{"id":20210,"date":"2026-07-02T00:45:46","date_gmt":"2026-07-01T23:45:46","guid":{"rendered":"https:\/\/innovatenews.site\/index.php\/2026\/07\/02\/i-used-an-old-ssd-as-scratch-storage-and-it-saved-my-main-drive-from-abuse\/"},"modified":"2026-07-02T00:45:49","modified_gmt":"2026-07-01T23:45:49","slug":"i-used-an-old-ssd-as-scratch-storage-and-it-saved-my-main-drive-from-abuse","status":"publish","type":"post","link":"https:\/\/innovatenews.site\/index.php\/2026\/07\/02\/i-used-an-old-ssd-as-scratch-storage-and-it-saved-my-main-drive-from-abuse\/","title":{"rendered":"I used an old SSD as scratch storage, and it saved my main drive from abuse"},"content":{"rendered":"
\n

The topic I used an old SSD as scratch storage, and it saved my main drive from abuse<\/strong> is currently the subject of lively discussion \u2014 readers and analysts are keeping a close eye on developments.<\/p>\n

This is taking place in a dynamic environment: companies\u2019 decisions and competitors\u2019 reactions can quickly change the picture.<\/p>\n

The last six months have shown that no component that goes into a PC is ever too safe from random price hikes. Unfortunately, it looks like flash storage is next in line. NAND prices have gone up nearly four times over the last nine months, so naturally, everyone’s looking at their old hardware a little differently than they used to before.<\/p>\n

This also means the price of replacing the components that you already own is also prohibitively high, and therefore, you must make all reasonable adjustments to make them last longer than the current inflation cycle. When it comes to NAND-based storage, the solution is quite simple, provided you have at least two SSDs. Here’s how I am protecting my storage from the wrath of my own intensive workflow.<\/p>\n

\"\"<\/figure>\n

One of the lesser-talked about problems of running a local AI workload on your main PC is feeding into the cycle of constantly writing to your SSD. Every model you pull through Ollama gets written to the SSD in full, which means that whenever you download an 8GB or 24GB model, it amounts to 8GB or 24GB if writes.<\/p>\n

If you experiment with as many models as I do, swapping between Qwen3-Coder, Gemma 4 and its various iterations along with a few others, that “write” activity can accumulate rather quickly. To make things worse, when a model and its context window exceed the available system memory (which, in my case, is just 16GB+32GB), Windows starts swapping to disk, and that swapping lands on the primary drive as sustained activity. This is known as virtual memory swapping. This means that, when a large model with a massive context window exhausts the 48GB pool, Windows has no choice but to dump the data into the Windows pagefile on the primary SSD.<\/p>\n

I first noticed this issue when I ran CrystalDiskInfo and found that I had written close to 1.9 Terabytes in three months, and my main NVMe had been absorbing all of it. At this rate, my 600TBW drive could potentially reach its rated endurance much sooner than I had anticipated. The obvious solution was to pull out my old 250GB Crucial SATA SSD and put it to work.<\/p>\n

A lot of memory-hungry programs can end up using your PC’s storage drive as a scratch disk to store temporary files when they run out of system memory, and that’s especially true for RAM-constrained systems. What most users aren’t aware of is the fact that the location of the temporary files can be specified and re-mapped, even to another drive on your system.<\/p>\n

Scratch disks were popular a few years ago, and the idea made a lot of sense as a way to improve performance on your main drive. The usual advice was to buy a dedicated scratch SSD, but the idea makes most economic sense if you already have an SSD that’s sat doing nothing, since the drive is already paid for and probably has enough capacity for the workloads that it needs to handle. The drive endurance gets spent on pagefile activity, browser cache, and other files that are disposable by design.<\/p>\n

\"\"<\/figure>\n

When it came to migration, the Ollama model directory was obviously the first move, followed immediately by the pagefile, since that was the biggest offender when it came to write activity. Windows handles pagefile relocation and makes it remarkably simple if you know where to look. All you have to do is change the virtual memory from your boot drive (where your OS lives) to the secondary drive.<\/p>\n

Now, to be clear, these steps solve two very different problems. Shifting the Ollama directory reclaims the massive amounts of static storage space on my primary drive by offloading those model files, whereas relocating the pagefile is purely about protecting the SSD from the debilitating high-volume write cycles caused by memory overflow.<\/p>\n

The uses of a dedicated scratch disk aren’t exactly limited to local AI inference either. It’s possible to move browser cache for Chrome, shader cache, and other temporary files created by Adobe suite such as Photoshop or Lightroom.<\/p>\n

Keeping your OS and games on separate SSDs might actually make sense<\/p>\n

There’s a certain irony in the fact that a half-decade old SATA SSD that most would’ve either given away or recycled has become one of the more valuable components in my setup. In this strange market, old storage that you find in your drawer can also be an upgrade for your system. Offloading disposable writes to the SATA drive that has already served its purpose is a small adjustment, but it may just buy my main drive many years of additional life for the cost of perhaps twenty minutes of configuration.<\/p>\n