Graviton is a Russian server provider declare The first AI and HPC server using domestic processors. The device can support up to eight computing GPUs to handle artificial intelligence and supercomputer workloads, but the manufacturer doesn’t recommend any specific GPUs (although they are easy to guess), possibly because it is illegal to obtain them during sanctions. Furthermore, it’s unclear whether the machine will be able to achieve competitive performance figures.
According to reports, the Graviton S2124B server is based on two undisclosed 48-core CPUs, running at 2 GHz and equipped with DDR4-3200 memory. Server News. The basic specifications of the processor may indicate Baikal Electronics BE-S1000 Server-grade chip, including 48 Arm Cortex-A75 cores, supports 2-way and 4-way symmetric multiprocessor (SMP) configurations. However, this particular version of the BE-S1000 appears to have a CPU clock frequency lower than 500 MHz its original frequencywhich may be the result of porting its design from TSMC’s 16FFC to a different production node at a different foundry. It is also possible that Baikal will run at reduced clock speeds to increase output or reduce power consumption.
Divide these performance figures by 8 and we can determine which GPUs Graviton plans to install in its S2124B machine. The performance figures for each accelerator – 60 FP64 TFLOPS for supercomputing and 3340 FP8/INT8 TFLOPS/TOPS for artificial intelligence – point to Nvidia’s H100 PCIe. In all cases, we are referring to Tensor core performance. That said, those using the S2124B will have to rely on Nvidia’s CUDA ecosystem. However, peak performance wouldn’t be possible without Nvidia’s support. Additionally, it remains to be seen how much performance these Hopper accelerators can actually deliver, given that the platform is Arm-based and relatively unknown.
In addition to its two CPU and eight GPU accelerators, Graviton’s S2124B can incorporate 12 SATA drives or 12 NVMe U.3 SSDs.
The Graviton S2124B server is currently available for order, and customers are invited to apply for testing opportunities, but its pricing is unknown. Additionally, it’s unclear whether Graviton will be able to supply Nvidia H100 GPUs.
Alexander Filchenkov, director of server and network systems at Graviton, said: “We are proud to consistently provide IT solutions that meet market needs in a timely manner.” “This time, we have successfully developed and manufactured a system for complex computing using domestic processors. critical servers. This product represents an important step in advancing domestic computing technology and will enable our customers to effectively address data processing challenges.”