Which 4TB NVME Gen 4 for gaming

Minstadave said:
I haven't got Starfield but will try with Hogwart's to see if this is replicatable.
Click to expand...

I've just tried this with 2x 990 Pro 1TB, one being the OS drive on PCIe 4.0 M.2 CPU slot, one on its own PCIe 4.0 PCIe connection and 1x SK Hynix Platinum 2TB on a PCIe 4.0 chipset M.2 slot. Zero noticeable difference between running it off any of those (I did a bit of running about beyond just the bit in the screenshot and I couldn't be bothered to include a screenshot for the SK Hynix as it is basically the same):

OS Drive:

Lhi8vGr.png


Games drive:

CEj0M8Y.png


0QNhWMX.png


32GB RAM definitely makes a big difference with Hogwarts though - my 16GB older system isn't bad at running it but definitely not as smooth - Intel 14th gen runs this game crazily well actually - far smoother than my older Xeon, or 10th gen setups (neither of which run it badly) or the 12th or 13th gen systems other family members have.

I don't have Starfield to try that and it wouldn't surprise me in that game to be fair given the state of Bethesda games :s but I've seen no difference between the OS drive and games drives on any of my setups generally since moving to faster SSDs and onwards.

EDIT: This is with the system tuned for quiet over performance though - but at 4-5ms frametime any significant OS induced hitches should show up pretty plainly.
 
Last edited:
It's a silly argument, you only need to understand how the storage subsystem works to know the load/speed/latency of permanent storage has zero effect on FPS. There's a chance it could effect FPS when/if it needs to load resources but if that happened the speed of your permanent storage is the least of your worries because you're either running out of RAM or your processor is causing bottlenecks.
 
Last edited:
Games will probably start to leverage NVMe drive speed more in the future. As game assets increase in size, drive speed will become important. Also, Direct storage can bypass the CPU/RAM and stream directly to the GPU, when this is commonplace the speed of the drive will have more impact. I keep my drives for years so for me it makes sense to get fast ones, within reason. I am looking to get a SN850X 4TB at some point to replace an old 2TB gen 3 drive.
 
Alan Wake 2 uses 2.8GB/sec read bandwidth whilst playing, and that game doesn't even leverage DirectStorage or other fancy storage tech, just raw bandwidth. No game to date leverages anything at or above PCIe Gen 3 read speeds (3.5GB/sec) and it's unlikely any will any time soon either.

Minstadave said:
Does anyone ever get remotely near the TBW of an NVME drive? I'm a pretty heavy user and have never even remotely close (like not even 10%) before they get upgraded. Write endurance really wouldn't influence my decision much.
Click to expand...
It's a non-issue. Your PC will be outdated before you reach your SSD's TBW to warrant replacing it. It's the equivalent of writing up to hundreds of GB to it every day for however many years necessary depending on its capacity.

Plus, once written (a game being installed), reading off the SSD makes zero difference to the drive's health.

Edit*
For ref I just did a quick test in all my other installed games for SSD reads:

Code: 
Starfield:
Game start:        1.6 GB/sec
Save load:        2.2 GB/sec
Active gameplay:    2.2 GB/sec

Cyberpunk:
Game start:        2.3 GB/sec
Save load:        1.6 GB/sec
Active gameplay:    493 MB/sec

RoboCop Rogue City:
Game start:        1.1 GB/sec
Save load:        1.2 GB/sec
Active gameplay:    1.1 GB/sec

Alan Wake 2:
Game start:        2.3 GB/sec
Save load:        3.5 GB/sec
Active gameplay:    2.8 GB/sec

Ratchet & Clank Rift Apart (DirectStorage):
Game start:        323 MB/sec
Save load:        8 MB/sec
Active gameplay:    10 MB/sec
 
Last edited:
Those active gameplay numbers are interesting, i would've expected to see much lower numbers.

Don't waste your time testing but it would interesting to know if level loads, moving around, standing still, and why those numbers are so high. Technically speaking if you have enough RAM/VRAM there shouldn't be much disk activity once the game loads, at least not in the GB/s as you'd expect all the assets that a game needs to be loaded from disk ahead of time.
 
It's almost certainly due to the traditional way games are coded to load assets. DirectStorage is still brand new but Ratchet is an example of everything being low when DS is on, and higher when off. I forgot to add the latter but here it is:

Ratchet & Clank Rift Apart (DirectStorage OFF):
Game start: 2.1 GB/sec
Save load: 204 MB/sec
Active gameplay: 559 MB/sec

As you traverse a game level you pass invisible loading boundaries which then streams in the next load of assets to render reading off the disk. The modern way to do it would be for the game to be clever enough to measure the amount of RAM and VRAM available and chuck everything into either which would be much faster still, but that requires additional development time which as we all know, is something of a unicorn given the state of most games at launch. Devs will just factor in the most common hardware configuration and run with that.

So even though I have a 24GB VRAM GPU, and 64GB RAM, there's no instance in any game where within reason, any of those memory areas are fully utilised when loading up a game, which means the rest is being streamed in as I play through a level. Sometimes that will result in traversal stutters as the engine loads it in too slowly causing frametime hitches (Dead Space Remake, Jedi Survivor), whilst in other games it starts off this way but then is patched by the devs after much outcry (Last of Us Part 1).

The most efficient way to go about this is use DirectStorage, the game talks directly to the GPU and the SSD low latency and massive bandwidth deals with the streaming direct to VRAM, which on a Gen 4 SSD can be up to 22GB/sec leveraging the technology, or something around 15GB/sec for a Gen 3 drive, still ridiculously quick.
 
Last edited:
Yea the streaming in of assets is what made me wonder about the moving around vs standing still bit, it seems obvious that there wouldn't be a lot of disk reads when standing still, to the point that just asking makes it seem like a stupid question but, for me, it would be interesting especially with the proliferation of SSD.

Like are we at the stage now where having a game installed on a HDD can effect frame-rates/times in some games.
 
I did monitor the HWINFO64 window whilst standing still and the read speed is basically 0 whilst not doing anything, or just walking around the same area as everything that needed to be loaded to render that area is already loaded into VRAM/RAM so disk read is not required until the next traversal boundary is reached.
 
mrk said:
I did monitor the HWINFO64 window whilst standing still and the read speed is basically 0 whilst not doing anything, or just walking around the same area as everything that needed to be loaded to render that area is already loaded into VRAM/RAM so disk read is not required until the next traversal boundary is reached.
Click to expand...
If you had 8/16GB RAM the SSD would get hit more, not so much at first load but overall. Windows cashes every file that’s loaded so if you access it again it might be in RAM (depending on free RAM). Having 64GB of RAM helps more than most would think, even if the software does not use it directly. My old PC has 64GB and I used that for > 5 years. I went with 32GB as DDR5 is still a bit of a mess, will upgrade at some point through.
 
  • Like
Reactions: mrk
You must log in or register to reply here.
Back
Top Bottom