Company News

Blackcore Review from Mike Yu of MEtech

We talk to Mike Yu of MEtech to find out why he thinks Blackcore's HFT servers are some of the best you'll find anywhere in the world.

Chloe Vidaurri

- 2 min read

We sat down with Mike, CTO of MEtech, to talk about Blackcore and why they will only use our systems for their clients. MEtech has been a long-term partner of Blackcore as they understand the HFT server market intimately and immediately saw the benefits that our systems offer.

The importance of cooling in HFT servers

As Mike explains, "To achieve the lowest latency, you may set up a CPU spike (busy loop) to allow you to get a fast speed and lowest latency to process the trading application. However, when a CPU spike has been configured in the system, the CPU temperature can increase significantly which may cause system jitter or burst as well as system crash if not cooled correctly."

Blackcore's bespoke liquid-cooling system delivers this stability and consistent performance across all cores. We know that this, coupled with our rigorous validation procedure for each system, delivers performance that is not guaranteed by others in the industry. 

Ensuring system speed across all cores

Mike goes on to highlight that, "Blackcore systems must pass extensive Prime95 and Linux stress tests with 100% CPU utilization across all CPU cores. Such as the 18core 10980XE system which can provide all 18 cores at 5.0GHz frequency without any instability. Other vendors may market things to show you that their 10980XE can reach speeds up to ~5GHz but it depends on the workload or not all cores to 5GHz. Blackcore can 100% promise their clients that their system speed is across all cores."

Why CPU and cache speeds are paramount

MEtech and its clients have been experiencing huge success with our new 5.4GHz Comet Lake systems, due to their unrivalled CPU and cache speeds. "CPU and cache speed is the key point for performance measure. As we know that CPU contains L1, L2, and L3 cache memory to allow the application to improve the hit rate without system memory access to speed-up your application performance. The Blackcore Comet Lake can achieve the highest CPU cache frequency of 5.0GHz now. In addition, AVX is another key point if you have the floating calculation for algorithmic trading, it can be improved by the huge speed."

An often overlooked part, critical to HFT server performance

Another key performance improvement comes from the RAM subsystem of Blackcore systems. An often-overlooked part of the puzzle that Blackcore and MEtech understand is vitally important. "The lowest latency memory architecture for Blackcore is a fantastic design. As we know that CPU cache is not always enough, for example, the Comet Lake only has L1 32K / L2 256k per core and L3 20M shared. If your application overuses these, you will hit system memory which means memory speed is very important. Blackcore now can offer the lowest latency memory access speed by DDR4-3800Mhz with CL15 at 7.8ns access speed that is the fastest one in the world at the moment."

Get in touch today to see how we can enable reliable overclocking performance for you.

About MEtech:

90% of MEtech’s customers selected our performance optimization service because MEtech Global not only sells trading hardware but also offers performance optimization services to allow our customers to have the best advantage for their trading box.

- Operation System Installation (Redhat/CentOS).

- Optimization Linux Kernel parameter and harden, ex, TCP, NIC, hardware IRQ.

- Advanced BIOS tuning for low latency system purpose.

- Customized proprietary lowest latency Linux kernel is to remove unnecessary kernel modules to get better performance and reduce system jitter.

- Application and hardware configuration match to achieve the best performance

MEtech Global also can offer trading services:

-Infrastructure Design

-Performance Optimization

-IT Outsourcing and Managed Services

-Low Latency Connectivity

-Financial Market Data

-Exchange Co-location

 

Chloe Vidaurri

- 2 min read

We use cookies, review our privacy policy here.