The topic I armed spare PCIe slots with cheap NVMe adapters, and they were the best… is currently the subject of lively discussion — readers and analysts are keeping a close eye on developments.
This is taking place in a dynamic environment: companies’ decisions and competitors’ reactions can quickly change the picture.
If you’re even remotely familiar with PC components, you might’ve noticed a bunch of PCIe (or Peripheral Component Interconnect express) slots on the motherboard. For most folks, only the largest among these slots is useful, as that’s where you’re supposed to plug your bulky graphics card. That said, bringing extra processing horsepower to a gaming rig isn’t the only use case where these expansion slots come in handy.
for example, I’ve got a PCIe network card that not only circumvented the faulty Ethernet controller on my motherboard, but it also enhanced the file transfer speeds on my LAN. But aside from this somewhat expensive investment, I’ve also installed a couple of PCIe port expanders on my home lab and general-purpose devices, and my PCIe-to-NVMe adapters were the most convenience-enhancing PC upgrade I’ve ever invested in.
Call me a cheapskate if you must, but I mostly go for cheap motherboards when building a new PC. Even then, I prioritize my budget on good-quality VRMs, easy BIOS flashback functionality, and other QoL aspects for the mobo. This means advanced features like triple M.2 slots remain out of my reach, as I’d have to shell out an extra $100–$200 for better-quality motherboards – money that I could instead allocate to my GPU budget.
Unfortunately, budget motherboards tend to cheap out on the M.2 connectivity front, and I’m not just talking about manufacturers cutting costs by restricting these slots to PCIe Gen 4 speeds, either. One of my AM4 motherboards has a single M.2 slot, while the one I use for my workstation features two of them. It’s not a deal-breaker, but considering all the LLM-hosting, gaming, and video-editing tasks I subject it to, I’d prefer to use fast NVMe drives over their (comparatively slower) SATA counterparts.
That’s before I even talk about my server-centric systems. I’ve bought a handful of recycled systems, including full-on Xeon mobos, and the dearth of M.2 slots on these otherwise gold mine server rigs is a real bummer. Heck, my dual-Xeon workstation only has one M.2 port where I can plug NVMe drives, which is far from enough to house the all the virtual guests I want to run on this behemoth of a system.
Considering that SSDs tend to slow down when they’re nearly full, I prefer to keep at least 15-20% free space on my daily driver. With Windows 11 already hogging dozens of GBs and my productivity tools (and more importantly, video-editing projects) taking up nearly as much free space as Microsoft’s bloated OS, I had to shift my games from the C: drive and to another NVMe drive I’ve added to the second M.2 slot on my rig.
If you’ve seen the ginormous file sizes on modern PC titles, it shouldn’t come as a surprise that just a handful of triple-A titles and texture mods are all I can keep on my secondary SSD without causing the free space to hit the red zone. But the real problem lies in my AI models.
Some of the larger models can easily hog dozens of GBs, and ever since I started experimenting with different quantization rates, the amount of space these clankers can end up occupying is insane. Then you have the image generation (or rather, upscaling) models I use for my Krita-based photo editing tasks. Let’s just say that the situation got so bad that, at one point, WizTree depicted the root folder housing my AI files as the single largest blob on its space utilization graph.
Since SATA SSDs would be somewhat slow for loading LLMs, I wanted to use NVMe drives at all costs. Fortunately, a cheap $15 adapter board and a spare PCIe slot were all I needed to hook my primary workstation with high-speed NVMe drives for my AI models.
A storage card may not seem all that impressive, but it’s a great investment for your server
Unlike my daily driver, where I’ve only plugged a single NVMe adapter to its PCIe socket, I’ve already armed my Xeon workstation with two of them. Now, I still keep my archived media on full-fledged hard drives, but things are radically different when it comes to my dev VMs, productivity containers, and random virtual guests I use for experiments.
Relying on slow HDDs for my server projects would reduce the responsiveness on my virtual guests, and I’d have to deal with slow startup speeds when spinning up new VMs for my DevOps experiments. Since I’ve already maxed out my SATA drive bays on my home server, bringing extra NVMe SSDs into the fray via PCIe adapters helped me accommodate all my virtual guests on my Xeon warmachine without overprovisioning the storage resources.