAN RTX, 2080 Ti GPU deep learning server.
Features:
– Preinstalled TensorFlow, Keras, PyTorch, Caffe, Caffe 2, Theano, CUDA, and cuDNN.
– Custom liquid cooling system (Dual CPUs and 8-10 x GPUs)
– Up to 3X times higher performance vs. air-cooled TITAN RTXs (no overheating).
– Low temperatures (50C at 100% load).
– Low noise level (50db at 100% load).
– 24/7/365 operation at 100% load.
– Dual 12-Core 2.10 GHz Intel Xeon Silver 4116 (up to Dual 28-Core 2.80 GHz).
– Up to 10 x NVIDIA TITAN RTX or RTX 2080 Ti.
– Up to 5 x NVLinks (up to +100% performance increase)
– ECC Registered DDR4 2666 MHz Memory (up to 768 GB).
– PCIe SSD or HDDs for storage.
– Rackmountable server chassis with redundant power supply.

Liquid-cooled test bench:
BIZON Z9000 rackmount server
GPUs: 4 x NVIDIA TITAN RTX + 4 x RTX 2080 Ti.
4 x TITANs will be active during the benchmarks to match the air-cooled test bench.

Air-cooled test bench:
Supermicro 4029GP-TRT2 rackmount server
GPUs: 4 x NVIDIA TITAN RTX.

_____________________________________

BIZON Z5000
4 GPU liquid-cooled deep learning desktop.
More details: https://bizon-tech.com/bizon-z5000.html
Deep Learning DIGITS DevBox 2018 2019 Alternative
Features:
– Preinstalled TensorFlow, Keras, PyTorch, Caffe, Caffe 2, Theano, CUDA, and cuDNN.
– 8-Core 3.60 GHz Intel Core-X (Latest generation Skylake X; up to 18 Cores).
– Up to 4 x NVIDIA RTX 2080 Ti, Titan RTX, Titan V, Quadro GV100
– FULL CUSTOM WATER COOLING FOR CPU AND GPU. Whisper-quiet (40db).
– DDR4 3000 MHz Memory (up to 128 GB).
– PCIe SSD and additional HDDs for storage.

Deep learning benchmark specs:
Model: Resnet 50
Tensorflow 1.13
Data: Synthetic
1000 iterations
Batch size: 64
FP32