NVME (m.2) / RAM Drive / RAID / SATA III SSD Game Load Time Comparisons

arestavo

[H]ard|Gawd
Joined
Mar 25, 2013
Messages
1,976
I completed this testing to help folks make a more informed decision when buying a new drive (or drives) when their primary focus is gaming.


I know this won't be popular with some, but I took several hours to complete game load time testing on an EVGA X99 Classified board (2.03 BIOS) between a 8 drive RAID 6 array with an Adaptec 8805 RAID controller running at PCIE 3.0 X8, a 512GB Samsung 950 Pro NVME drive in a PCIE 3.0 X4 expansion card with passive heatsink (running at PCIE 3.0 X4 speeds), a SATA III 512GB Samsung 850 EVO SSD (RAPID mode isn't enabled), and a 42 or 55.8GB RAM drive (64GB of 3200MHz DDR4 16-18-18-32).

Full computer specs can be found in ModRigs (and my sig!).

All games were moved using Steam Mover (an awesome tool!) - it does work on both Steam and Origin games (and should work on any game as long as you point it to the correct game directory).

All results are given in seconds - e.g. 13.44 seconds, or 13 point 44 seconds. Times were taken using a stopwatch app (yes, there is a slight human reaction time error inherently built into each result).

All game loads were tested three times in a row.

No, I can't test all of the games with a RAM drive since many games are larger than 50GB (I only have 64GB of RAM you know! - unless you want to spot me 128GB or the cash for it!).

Yes, I'll test a game for you if I have it! (just ask me via PM as I may not check this thread often)


Batman Arkham Knight 1080P
Savegame of 5% at Ace Chemicals, in game settings manually set to max and 1920X1080.
My timer was started when I clicked Continue Story and was stopped when I was loaded into the game.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
9.17
9.06
8.89

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
9.75
9.14
9.07

512GB 850 Evo (empty except for B:AK)
9.69
9.44
9.38

42GB RAM drive (3200MHz DDR4 16-18-18-32)
9.41
8.94
8.94


Batman Arkham Knight 4K
Savegame of 5% at Ace Chemicals, in game settings manually set to max and 3840X2160.
My timer was started when I clicked Continue Story and was stopped when I was loaded into the game.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
9.91
9.02
9.10

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
9.95
9.34
9.22

512GB 850 Evo (empty except for B:AK)
9.71
9.12
9.26

42GB RAM drive (3200MHz DDR4 16-18-18-32)
11.56
9.16
11.27 (ran several times more to see, always between 9 and 11 seconds)


Shadows of Mordor 1080P
Storymode savegame of 3% (24 mins), in game settings manually set to max and 1920X1080.
My timer was started when I clicked Continue Story Mode and was stopped when I saw (ESC) Skip in the lower left corner.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
6.48 (I restarted the computer after this first result to be sure there wasn't anything cached as I did the RAM drive first, the NVME second, then the RAID third)
6.85
6.24

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~135GB free)
5.86
5.80
5.88

512GB 850 Evo (empty except for SOM)
5.86
5.99
5.88

55.8GB RAM drive (3200MHz DDR4 16-18-18-32) (~62 of 64GB in use)
6.08
6.32
6.37


Shadows of Mordor 4K (appears to be DSR)
Storymode savegame of 3% (24 mins), in game settings manually set to max and (200%) 3840X2160.
My timer was started when I clicked Continue Story Mode and was stopped when I saw (ESC) Skip in the lower left corner.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
5.95
5.91
5.88

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~135GB free)
6.08
5.97
5.85

512GB 850 Evo (empty except for SOM)
6.18
5.85
5.96

55.8GB RAM drive (3200MHz DDR4 16-18-18-32) (~62.5 of 64GB in use)
6.01
5.95
6.34


Batman Arkham Origins 1080P
Savegame of 0% at Blackgate Prison, in game settings manually set to max and 1920X1080.
My timer was started when I clicked Continue Story and was stopped when I was loaded into the game.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
5.31
4.60
4.59

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
5.30
4.49
4.61

512GB 850 Evo (empty except for B:AO)
4.62
4.56
4.35

42GB RAM drive (3200MHz DDR4 16-18-18-32)
5.30
4.69
4.66


Batman Arkham Origins 4K
Savegame of 0% at Blackgate Prison, in game settings manually set to max and 3840X2160.
My timer was started when I clicked Continue Story and was stopped when I was loaded into the game.


8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
5.56
4.70
5.07

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
5.06
4.76
5.02

512GB 850 Evo (empty except for B:AO)
5.63
4.86
4.88

42GB RAM drive (3200MHz DDR4 16-18-18-32)
5.63
4.72
4.89


Fallout 4 1080P (no mods)
Used a save game from one hour thirty minutes into the game. (1920X1080 and all game settings manually set to max)
Timer started when press Enter to Confirm (load) was pressed and stopped when the game loaded in.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
29.11
22.61
22.91

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
7.65
7.01
6.68

512GB 850 Evo (empty except for FO4)
7.49
6.81
6.67

42GB RAM drive (3200MHz DDR4 16-18-18-32)
6.9
6.54
6.63


Fallout 4 4K (no mods)
Used a save game from one hour thirty minutes into the game. (3840X2160 and all game settings manually set to max)
Timer started when press Enter to Confirm (load) was pressed and stopped when the game loaded in.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
27.83
22.74
23.07

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
8.33
7.87
7.75

512GB 850 Evo (empty except for FO4)
8.98
7.92
7.88

42GB RAM drive (3200MHz DDR4 16-18-18-32)
7.89
7.92
7.93


DOOM 1080P
DOOM Campaign, Advanced Research Complex save game (1920X1080 and in-game settings manually set to max)
Timer started when Continue Game pressed and stopped when press spacebar was displayed in the upper right corner.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
13.12
11.07
11.02

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
11.86
11.64
11.67

512GB 850 Evo (empty except for DOOM)
12.26
12.13
12.02


DOOM 4K
DOOM Campaign, Advanced Research Complex save game (3840X2160 and in-game settings manually set to max)
Timer started when Continue Game pressed and stopped when press spacebar was displayed in the upper right corner.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
12.99
11.18
11.28

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
11.89
11.78
11.73

512GB 850 Evo (empty except for DOOM)
12.22
12.24
12.19


Dragon Age: Inquisition 1080P
1920X1080 and in-game settings manually set to max.
Timer started when Continue was pressed and stopped when game loaded.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
25.16
12.87
13.00

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
13.09
12.52
13.06

512GB 850 Evo (empty except for DA:I)
16.13
12.59
13.02

42GB RAM drive (3200MHz DDR4 16-18-18-32)
12.96
13.08
13.07


Dragon Age: Inquisition 4K
3840X2160 and in-game settings manually set to max.
Timer started when Continue was pressed and stopped when game loaded.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
20.11
20.08
19.29

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
22.24
19.68
19.63

512GB 850 Evo (empty except for DA:I)
18.09
19.63
18.71

42GB RAM drive (3200MHz DDR4 16-18-18-32)
21.82
18.96
19.73


Titanfall 1080P (Multiplayer)
All match loads were into the Attrition map, in game settings manually set to max and 1920X1080.
My timer was started when I clicked on the attrition map and was stopped when I was loaded into the match, with two exceptions where I had to wait several minutes in a lobby to get a match - for those I started the timer when the match counter read 0.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
12.24 (no lobby)
11.56 (no lobby)
11.97 (no lobby)

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
9.25 (lobby load, timer started at 0)
11.88 (no lobby)
12.31 (no lobby)

512GB 850 Evo (empty except for Titanfall)
7.28 (lobby load, timer started at 0)
12.11 (no lobby)
12.52 (no lobby)


Titanfall 4K (Multiplayer)
All match loads were into the Attrition map (edit: all games loaded into were already in progress, no lobby time), in game settings manually set to max and 3840X2160.
My timer was started when I clicked on the attrition map and was stopped when I was loaded into the match.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
14.38 (no lobby)
12.67 (no lobby)
12.59 (no lobby)

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
14.32 (no lobby)
12.35 (no lobby)
12.81 (no lobby)

512GB 850 Evo (empty except for Titanfall)
14.73 (no lobby)
13.69 (no lobby)
13.57 (no lobby)


To back up my overall findings and prove that I'm not full of it, here are some reviewers who actually did some game load testing as well as Windows boot time testing - which almost NO reviewers actually do (I wonder why that is, ehhhhhh?):

http://www.legitreviews.com/game-load-time-benchmarking-shootout-six-ssds-one-hdd_204468

http://techreport.com/review/30813/samsung-960-pro-2tb-ssd-reviewed/5

Samsung's 950 Pro 512GB SSD reviewed

Samsung SSD 950 Pro review | PC Gamer

Toshiba's OCZ VX500 512GB SSD reviewed
 
Last edited:
Wow you really went the distance! I am pretty surprised about the ramdisk being within spitting distance of an 850 evo. Am I interpreting right where it seems the cpu, ram, or PCI-E subsystem is bottlenecking now?
 
Wow you really went the distance! I am pretty surprised about the ramdisk being within spitting distance of an 850 evo. Am I interpreting right where it seems the cpu, ram, or PCI-E subsystem is bottlenecking now?

Thank you!

It very well could be that. I'm leaning more towards games not being able to utilize the massive queue depth performance that NVME drives have (and I assume that RAM drives have as well), because once we virtually eliminated seek times game load times plateaued.
 
Last edited:
I have no idea how exactly the RAM drive works when you look at the queue, with SATA and NVMe as comparisons. But it shines in of course with latency. Millisecond? HA!
 
From what I can tell, the 512GB 850 Evo is nearly as fast as the 512GB 950 Pro NVME for nearly half the price.
 
My guess would be that the OS is still performing limited caching actions in the background when you use a RAM disk. This would mean that you have more RAM load then strictly needed.

Caching RAM in other RAM is kinda pointless. You could look into disabling read and write cache for your RAM disk and test if that affects the result.
 
My guess would be that the OS is still performing limited caching actions in the background when you use a RAM disk. This would mean that you have more RAM load then strictly needed.

Caching RAM in other RAM is kinda pointless. You could look into disabling read and write cache for your RAM disk and test if that affects the result.

Disk Write Caching - Enable or Disable in Windows 10

Is that what you're talking about? If so, I'll have a go at it tonight if I have the time!
 
From what I can tell, the 512GB 850 Evo is nearly as fast as the 512GB 950 Pro NVME for nearly half the price.

The main benifit is higher IOPS especially at high queue depths. This type of workload is not typical for games or most single desktop applications.
 
This makes my Boxing Day score of a 115 850 Evo 500gb so much sweeter last December. Thanks for the hard work!
 
This makes my Boxing Day score of a 115 850 Evo 500gb so much sweeter last December. Thanks for the hard work!

Thanks!

Did you reboot between switching drives?

All drives were present in the system so no reboot was needed, just had to move the games from one to the next. The RAM drive was disabled until the RAM drive was tested.

I didn't test each game with the exact same drive order each time either, but I did test 1080P/4K or 4K/1080P one after the other.
 
Last edited:
I was worried about cache effects between tests. I mean if you tested 1 drive then copied the data from that drive to the next then tested again with the next drive you may be measuring cache performance.
 
Wow you really went the distance! I am pretty surprised about the ramdisk being within spitting distance of an 850 evo. Am I interpreting right where it seems the cpu, ram, or PCI-E subsystem is bottlenecking now?
I would hazard a guess of NTFS being from 1993 not helping matters.
 
So what exactly needs to be done to take advantage of the hardware we have now? I complete re-write of the file system? Game software needs to be coded different?

What is holding back games from loading pretty much instantly?
 
So the other drives tested ran with full 64gb of ram while the ram drive config ran with 24gb of ram (and the rest in a ram drive)? Just want to be sure I'm reading the data correctly.

Also any chance of throwing 2x SSDs in RAID 0 and testing it for reference? The big HD RAID is fun to look at but won't be as common a config compared to 2x SSDs in RAID 0 IMO.

EDIT: typo!
 
Last edited:
So the other drives tested ran with full 64gb of ram while the ram drive config ran with 24gb of ram (and the rest in a ram drive)? Just want to be sure I'm reading the data correctly.

Also any change of throwing 2x SSDs in RAID 0 and testing it for reference? The big HD RAID is fun to look at but won't be as common a config compared to 2x SSDs in RAID 0 IMO.

Correct. Even the RAM drive tests only saw me use about 55GB RAM total as seen with Afterburner.

I had two striped SSDs (and four 32GB drives once - but that was when SSDs were fairly new and drivers were so-so) some time ago. Yes, I am that guy who doesn't believe in saying "RAID" 0. While I don't have any data from that time, the game load times were no different than with a single drive.

Think about it - two SSDs optimally running at ~550 MB/s each, multiply by two and.... still slower than the top end NVME drives for sequential read and write speeds. And my data, and reviewers provided at the end of my first post, shows that sequential read/write speeds don't mean much for game load times.

If I could find it (it came out a couple of years ago) there was an article showing that striping SSDs can actually hurt game load times. I don't think I'll be able to find it, and it would be out of date anyways since drivers matured and OSs have changed.
 
Last edited:
Thank you for doing this. I love this kind of stuff that really gets to the truth of the matter.
 
Correct. Even the RAM drive tests only saw me use about 55GB RAM total as seen with Afterburner.

I had two striped SSDs (and four 32GB drives once - but that was when SSDs were fairly new and drivers were so-so) some time ago. Yes, I am that guy who doesn't believe in saying "RAID" 0. While I don't have any data from that time, the game load times were no different than with a single drive.

Think about it - two SSDs optimally running at ~550 MB/s each, multiply by two and.... still slower than the top end NVME drives for sequential read and write speeds. And my data, and reviewers provided at the end of my first post, shows that sequential read/write speeds don't mean much for game load times.

If I could find it (it came out a couple of years ago) there was an article showing that striping SSDs can actually hurt game load times. I don't think I'll be able to find it, and it would be out of date anyways since drivers matured and OSs have changed.
I think he's hoping for RAID0 with two 950 pros
I found a couple reviews, same MB no mention of game loading that I could see
Triple M.2 Samsung 950 Pro Z170 PCIe NVMe RAID Tested - Why So Snappy? | PC Perspective
Samsung 950 Pro M.2 PCIe Gen 3x4 NVMe SSD RAID 0 Report
 
Correct. Even the RAM drive tests only saw me use about 55GB RAM total as seen with Afterburner.

Think about it - two SSDs optimally running at ~550 MB/s each, multiply by two and.... still slower than the top end NVME drives for sequential read and write speeds. And my data, and reviewers provided at the end of my first post, shows that sequential read/write speeds don't mean much for game load times.

Right, I'm with you on the NVME v. RAID0 drives in practical use (see my current hot (NVME)-warm(RAID0 SSD)-cold storage setup (RAID1HD)), but it would still be of comparative interest as a data point to people who due to budgetary reasons have multiple small SSD drives looking to RAID. Not everyone can afford to NVME their way out of slow data. <- Thats all I meant :-D
 
Steam Mover uses MKLink Windows command. So the game will automatically go to the original saved game location (what ever the speed of that drive is) then get re-directed to the proper file location via MKLink to read/write the data to the new location every time the game pulls or writes data. So you still have in play the original drive as far as I can tell in addition to the new location after you move the files, also a Windows command being done each time (this adds latency in itself and cpu usage) for every file read/write and could give a consistent bottleneck. I am not sure your data totally represents the capabilities of the drives.

Another option would to just change the drive letter being tested to the right location with each drive game directory mapped the same. Meaning your different drives all have the same games in the same directory and you just change the letter designation on the drive to match what Steam will use to run the game. That way the game would always run like it was installed on that drive originally.
 
also a Windows command being done each time
A system call? That's being done every time any file is accessed.
If they all have the same overhead, then wouldn't the difference be negated?

Steam does support multiple library folders, so there isn't a need to use junctions. All you need to do to move a game is to move the game folder and the .acf file then start steam.
 
A system call? That's being done every time any file is accessed.
If they all have the same overhead, then wouldn't the difference be negated?

Steam does support multiple library folders, so there isn't a need to use junctions. All you need to do to move a game is to move the game folder and the .acf file then start steam.
I am not sure, if it is a system pointer in memory of correct file location even though Steam thinks it is the original directory it may not make a difference. If the command has to be run each time or actual location use for the re-direct then there will be some latency involved. Really don't have time to test.

What is .acf file? Where are those?
 
I'm probably just talking out my ass :), although I doubt it would matter for loading those games (most games stick all the data files into a bunch of large archives). For a whole bunch of separate files it could make a noticeable difference?
What is .acf file? Where are those?
In the steamapps folder there's a bunch of appmanifest_*gameid*.acf files, Steam uses them to determine (among other things) which library a game is installed in.
Steam>Settings>Downloads> you can add new library locations there
 
Only question I have is seeing CPU utilization, is loading the game using a single core or multiple cores and how much CPU is being used. I am assuming it is CPU bottleneck but some interesting results all around! Well done arestavo!
 
Steam Mover uses MKLink Windows command. So the game will automatically go to the original saved game location (what ever the speed of that drive is) then get re-directed to the proper file location via MKLink to read/write the data to the new location every time the game pulls or writes data. So you still have in play the original drive as far as I can tell in addition to the new location after you move the files, also a Windows command being done each time (this adds latency in itself and cpu usage) for every file read/write and could give a consistent bottleneck. I am not sure your data totally represents the capabilities of the drives.

Another option would to just change the drive letter being tested to the right location with each drive game directory mapped the same. Meaning your different drives all have the same games in the same directory and you just change the letter designation on the drive to match what Steam will use to run the game. That way the game would always run like it was installed on that drive originally.
A system call? That's being done every time any file is accessed.
If they all have the same overhead, then wouldn't the difference be negated?

Steam does support multiple library folders, so there isn't a need to use junctions. All you need to do to move a game is to move the game folder and the .acf file then start steam.

Fallout 4 1080P (no mods) (Steam + FO4 moved)
Used a save game from one hour thirty minutes into the game. (1920X1080 and all game settings manually set to max)
Timer started when press Enter to Confirm (load) was pressed and stopped when the game loaded in.

Tested with steam install + FO4 moved (no other common folder steam games). Verified that other games could not be opened by trying to open one.

Order of testing: NVME, SSD, RAM, RAID. RAM drive only enabled for RAM drive test.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
30.25
23.34
21.74

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~135GB free)
7.54
7.26
6.9

512GB 850 Evo (empty except for Steam + FO4)
7.33
6.80
6.95

48GB RAM drive (3200MHz DDR4 16-18-18-32)
7.06
6.76
6.89



EDIT: Here's my original test for comparison:

Fallout 4 1080P (no mods)
Used a save game from one hour thirty minutes into the game. (1920X1080 and all game settings manually set to max)
Timer started when press Enter to Confirm (load) was pressed and stopped when the game loaded in.

8 Hard Disk RAID 6 Array (WD Black 5TB 128MB Cache 7200RPM drives) (18.1TB of 27.2TB free)
29.11
22.61
22.91

512GB 950 Pro NVME in PCIE 3.0 X4 slot (with passive heatsink) (boot drive, ~75GB free)
7.65
7.01
6.68

512GB 850 Evo (empty except for FO4)
7.49
6.81
6.67

42GB RAM drive (3200MHz DDR4 16-18-18-32)
6.9
6.54
6.63
 
Last edited:
Only question I have is seeing CPU utilization, is loading the game using a single core or multiple cores and how much CPU is being used. I am assuming it is CPU bottleneck but some interesting results all around! Well done arestavo!

I'm sure they are mostly heavily single threaded. That's the standard anyways.
 
I just find it very interesting how close all the drives are - meaning a bottleneck is some place else keeping the speeds down. DX? Virus software? Network connection speed? The PCIe drives can perform way better then the SATA ones - what is the bottleneck for games?
 
Game developers not wasting money by only targeting HDDs, and to some degree SSDs (and that NTFS thing I mentioned above).
 
I would hazard a guess of NTFS being from 1993 not helping matters.

Nail on the head right there.

I develop cross platform QT applications (same source code and compiler on W7 and Ubuntu 16.04 LTS). Compilation on Ubuntu* is easily 15 times faster then on W7.

NTFS is horrible when it comes to big sets of small files. In fact it is pretty horrible no matter what way you look at it.

Until lots of thumbs get pulled out of lots of asses at MS HQ, and a new modern filesystem is developed, performance will never be stellar. Or even better, they could start supporting OpenZFS.

* Using OpenZFS on ubuntu.
 
I tried to get ReFS working on my 850 EVO to test if there was a difference. Unfortunately it failed to format every time.

I would have loved to test on Ubuntu or Mint as well, but Fallout 4 doesn't have Linux support. And with that game being the only one with meaningful load time differences (that I've found), I'd rather not wipe all my data from my gaming PC to test the few games that would work and likely still show minimal differences.
 
Thanks OP, this post moved me to a 1TB raid 0 M.2 MX300 volume opposed to spending 50% more on a 50% smaller volume that would have provided no real world benefit.
 
Very glad that you found the info helpful in your storage upgrade decision!
 
Definitely made me reconsider wasting money on a 960 EVO M2 1TB when a $240 Scandisk or Micron M2 1TB at 1/2 cost will give virtually the same results. Thanks for all the data.
 
Back
Top