Testbed v2 Updates

As stated in my SSD Capacity Scaling Across 5 SSD Models piece, that was my final review on my original testbed. Before discussing how I arrived at the current platform I’ll simply lay it out in table form:

I’ve had a lot of internal back and forth about what form my next generation testbed should take. I debated buying a 5000-series AMD CPU, but due to supply shortages and cost I opted to just stick with the 3700X I had on-hand. Also, the MSI MAG X570 TOMAHAWK WIFI comes with some M.2 heatspreaders. They really don’t have much in the way of added surface area or mass, but I’m going to use them anyway.

A primary consideration was a newer platform to gain access to PCIe 4.0 drives. I settled on the X570 platform quite a while ago so that I’d potentially be able to have 2x M.2 slots using PCIe 4.0, despite one of those slots being connected to the chipset on most motherboards. The idea was that I’d be able to use a fast PCIe 4.0 drive as the source disk for any disk-to-disk copy operations, though this approach has some drawbacks. First of all, I’d be limited to the maximum performance of PCIe 4.0 drives available right now. Between the Samsung 980 Pro and the Western Digital SN850 there are certainly some capable drives, but upcoming controllers might soon outpace the existing flagship consumer SSDs.

Even waving away this concern, this would mean that I’d need to buy at least two flagship drives, one to live in the testbed and the other to perform tests on. I simply can’t justify spending that much money on this project right now. Additionally, I might find that the 2nd M.2 slot routed through the chipset might sufficiently bottleneck performance for my purposes.

At least for now, I’m dismissing the idea of using 2x M.2 SSDs for my future tests. However, that doesn’t mean that I don’t have a plan for minimizing bottlenecks for some of the tests. For one, although the source SSD isn’t particularly fast, it was quite handicapped in the disk-to-disk copy tests in Testbed v1.0 by the PCIe 2.0 x2 link. On X570 it’ll have access to the full PCIe 3.0 x4 bandwidth the drive is capable of, though the 128GB SK Hynix BC501A definitely doesn’t saturate the bus.

As the keen among you may have noticed, this testbed has 64GB of RAM. That’s significantly more than would be needed for a simple SSD testbench. As an added bonus, during my retests, which I am dubbing “Methodology v1.1” I’m making a minor modification. In addition to removing the performance bottlenecks on the PCIe bus for the source disk during the copy test, I’ll be running the copy tests with a RAMdisk as a source. This will better illustrate drive performance for the sample datasets without having to worry about the performance of the originating media.

In short, Methodology v1.1 will simply be Methodology v1.0 on Testbed v2.0, with the addition of a disk-to-disk copy test performed from a RAMdisk to the drive being tested.

It is my plan to retest each of these drives on the updated testbed and then compare the results to the previous results from the old platform. This will set up a baseline for what kinds of performance differences to expect from the updated platform.

I hope to release a preview of what this will look like ahead of a roundup article containing the results of all previously tested drives.

After completing this effort I’ll work on a more thoroughly revamped set of tests, retest everything again, and then look to acquire and test additional drives.