ASRock AM4 Rack X570D4U-2L2T

Product Information

Each Sysbench VM is configured with three vDisks: one for boot (~92GB), one with the pre-built database (~447GB), and the third for the database under test (270GB). From a system resource perspective, we configured each VM with 16 vCPUs, 60GB of DRAM and leveraged the LSI Logic SAS SCSI controller. The above usage scenario still leaves me with one available open-ended PCIe x4 slot for expansion, which is not bad at all. Ok, so this board is not perfect. I will start with what I believe to be a design flaw that I currently suffer through, but that may be a deal breaker for many prospect buyers. There is a fundamental and inherent layout and power delivery design incompatibility: you have three PCIe lanes that fight for space and electronics, and there are no ideal configurations to maximize the potential of this board. It's currently running unRAID booting from the internal USB3 (needed to buy an adapter) with 8 drives in to a SAS controller sitting in the x4 PCIe expansion. unRAID boots to the internal graphics which is nice and I have a R9 Nano for a Windows VM and an old AMD card for Server 2016 VM. I have a USB soundcard in one of the rear USB ports and a small hub in the other for the keyboard and mouse.

Supports PCIe3.0 x16 when using AMD Ryzen™ 5000, 4000 G-Series Desktop Processors with Radeon™ Graphics Looking at performance, we used two SN850 SSDs to get a feel for the peak system capabilities. For our Application Workload Analysis, we did the testing a bit differently using only 2 VMs for SQL Server and then 4 VMs for Sysbench. SQL Server gave us an aggregate average latency of 2ms. For Sysbench we saw aggregate scores of 3,210 TPS, 36.5ms for average latency, and 264.5ms for worst-case scenario latency. For AMD Ryzen Desktop Processors with Radeon Graphics, ECC support is only with Processors with PRO technologies.

Conclusion

Each SQL Server VM is configured with two vDisks: 100GB volume for boot and a 500GB volume for the database and log files. From a system resource perspective, we configured each VM with 16 vCPUs, 60GB of DRAM and leveraged the LSI Logic SAS SCSI controller. When it comes to benchmarking storage arrays, application testing is best, and synthetic testing comes in second place. While not a perfect representation of actual workloads, synthetic tests do help to baseline storage devices with a repeatability factor that makes it easy to do apples-to-apples comparisons between competing solutions. I’ll look into a better LSI HBA option…open to recommendations as I can’t say I am very familiar with their line. That was supposed to read LSI 9207 by the way – sorry for the typo! Or maybe I just use a breakout cable on the second port of that card with SFF-8087 to SATA so I can pass through some SATA SSDs…not sure if there is a way to pass through NVME that way. With the Sysbench OLTP, we recorded an aggregate score of 3,210 TPS across 4 VMs with the individual VMs ranging from 759.3 TPS to 876.6 TPS.

Whilst there seems to be a lot of bad points, I have managed to overcome all of them, with the exception of the RAM which I hope will be fixed with a BIOS update.

Parameters

For basic functionality though, it works well for remote connect into the server, power it on, and install an OS remotely. For the use cases of the platform that will be perfectly fine for most buyers in this class. Out of band management isn’t something typically found in workstations, so even in this limited form, it’s nice to have. So, BMC functionality is there and works pretty well, at least the web management interface. SSH allows only a single session at a time and my system thinks it has an existing active session when it does not. Haven't tried sol. The remote KVM works. I have not tried to install an OS onto the server using the BMC but that is supported by the web interface. Since this is based on Ryzen, it is a single NUMA node design. AMD EPYC 7001 8-core servers, such as those based on the AMD EPYC 7251 had four NUMA nodes which created a lot of inter-die traffic. With Ryzen, one does not have to worry about that on these lower-cost platforms. SLOT5: PCIe4.0 * x1 [FCH] *Supports PCIe3.0 when using AMD Ryzen™ 5000, 4000 and 3000 G-Series Desktop Processors with Radeon™ Graphics

Along the front are four bays for hot-swapping 3.5” or 2.5” SATA drives. This will go a long way for adding capacity while keeping the price down. Assuming you want to max out capacity on this guy, four bays isn’t enough. The good news is that there are three more internal 2.5” bays. These bays are fixed, not swappable, and they too are SATA. This gives users even more options for adding low-cost capacity ideally with 2.5” SSDs or slim HDDs if you must. Now a year later, I discovered after a prolonged power outage that drained my UPS and forced a restart (which completely failed) that the ATX power supply in my server was bad.write saw a much better start in the latency department being as low as 22.2µs, here the sever went on to peak at 311,441 IOPS and a latency of 790µs before a slight drop off. SLOT5: PCIe4.0* x1 [FCH] *Supports PCIe3.0 when using AMD Ryzen™ 5000, 4000 and 3000 G-Series Desktop Processors with Radeon™ Graphics This is a highly revised review. Originally, I was disappointed by this board due to problems that I was having... inconsistent boot experiences with some outright failures, the BMC not functioning (not taking an IP lease and appearing as an unknown device), and unreadable screen resolution and display problems. Quite awful. These workloads offer a range of different testing profiles ranging from “four corners” tests, common database transfer size tests, as well as trace, captures from different VDI environments. All of these tests leverage the common vdBench workload generator, with a scripting engine to automate and capture results over a large compute testing cluster. This allows us to repeat the same workloads across a wide range of storage devices, including flash arrays and individual storage devices.