Storage Game Loading Test: PCIe 4.0 SSD vs. PCIe 3.0 vs. SATA vs. HDD

[DyneCorp]

"For sequential writes, the story doesn’t change much either, although this is peak performance from writing into those DRAM caches."

Just a correction:

The data is being written onto the SLC cache buffer, not the DRAM cache buffer. No user data is written to or passed through the DRAM cache; DRAM in an SSD is used for logical block addressing only.

This is why DRAM has no affect on game load times.

SLC caching is where the controller treats part of the 3-bit TLC NAND as 1-bit per cell SLC NAND; this increases speed and endurance. Most of the drives tested have multi-tens of gigabytes of SLC cache; a couple have over a hundred gigabytes of SLC cache.

 

[defaultluser]

Try this: Make a gigabyte's worth of files in the range of 128 to 512 bytes, then record the speed of reading and writing that folder with the same set of drives.

Crystal disk mark random read performance Q1T1 is supposed to be that "worst-case scenario, " where you have a single video game that is not doing a very good job multi-threading a load.

It is pretty accurate, as the slowest SATA SSD is 30x faster than a hard drive!

Also of no surprise is the fact that NVME is only 2x better at reads, as the theoretical increase in is 2us latency for NVME (versus 6us latency for SATA6),

These drives should be capable of more throughput, but when you're feeding a Game Engine, there's only so much block I/O these games actually use . We will have to wait and see how much better Sony's hardware transfer system can get around those limits?

 

[EdmondRC]

Unfortunately, having just built my PC last year, I'm stuck with PCIe 3.0 for a while. I don't expect I'll upgrade for at least five years. The good news is I have two nvme drives, a 500 GB WD black SN750 and a 512 Adata XPG Spectrix s40g (which I think is the same as the SX8200 except it has an RGB feature). The load times I get in games are better than acceptable in my opinion, I don't feel like I wait long at all. Cutting it in half would be nice of course, but since the load times are so short to begin with its not essential. Of course, with the PS5 leading the way with required fast load times that will actually change the in game experience rather than just the loading experience, I guess there is a good chance that there will be some games, at some point, that I will not be able to play or at least not be able to play optimally. I never would have thought last year, when buying near the top of the line nvme drives that this year I would be looking back and wishing I had waited a year to build my PC. Oh well.

 

[MaxSmarties]

Long story short: just buy an SSD. It doesn’t matter which model.

 

[Bp968]

Id like to point out not all QLC drives are the same. I have a intel 660p 2tb with a dram cache and in 99% of use cases it easily matches the top TLC drives. I bought the 2tb for 180$ 1-1.5 years ago. At that price it was similar to a 1tb TLC nvme drive. As long as you leave a few hundred GBs available for the SLC cache its extremely fast.

I recently bought a 2tb TLC 8600 pro drive (the adata drive) for like 220$ and didn't see the 660p or 665p at nearly as attractive a price as when I bought mine so I would (and did) get the TLC drive instead. But if its enough of a difference to buy the next size up I would absolutely buy the QLC drive.

I really hope we will see a large uptick in demand for larger NVME drives and ill be able to grab a 4TB nvme drive for 200-250$ this time next year. The 2tb tlc drive is for my wifes 5900x build this year so ill have a free M2 slot next year in both systems. (And my primary system is stuffed full of radiators so I had to excise the 3.5" hard drives from inside the case and I am missing the extra storage since I enjoy installing many hundreds of steam games all at once.. lol)

 

[duckofdeath]

Sabrent Rocket 4.0 FTW \o/
Budget overkill-power for the masses.
Got one myself on a sale early this year, because you can never have too much excess of power. I think?

 

[Angga B]

I use wd red 4tb hdd to store my game and use primocache to make 100GB L2 cache out of my 256Gb boot sata ssd and top it with 1GB L1 cache out of my 16GB ram.

I time all my game loading, direct from ssd, direct from hd and the primocached setup. The latter does come up as fast as the first.

And no this is not advertising. I pay full license and I am happy.

 

[Gezzer]

Could you of picked a worse HDD? Like really??
A 5400RPM, you might as well stuck a bolt threw it while you were at it. No one building a gaming system with a HDD storing some of the games will use a drive like that, it's for archiving, not games.

You should of added something like this to your test suite.
https://www.newegg.ca/black-wd6003fzbx-6tb/p/234-000G-000W5
Then you would of had a better snap shot. And yes the SSDs would still be faster than the HDDs, but not by nearly the same margin. As it stands IMHO this isn't a real scientific test, just one meant to prove the already reached conclusion... nothing more.

 

[poohbear]

Thank you for this much needed review and your recommendations at the end!!! So much marketing hype out there and great to see you guys sift through that!

 

[arrowflash]

This review confirms something I already knew: In real world usage with games and applications, there's virtually no performance difference between SATA and PCIe SSDs. I've always prefered SATA SSDs because I simply think they're much neater and easier to manage from a hardware standpoint. PCIe SSDs always looked to me like a retrocess back to the 1980s and early 1990s when people were buying hardcards (3.5-inch HDDs bolted to controller cards) and plugging them directly into ISA slots.

I completely disagree with Tim Schiesser in this article when he says that it's worth paying a premium for a top end PCIe 4.0 SSD to use as a boot drive. Me and most power users I know only boot their PCs once per day at most, and of course there are many who never ever turn it off. Unless you're one of those people who turn off their PCs every time they leave their desks, you'll be paying a premium for gaining 3 or 4 seconds per day at most. One might argue that you also gain time when loading applications, but then again, for daily use applications I usually leave them open in the background. Plus, my SATA SSD already loads most programs virtually instantly.

Also, like others have already said: You couldn't have picked a worse HDD for this review. How about including a 7200 rpm model or even a 15000 rpm SAS HDD. I'd love to see how they'd fare against the slowest SSDs.

 

[Lounds]

Its going to be several years before we have more than one game that actually utilizes more than SATA6 bandwidth. By that time, my systems will be ready for a PCIe4 update.

But I told all those folks worshiping that PS5 SSD - it's going to be processor-limited in most cases. Even the R9 3900X is not good-enough to more than slightly exceed the SATA6 drives in a couple games.

If current console-pushing games can't really show more than 10% boost over SATA6 SSD, then it's going to be another 5 years before optimized PS5 titles push us over that load bandwidth hump. And even then, it's going ot be satisfied by cheaper PCIe 3 drives (because they will hit that CPU limit way before PCIE4.)

The PS5 has a dedicated controller for streaming data, the CPU will not handle this task in the PS5, I assume it will be the same for the XSX. It'll be a while before PC catches up to the game load speeds of the new consoles due to PC bottlenecks.

 

[adsasdu8ads7adsy]

"Instead, the main limitation seems to be things like how quickly the CPU can decompress assets, and how quickly it can process a level before it’s ready for action, rather than how fast it ca.n read data off storage." If you are going to mention the CPU as the bottleneck, then why not specify the CPU or ideally the entire system used for the benchmarks? This seems like missing information that's critical.

 

[pildorman321]

Currently using an old sata ssd. will put my 1.TB EX950 back when mocosoft / nvidia stops napping with directstorage and RTX IO

 

[ddferrari]

Currently using an old sata ssd. will put my 1.TB EX950 back when mocosoft / nvidia stops napping with directstorage and RTX IO

So do you think that you're "punishing" Microsoft and Nvidia by not using your NVMe lol, or are you just looking to extend its life from 10 years after you don't own it anymore - to 11 years after you don't own it anymore?

 

[pildorman321]

So do you think that you're "punishing" Microsoft and Nvidia by not using your NVMe lol, or are you just looking to extend its life from 10 years after you don't own it anymore - to 11 years after you don't own it anymore?

Neither. I Have SEVERAL SSDs ( SATA, M2 SATA, NVMe, Gen 3, Gen 4) Just like use the right tool for the job.

 

[RobMonteforte]

The issue is that the people who comment on the forums believe that Game Loading Times means that the whole issue of OS performance, especially the pagefile.sys access, is identical regardless of drive used. The OS just doesn't load into memory and stay static, and with the pagefile turned off, there is still need for it. For a file used thousands of times a second, its read/write access time matters.

 

[DonquixoteIII]

This is the second time TechSpot has done one of these types of articles, and each time you have included the slowest possible HDD. Use a 10k or faster for believability.

Then there is the matter of what you are actually testing. In the real world, (say, loading games) the units tested have to deal with a ton of smaller files, and you do not test for that in your initial load testing. So, in a sense you are showing the readers an apples and oranges testing regime. Maybe include a RAM disk for completeness... It will show barely any difference in the loading times, because ALL 'drives' suck when it comes to loading small files.And they pretty much suck equally.

Try testing your group with real world synthetic tests first, maybe a gigabyte's worth if mixed 128 bit through to 2k file sizes, THEN run your game loadings. Surprise, they are pretty much the same....

And, don't forget to run the tests AFTER your drives have been fully loaded with data first....

Then there is the sequential loading, where the second load is faster than the first, due to your OS doing some caching, and the third load a bit faster still...

 

[Ben Myers]

Belaboring the obvious, are we????

 

[msroadkill612]

A self evident & verbose answer to a stupid question.

In the current dominant gaming paradigm, with adequate CPU & ram, the only significant role for the advantages of direct cpu linked nvme, is the arguable bother of game load times - sequentially reading large blocks from storage to fill gpu cache & some system ram.

Guess what Sherlock? Its not a lot.

The one thing that doesnt change, is that paradigms change.

As mainstream resources up over time, games will adapt to provide a richer experience.

A more intelligent question to test, would be perf in the more futuristic & challenging MS flight simulator, on a rig w/ a struggling cpu w/ modest system & gpu ram - to force the active participation of storage during game play - just as trend setting MS FS does I hear - scenery etc is loaded dynamically from storage/nvme during play.

We know the raw numbers are radically better for the new tech storage. This is just a dopey test for them.