Servers used in solving problems of Linear Algebra and machine learning (inference and forecasting) for recurrent neural networks must have high fault tolerance, wide functionality and good performance.
All these requirements are met by the Supermicro SYS-1029GQ-TVRT - a 1U form factor server equipped with two Intel Xeon Scalable processors, an X11DGQ motherboard, 2000W (Titanium) power supplies and up to 2 hot-swappable SATA.
Supermicro SYS-1029GQ-TVRT appearance
Overview of the SUPERMICRO SYS-1029GQ-TVRT
The server is equipped with two 1st and 2nd generation Intel Xeon Scalable processors, which are represented by a wide range of models, from Bronze to Platinum. Each of these processors is equipped with six DDR4 DIMM slots, which allows support for Optane DCPMM technology (Optane PMem 100) with a capacity of up to 512GB, which, together with DRAM, can increase the total amount of system memory up to 6TB. Another advantage of Intel Xeon processors is the Intel Deep Learning Boost technology with vector neural network instructions, which significantly increases the performance of artificial intelligence inference.
The X11DGQ motherboard has a unique feature - it supports processors with TDP up to 205W; at the same time, for the platform itself, the TDP of supported processors ranges from 70 to 165 Watts. It means that some high frequency Intel Gold series processors (6144, 6146, 6244, 6246, 6250 and 6256) are not compatible with this motherboard. The X11DGQ provides internal storage and I/O sections, as well as front-panel SATA connection and a RAID controller connection directly to the motherboard.
Supermicro PSU, Titanium, 2000W, hot swap PWS-2K05A-1R, used in the presented server, is high-performance and safe. Efficient cooling system, compact size and hot-swap capability makes this device an ideal accessory for Supermicro.
Installed in a SATA 3 (6 Gb/s) server with RAID 0, 1, 5, 10 provides higher frequency and latency minimization, shows improved device power management and faster data transfer when working with SSD (not suitable for SATA 3 hard drives).
The front and rear panels of the device have a fairly simple and understandable principle of operation.
The front and rear panels of the device have a fairly simple and understandable principle of operation.
On the front panel, on the right side, there is a Power Button - supplying / disconnecting power to the server (when the main power source is turned on, the backup is saved) and the UID Button button (turning on the LEDs). LED signals mean:
- red - overheating;
- flashing red (1Hz) - problems in the operation of cooling systems;
- flashing red (0.25Hz) - disturbances in the power system;
- blinking blue - activation of UID from IPMI for visual detection of the server in the rack;
- NIC1 LED - network activity for 1 port;
- NIC2 LED - network activity for 2 ports;
- HDD LED - hard drives are active;
- Power LED - power supply.
The front panel also has two 2.5-inch slots for SATA3/SAS3 (accessories are required to install SAS drives). The IPMI password is located on the server tag above the disk slots.
The back panel Supermicro SYS-1029GQ-TVRT
On the back panel are:
- rear I/O ports - VGA port, Dedicated IPMI LAN, USB0 (3.0), USB1 (3.0), USB2 (3.0), USB3 (3.0), LAN2 (10GbE), LAN1 (10GbE), Unit Identifier Switch (UID);
- low profile PCI-Express 3.0 x8 LP expansion slot (for CPU2) and two full height PCI-Express 3.0 x16 FHHL expansion slots (for CPU1 and CPU2);
- two power supplies: PWS-706P-1R (1U 750W redundant power width 54.5mm (Platinum), HF, RoHS/REACH, PBF).
Specifications Supermicro SYS-1029GQ-TVRT
Specifications Supermicro SYS-1029GQ-TVRT are listed below:
| Functionality | Technical details |
| Form Factor | 1U rack server |
| CPU | 1st-2nd generation Intel® Xeon® Scalable processor |
| Number of processor cores | Up to 28 |
| Video cards and accelerators |
Up to 4 NVIDIA Tesla V100 32GB SXM2 Volta,
Up to 4x NVIDIA Tesla P100 16GB SXM2 Pascal |
| Memory | Up to 6 TB (12 DIMM slots, DDR4-2933 MHz, Intel Optane DCPMM support |
| Chipset | Intel® C621 |
| System BIOS | AMI 32Mb SPI Flash ROM |
| RAID Controllers | SATA3 (6Gbps) with RAID 0, 1, 5, 10M |
| Expansion slots | 4 PCI-E 3.0 x16 slots |
| Remote Administration | Supermicro Intelligent Management Remote Administration Module |
| Network connectivity | Intel X540 Dual Port 10GBASE-T |
| Hard drives | Up to 2 SAS/SATA hot-swappable drives |
| Power supply | included 2 Supermicro 2000W (Titanium) power supplies |
| Dimensions | Height 1.7" (43 mm); Width 17.2" (437 mm); Depth 35.2" (894 mm) or 39.3" (997 mm) with rails. |
Software management is through Intel Node Manager, IPMI 2.0, KVM with dedicated LAN, NMI, SSM, SPM, SUM, SuperDoctor5, and Watch Dog. ACPI/APM is responsible for power management.
Advantages of the SUPERMICRO SYS 1029GQ
To accelerate machine learning, graphics, and data science on Supermicro SYS-based systems, these servers use Tensor Core and CUDA GPUs, the Tesla P100 and Tesla V100.
NVIDIA TESLA P100 SXM2
The P100 accelerator is the most often used in creating remote workstations that work with resource-intensive applications (for example, for 3D graphics or video transcoding).
NVIDIA TESLA P100 SXM2
The minimum clock frequency of this card is 1328 MHz, and when switching to turbo mode, it can be overclocked to 1480 MHz. RAM is 16 GB at 1408 MHz, bandwidth speed is 720.9 GB/s. This card houses 15.300 million transistors (3584 CUDA cores) to support 16-nm process technology and 10.609 gflops of floating-point operations.
NVIDIA TESLA V100 SXM2
Tesla V100 belongs to the next generation of GPUs, so the performance of these graphics accelerators has become much higher, reaching 15.7 teraflops in single precision operations.
NVIDIA TESLA V100 SXM2
This is due to the transition to a 12-nm process technology, an increase in the number of transistors to 21 billion and the number of CUDA cores to 5120. Tesla V100 is used in training neural networks and in high-performance computing (HPC) in various fields of AI. At the same time, due to greater productivity, these processes are carried out much faster and more efficiently.
Thus, Tesla V100 shows the best results in solving problems of Linear Algebra and machine learning (inference and forecasting) for recurrent neural networks with long short-term memory. This is due to the greater adaptability of tensor cores to mixed-precision matrix block multiplication, as well as better magnification of the Tesla V100 hidden layers and reduced computational accuracy. In all cases, testing was carried out using a single card, but, according to the developers, combining several GPUs will give maximum acceleration results.
NVLink-optimized servers provide the best performance and flexible scaling for data centers running HPC and deep machine learning applications.
Who is Supermicro SYS suitable for: scope
The Supermicro SuperServer 1029GQ-TVRT is optimal for solving Linear Algebra and machine learning (inference and forecasting) problems for recurrent neural networks.
If you need to purchase the presented equipment, please contact the specialists of our company. We will select the required configuration of the Supermicro SuperServer SYS-1029GQ-TVRT server for any task.
