Load times with gen 4 nvme 7400 MB/sec vs Gen 3 3400 MB/s, surprising results or maybe not (no change)

I was running out of space on my 2TB XPG 3400 MB/sec gen 3 nvme (https://www.amazon.com/gp/product/B07TY2TN64/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&th=1) ,

so i opted to go for the 4tb 7400 MB/sec version upgrade. (https://www.amazon.com/dp/B0B2FSKTHK?ref=ppx_yo2ov_dt_b_product_details&th=1)

I run the sim exe on the C OS drive, which is also a 3400 MB/sec nvme (perhaps a factor here, being the os is gen3).

Su12 load times on the old 3400 drive are 2min 21 sec for what is basically 1.4TB of MSFS data on a fresh win 11 install.

Su12 load times on the new 7400 drive are… the SAME. Nearly exactly the same.

I guess its not surprising as often with simulators, the load times for files arent optimized to match the hardware (i think) or the ssd isnt the bottleneck.

Has anyone else made the upgrade in this manner and seen the same? (I was able to verify with crystal diskmark v8 the 7400 speed on reads is accurate)

**or has anyone found that if you do both the OS and the area where MSFS files are located then the speeds jump (so the OS has to be the same speed)

**my hardware: i9-12900k at 4.9ghz, 6400mhz ddr5 ram, 4090 Gigabyte OC, mb msi z690 meg unify-x

(so since currently using ddr5 6400 which is around 52 GB/sec which is 6400 MB/sec (vs the ram which is 7400 MB/sec), since the old nvme was around 3400 MB/sec now its fully on par and then some )

1 Like

MEG Z690 UNIFY-X?

Yep thats the one

that have nothing todo with that… There is just no big real benefit of all that advertinsing numbers ( like the frequencies in RAM ). On the paper the numbers looks great, but in real it have not the same ‘double it’ effect, also because “MB per Sec” is not all what count and often there is a big portion of advertising included ( how many seconds can the controller work in that speed, caches, read/write/ , etc. ). So, I not expect really a relevant difference. Additional a huge part of startup msfs seems the update check, etc… I remember similar discussions in other game forums for e.g. the 970 EVO vs 980 pro , where users also not noticed any kind of difference ( and not a simulator ).

In my former PC I had MSFS data on a SATA SSD. In the new PC it is one a 980pro nvme , and the loading time of msfs is nearly the same ( I not count the seconds :wink: ).
If we check e.g the taskmanager, we see a short-time high read-rate, and thats it. I buy the “pro” variants for drives which are important or where I expect that more data are written ( the pro’s have usually more warrenty ) or also where I expect that huge files needs to be transfered ( my video projects ). For all others I still have “old” SATA SSDs and not care about that the game possible loaded 2 seconds slower :wink:

Also lots of gameing magazines reported about that facts…
here a compare of loading time SATA vs pci3 vs pci4


source: Samsung SSD 980 Pro im Test - Extrem schnell, aber (noch) unnötig

There are tons of other magazines, man yof the youtube guru-advertising-influencer videos, etc. about that :slight_smile:

2 Likes

Ah nice comparison graphic, i hadnt seen many of these. I guess its nothing related to the fact the OS drive is still gen3 either then. Just how it is.

I did find on both drives the “bare bones” load time, and by bare bones in my case it was more like 250GB of official content (vs 122 or so really bare), was identical at 1min30sec too as a side note, though thats expected as well.

yep, its often just not much data that the benefit of the higher bandwith can be noticed. If you copy a 200GB video file, the situation is different - as long the ssd controller can hold the speed ( temps are a huge topic for nvme’s ).

I also re-read your post about DDR vs NVME: it is >50GB/s for your 6400’er in Single-Channel , but usually we use Dual Channel Setup ( hopefully ) . So ist more like 100 vs the max of pci4 is ~7.9GB/s ( 4 lanes) . Thus RAM is still much faster and have not the most of the issue of the ssds chips, etc.

1 Like

Well I was thinking about getting a 2TB pcie gen 4 NVME for the sim now that I have a mobo and such that are rated to handle it. Currently using a 1TB PCIe gen 3 drive for everything, but maybe I’ll just save my money with a gen 3 drive or SATA SSD.

in case it was planned only for MSFS with expect to speed something up, I would safe the money.
Where I some time recommend a second drive, is to “safe” the lifetime of the main-drive. But thats also more optional than necessary.

Well the differential on 4tb gen4 at 5000 MB/sec is at a price of like $224 vis the gen4 4tb 7400 MB/sec was about $350, so i viewed it as worthy, not a huge difference, not with 4tb anyway, with the potential “future” benefit of the added speed, though for windows file transfers its faster (i guess for sure if both drives were 7400), point being that there really werent any gen3 4tb’s at a significant cost decrease over the slightly slower gen4 4tbs out there.

1 Like

I think any differences between gen3 and gen4 are more likely to be felt loading large scenery files into ram while flying at speed or down low over PG. IMO as yet initial load times are more likely relevant to single core cpu performance unless you only have mechanical HDDs which really are too slow. When we get it Direct Storage should speed things up considerably.

Edit: I’ve just read the OP again, if I understand rightly he has MSFS installed on the PCIe3 which as this makes up by far the largest part of the loading time the only time he could possibly save is for cpu to read his register of content and compare versions to the sim’s latest, meaning no more than seconds. Going back a couple of years I remember swapping my NVMe drives and with an empty community and just the Toprudder 103 solo added the gen 3 was 20 seconds slower than the gen 4’s 105 seconds. Of course much has changed since then including the rest of my system.

thats a good point. If all preconditions meet ( win11, nvme, up2date driver ) then the loading time can reduced. But also here I not realy expect a “click and game is on”, because the startup is more than loading textures into the VRAM. But currently MSFS does not use the DirectStorage-API and so users not need to hurry and can wait till the point it is implemented and getting feedback how much faster the loading time is ( and can then decide whether its worth to spend 300$ for a reduce of e.g. 30sec of game-load-time :wink: )

Just for added info … for smooth high quality 4k recording @ 60fps you really need PCIe4 NVMe but your mobo must also be compatible. PCIe3 is ok for 4k 30fps or for lower quality.

ähmmm… nope… the “normal streamer” does not record raw-data ( uncompressed 4K ) , instead they record already h.264,265,… or what ever files ( often eg. the nvidia encorder is used ). High quality for that youtube worst-re-render-advertising-platform is already 100mbit and that can an old HDD handle. Also a 4K raw stream is may be 2-3GB/s ( I remember more about, without recalc, about 12Gbit (b vs B ) ) and that can PCI3 also do.

at 60fps it’s at least 3GB but last time I looked PCIe 3 will not sustain that over any extended amount of time.

Just looking in Task Manager statistics for yor MSFS drive you could see that there
nothing near than high disc usage during loading into main menu of MSFS.

Upgrading from a SATA SSD (approx 350MB/s) to NVME SSD (approx 3500MB/s) for me
gave near to zero performance gain.

My guess: MSFS reads only the small JSON and XML files to parse the content to expect and build
the memory library and base world to work with during flights.

To validate this i tried adding 10 big airports (in sum same size on disc as fstltl lib) and measure time compared to the fsltl library with 1500 mini sceneries (every aircraft livery combination has its own xml/json files) … the 10 airports have minimal impact on loading time, the fsltl files double loading time.

The only time where i expect improvements i during load into flight and scenery loading during flights with less impact to fps.

1 Like

Direct storage can’t improve loading into menu, because MSFS is loading mainly the
base information (xml,json files) during startup and the long time it needs is to organize
all the packs. Maybe it improves loading into flight from menu, but i don’t think the
improvement will be worth many $ for storage optimization.

Other things like panning around texture loading stutter maybe can be improved.

I think Asosbo will not open the pandoras box of direct storage, because there are many
things that will stop direct storage from work, like wrong drivers, bios settings, …

I have updated to Win11 to make my PC direct storage compatible and found out some
default bios settings for the NVME slot will block it, because intel RST was active. Switching to
AHCI stops windows from booting. It need some steps to change a running system from
Intel RST drivers to MS AHCI drivers.

1 Like

I found a wiki for your 11.9Gbit :slight_smile:

( and we speak not about DCI :wink: )

@DensestSnail693 ah, there was a little edit. Yep, let assume 3GB. … ähmmm… I removed the confusing sentence with SATA :face_with_hand_over_mouth: … I have in mind PCI3 :laughing:


@GoSammy1971 yes, the high load is not allways. Its some release ago than I checked that. I also assume that in most cases only the “index” is checked.

And also for DirectStorage: yes… the loading time into the game is may be also not realy affected, just because at these point no huge textures must be loaded. So, yes: more exact to say is: the loading into the flight becomes possible faster. I also assume the effort isnt it worth, but users hear somewhat, creates whishes over wishes and then the priorities goes some times strange ways ( on other side its also a publisher behind that game, and these publisher is a big fan of direct-storage :wink: ).

1 Like

I stand corrected then,12gigabit = 150Mbps so yes gen 3 should be able to.

So I guess the technology is changing faster that I am :weary:

1 Like

more funny is PCIe5… all speak about, nobody have realy a solution that works stable and more important, nobody currently need that. In special not, if mainboard manufactors like Gigabyte are not able to produce mainboards, where you can use all lanes of PCI4 for your GPU ( not a shared port , and the fix from that manufactor is a bios with a default setting, where the lanes are limited ). What should then works fine when the first PCI5 card arives :slight_smile:
Therefore the hint for the users: dont hurry. If users now buy a new drive, then I would also choose a pci4 one. They are faster and so its bit more future-proof. But upgrade the system with a new nvme and think that all becomes faster , like a magic wonder, that will not happen ( except users come from a hdd :slight_smile: )

1 Like

It would be interesting to optimize the drive’s cluster size for best results.