I wish I could buy a 2TB Nvme for $107 in 2026

Basically a cheap Pcie 3.0 Nvme is enough for 99% of all users and beyond that is luxury over nescessity

This is good content that a lot of people need to hear especially with the cooling requirements of 5.0

The problem with SSDs is that people tend to only pay attention to sequential read/write speeds... because that's what these companies push. However, for gaming, small file transfer speeds and IOPs are far more important and there are diminishing returns once you move past a SATA to NVMe, which is why the differences are pretty marginal.

Pcie 4.0 nvme's are as cheap as 3.0 so just get a Pcie 4.0 nvme ssd and call it a day

Not even using an ancient 10k rpm Raptor HDD for the test... now we will never know the real performance king.

The SSD prices quoted in this video aged like fine milk!

Above all, get a big one. 2-4TB M.2 drives are the most cost effective SSD's today.

DRAM cache helps when verifying many small files because Steam does not happen to read the files in the same order they are actually stored on the drive. DRAM-less drives have a much smaller cache for mapping tables, no more than 64 MB, while DRAM drives usually have 2 GB, which is enough to store the complete mapping tables. Whenever a DRAM-less drive has to read some data from a place that it could not predict and that is not in the cache, it also has to read that part of the mapping tables from flash, i.e., one read request from the PC results in two flash reads.

The DRAM cache is used only for mapping tables, not for actual data. Writes are faster because the drive can more quickly determine the correct place to write to.

My man is holding $1000 of ssds in his thumbnail lol

I'm glad you had a write test at the end that showed the potential problems of QLC.
I strongly recommend everyone to avoid QLC drives, because TLC drives are only a little more expensive, but are going to last MUCH LONGER, be MUCH MORE consistent at writing, and less reliant on cache.
QLC is so slow to write directly to, the drive has to write everything to cache first, and most drives actually provision QLC as SLC. First the drive writes files in 'SLC mode', using theoretically up to 25% of free space (Q = quad, S = single). Later on, during downtime, it slowly re-writes the data densely in QLC mode. This means two big problems: You end up with A LOT of writes to your SSD, just using it normally, and as you fill up your drive, you lose more and more SLC fast-write cache space, so it will begin to feel more and more like a terrible spinning disk HDD. If you fill you drive up to 80%-ish, and then use it fairly intensively, you can end up putting a massive amount of writes onto a fairly small portion of the drive's memory modules, leading to failure long before you expected it.

Just. Use. TLC. or better.

Wait! I've been asking for this for years! You finally did it!!!!! Thanks!

Just get a gen 4 SSD with dram and you are set for probably 15 years.

That Ratchet and Clank example was crazy. That’s a good example of direct storage.

Please don't just focus on loading times, but consider also testing the 0.01% low frametimes (*NOT* just the 0.1% or 1% lows), the frametime pacing, and measure how long it takes for stuff to 'pop in' for games that use data streaming techniques during gameplay.

If possible, please compare the results with that obtained by one of the last (PCIe 4.0) Optane drives, to give an indication of how modern SSDs compare and how much performance headroom is available.

In this era of games, i think you guys should also start testing framerates/frametimes too. Some games are worse or unplayable on hard drives. Also should've tested Marvel Rivals. Using a HDD in that game is the same as throwing since you load 30 seconds into a match basically.

Good video, often see too many people either wanting to buy, or recommending other people buy high end NVMEs, like the Samsung 990 Pro, in new mid range gaming builds where the budget just doesn't allow for it.  I keep trying to explain that any old cheap gen 3 NVME will work just a well, and that the faster NVMEs are mainly for productivity workloads, but they often don't listen.  Will be nice to have a video to link as an example now :)

Hey, looks like my 4TB Samsung SSD (it's still sata, but hey I got it for $80) won't be going anywhere soon.

The Total War series, where you load into battles and then load back into the world map, make a great series of games to test for this sort of thing, should you do this test again someday. Going from HD to Sata SSD for TW Warhammer was a huge quality of life improvement a number of years back.

Best benchmark i have seen in decades