Recommended Server Hardware
For servers and cloud setups running Row64.
Compatible Platforms
Row64 runs on many platforms, but the underlying hardware impacts the performance.
Recommended RAM
The recommended amount of RAM depends on your target for the maximum number of records. Note that faster RAM speeds will improve dashboard loading times.
If Row64 runs out of RAM, it will move infrequently accessed storage to your system's mass storage (i.e., SSD).
Recommended CPU
Row64 Server works with x86-64 and ARM architectures, either as local hardware or in virtual machines. It is optimized for data center, server, and desktop CPUs.
CPU Multithreading
By default, Row64 Server is multithreaded in several ways:
More details on Row64's multithreading can be found at the following documentation:
Optimization Overview
Recommended Number of Cores
- Minimum: 1 core
- Recommended: 8 cores
- Optimal: 16+ cores
GPU Display
By default, the Row64 platform uses the GPU for display. GPU display is not required on the server, but many of the diagnostic tools do use GPU display, however, including the following:
-
ByteStream Viewer: Diagnostic viewer for Row64 server files
-
Benchmark Analysis: Studio can be used on the server to analyze Row64 Bench logs
More details on GPU display can be found at the following document:
Optimization Overview
GPU Compute
Row64 Server has many GPU Compute features which are turned off by default. The optimal way to use GPU Compute is to benchmark your hardware and determine the exact crossover points where GPU Compute becomes faster. Due to the latency of the PCIe bus, smaller recordsets tend to run faster on the CPU, so you will want to calibrate to the exact crossover point.
More details can be found at the following document:
Optimization Overview
Recommended GPU
Row64 Server uses Vulkan GPU Compute for low-level speed and cross-hardware compatibility.
We have discovered issues with AMD drivers on Linux, so we recommend Nvidia GPU hardware for Compute in V3.5. For GPU display, however, both Nvidia and AMD work well.
GPU Critical Factor - Generation
From our past testing of GPU Compute with dashboard workloads, we recommend basing your choice primarily on GPU architecture generation (also referred to as the GPU's series).
We conducted a study where we compared 20 series and 30 series cards running Row64 GPU Compute algorithms. In the following diagram, notice how the generation of the card (the first number) had a much larger impact, and teh performance level of the card (the second number) had a minimal impact.
We also tested a $5,000 data center card - the A6000. Surprisingly, it also followed this rule. Since it was an Ampere generation card, the performance was comparable to a 3090 or 3070.
Mass Storage
Higher SSD speed is ideal. Hard disks are not recommended.
When Row64 runs out of RAM, it wil load dashboard assets with less use to the SSD.
The recommended minimum storage is 1TB.
Data Center CPUs and GPUs
Row64 is optimized for data center CPUs and GPUs. The performance will correspond with the speed of the underlying CPU and GPU hardware.
Server Ideas
Data Center CPUs
The AMD EPYC 9965 CPU with 192 cores was tested to be the best CPU for multithreading when it was released in Q4 of 2024. This processor is considered to be a great leap in performance and power efficiency.
In light of this, we anticipate completing more benchmarking and analysis with this CPU. In V3.5, Row64 is pushing towards breaking records with multithreading.
Chips like this are generally behind the scenes running in the cloud. For high-performance workloads, you might consider tuning cloud specifications to the CPU.
Office Server CPUs
We have conducted many performance tests and demos on the AMD 7950X. We found that it is fast, and the 144MB L3 cache appears to accelerate dashboard performance, since it loads more data out of L3 instead of RAM. The 7950XD version has more L3 cache, which could be beneficial for certain large-scale workloads.
The 9950X(D) 16-core processor is a more recent version, and depending on the workload requirements, may be needed for certain large-scale workloads.
We plan on doing additional research on this topic in the near future.
Server Testing
Portable VM
Laptops with integrated GPUs are fast enough that they can run Studio on the desktop and Server through VirtualBox or WSL 2.0 simultaneously. This works well, and is an easy way to test your server and cloud setup while only managing a single laptop.
The best setup we've tested is an ASUS Vivobook 16" laptop with an AMD Ryzen 9 8945HS CPU. It costs under $1,000 and weighs 3.31 lbs.
More details on setting up VirtualBox can be found at the following document:
VirtualBox Setup
More details on setting up WSL 2.0 can be found at the following document:
WSL 2.0 Setup
Mini PC
For those who want to install Linux from USB and have a bare-metal server, a mini PC is a portable, fast, and inexpensive solution.
A model we often use is the Minisforum Venus UM790 Pro Mini PC with an AMD Ryzen 9 7940HS CPU. These work well with cheap, portable monitors, and have better typing speeds than VMs.
Portable Network
For server work from multiple locations, a travel router can be a good solution if your Studio and Server machines need to connect.
The best setup we've found is the GL.iNet GL-AXT1800 (Slate AX) Portable Gigabit Travel Router. It's excellent for working with the entire Row64 platform from multiple PCs when traveling.
Optional Workflow: VM Staging
Working from a local VM can simplify dashboard creation, allowing users to create many test projects and versions of dashboards before publishing. Both VirtualBox and WSL 2.0 work well with this type of setup.