Sign in to follow this  
SAAB340

The Differences Between Hdd, Sshd And Ssd For Photo Scenery

Recommended Posts

We all know that photo scenery takes up vast amount of space. But what are the impacts on photo scenery performance depending on what type of drive we chose to store all those gigabytes of scenery on?

 

I’ve once more looked into that and these are my updated findings:

 

TextureloadTest.gif

 

The first column shows texture loading, which basically is how long time it takes for FSX to load the ground textures to full sharpness, so a lower value means that you can fly faster before your ground textures start to go blurry. The second column shows how long time it takes to load a flight. You can also see the performance at different relevant affinitymasks.

 

The affinitymask also makes a lot of difference (as I’ve previously shown). The results are grouped in to 3 different sections relevant for different CPUs with each section having two different affinitymasks. Depending on what other add-ons you use apart from your photo scenery you might have to keep a vacant core for best performance. That’s the top affinitymask in each section.

 

The top two affinitymasks are relevant to an i5 quad-core without hyperthreading or an i7 quad-core with hyperthreading turned off, where the AM=84 results are equivalent to AM=14 on an i5 and the AM=85 results are equivalent of AM=15 on an i5.

 

The middle two affinitymasks are relevant to an i7 quad-core with hyperthreading active.

 

The bottom two affinitymasks are relevant to a ni7 hexa-core with hyperthreading active.

 

The WD Green is almost always the slowest. Using a WD black instead gives a healthy boost in texture loading. The Velociraptor is indeed the fastest mechanical drive but the texture loading improvement over the WD Black is very minor while the price increase is very, very steep. With the SSD you do pay even more but the texture loading improvement is substantial compared to all mechanical drives, not to mention the stellar load times. The SSHD is a bit of a mixed bag. It’s slower than the WD black when it comes to texture loading so not great to defeat blurries, but at the same time it does help a lot with load times. In the end it is a fairly low end 7200rpm drive with a very small and slow 8GB SSD cache on it.

 

So if you find that you have to fly really slow in order for your ultra-res city not to go blurry and you use an i5 with AM=14 and have the scenery on a WD Green you are absolutely right.

 

The upgrade path for better texture loading would be firstly an i7 quad-core CPU using it with hypertreading active and the appropriate affinitymask. Once you have that its time to go SSD to improve the texture loading. Once you’ve done that you can finally upgrade to an i7-hexa core to get even faster texture loading. Who said that FSX is a cheap hobby? =)

 

The SSD is also absolutely outstanding when it comes to load times

 

Some people might have seen that I’ve already looked into this topic previously. There part of my conclusion was: As you can see, texture loading is affected by the different drives, but only a little. The SSD only option actually performs as when everything is already cached in RAM (remarked as Hot). Load times change more with the different drives. The impact of different drives is also gets greater as we use more cores/threads.”.

But in the end I also stated: “I’ve also noticed that even though the HDDs are capable of almost the same measurable texture loading, in reality the actual texture loading can be inconsistent and a bit irregular when looking out of the window during a flight. It can be as fast as with the SSD, but some times, texture loading lags behind a bit using the HDDs.”

I’ve also seen several reports here where people say they have poor texture loading with photo scenery. 

 

That’s why I decided to re-visit the subject once more with this slightly revised test to see if I could get some data that mimic what I’ve actually been observed flying. The drives tested this time are 4TB WD Green HDD, 2TB WD Black HDD, 1TB Velociraptor HDD, 2TB Seagate SSHD and 1TB Samsung 840EVO SSD.

 

Worth to note is that the main FSX installation is always installed on a 160GB Intel X25-M G2 SSD together with Windows. It’s just the scenery that I’ve had on the different drives.

 

The new test uses almost the same procedure as I’ve used in my previous tests where I quickly slew the aircraft away and then measure how long it takes for the ground textures to sharpen. What I’ve changed is where I conduct the test. It’s no longer done over a few islands of high res photo scenery. It’s now done over LA using Mega Scenery Earth 2.0 and Ultra Res cities San Diego and Los Angeles. The flight is loaded over the Camp Pendleton area just north of San Diego and quickly slewed to overhead the San Marino area of LA. That means that there are miles and miles of high-res photo scenery textures to be loaded around the aircraft in all directions this time. I also use external spot view now. The shown result is an average from 3 different measurements. LOD setting is 9 and I have a minimally tweaked .cfg with just highmemfix, BP=0 and affinitymask added. FPS is locked to 30 in game.

 

As I’ve acquired more and more Mega Scenery Earth 2.0 states I’ve now pretty much filled my 1TB SSD. For the test I have the Horizon UK volume 1-8 photo scenery, Mega Scenery Earth 2.0 states CA, CT, DE, FL, ME, MD, MA, NH, NV, NY, NC, OK, OR, RI, TX, VT, VA, WA, Ultra Res Cities San Diego and Los Angeles photo scenery, and FSGlobal Ultimate Europe and the Americas terrain mesh active in the scenery.cfg and stored on the tested drives. It uses 797GB of space.

 

I also tested what happens if I remove all add-on sceneries from the scenery.cfg apart from the CA, NV, OR, WA, San Diego and LA photo scenery, with and without the FSGlobal Americas terrain mesh.

 

ReducedScenery.gif

 

There is an improvement in texture loading to have less scenery active. Especially with a high access time mechanical HDD. The SSD however shows minimal improvement from deactivating scenery. Using the FSGlobal mesh causes a minimal slowdown of the texture loading.

 

Load times are however significantly improved on all drives by deactivating scenery. The regular HDDs sees the load time cut to a 3rd. But that is still longer than the SSD with everything active. The SSHD and the SSD sees “only” a cut in half but the SSD is still very much superior to any other drive.

 

When I use Process Monitor to see what files that are actually accessed during the test I can see why you see a noticeable texture loading improvement on the harddrives by deselecting scenery. One might think that FSX only accesses data for what’s really close to you. It’s not. FSX actually accesses FSGlobal mesh files as far away as southern Turkey and photo scenery tiles as far as parts of Virginia and North Carolina when flying in California with LOD9. So if you fly in central US it will basically access photo scenery textures from all of your active US states so the HDD arms will get a good workout indeed.

 

As you can see, the storage does affect texture loading quite a bit more than what I previously managed to measure. This improved testing procedure shows results that are in line with what I’ve actually seen as I fly along. The SSHD is as I’ve said before a very mixed bag and produces results that are very variable. It’s been a real challenge to create a testing procedure that measures how it performs during normal FSX use when most data would be loaded from the HDD part of the drive and not from the small 8GB SSD part. The reason the SSHD shorten the load times significantly compared to regular hard drives is that during initial load FSX ‘checks’ all the active scenery files. This checking only reads a small section of each file and these LBAs will be put in to the 8GB SSD cache after you’ve used the SSHD for a few flights. If I repeat my test with the SSHD it will also start to cache the actual scenery accessed during the test on its SSD part. In order for me not to measure texture loading from the SSD part I repeatedly load a flight over the UK in order to fill the rest of the SSD cache with that data before returning to my texture loading test. The SSHD can perform a lot worse as well. If you have something else apart from FSX accessing all files it will start caching LBAs that are not helpful for FSX at all. For example, to run a virus scan. When that happens load times gets a lot longer.

 

This is a chart that shows how much the SSHD can vary in performance.

 

SSHD.gif

 

The top result shows how the SSHD performs when all the wrong sectors are in the SSD cache.

The middle result is what I’ve used for the all the previous charts. Initial load data is in the cache.

The bottom result shows how the SSHD performs when the SSD cache is filled with test data.

 

I hope this information can be of help when you chose what storage to put your photo scenery on.

 

Lars

Share this post


Link to post
Share on other sites
Help AVSIM continue to serve you!
Please donate today!

We all know that photo scenery takes up vast amount of space. But what are the impacts on photo scenery performance depending on what type of drive we chose to store all those gigabytes of scenery on?

 

I’ve once more looked into that and these are my updated findings:

 

TextureloadTest.gif

 

The first column shows texture loading, which basically is how long time it takes for FSX to load the ground textures to full sharpness, so a lower value means that you can fly faster before your ground textures start to go blurry. The second column shows how long time it takes to load a flight. You can also see the performance at different relevant affinitymasks.

 

The affinitymask also makes a lot of difference (as I’ve previously shown). The results are grouped in to 3 different sections relevant for different CPUs with each section having two different affinitymasks. Depending on what other add-ons you use apart from your photo scenery you might have to keep a vacant core for best performance. That’s the top affinitymask in each section.

 

The top two affinitymasks are relevant to an i5 quad-core without hyperthreading or an i7 quad-core with hyperthreading turned off, where the AM=84 results are equivalent to AM=14 on an i5 and the AM=85 results are equivalent of AM=15 on an i5.

 

The middle two affinitymasks are relevant to an i7 quad-core with hyperthreading active.

 

The bottom two affinitymasks are relevant to a ni7 hexa-core with hyperthreading active.

 

The WD Green is almost always the slowest. Using a WD black instead gives a healthy boost in texture loading. The Velociraptor is indeed the fastest mechanical drive but the texture loading improvement over the WD Black is very minor while the price increase is very, very steep. With the SSD you do pay even more but the texture loading improvement is substantial compared to all mechanical drives, not to mention the stellar load times. The SSHD is a bit of a mixed bag. It’s slower than the WD black when it comes to texture loading so not great to defeat blurries, but at the same time it does help a lot with load times. In the end it is a fairly low end 7200rpm drive with a very small and slow 8GB SSD cache on it.

 

So if you find that you have to fly really slow in order for your ultra-res city not to go blurry and you use an i5 with AM=14 and have the scenery on a WD Green you are absolutely right.

 

The upgrade path for better texture loading would be firstly an i7 quad-core CPU using it with hypertreading active and the appropriate affinitymask. Once you have that its time to go SSD to improve the texture loading. Once you’ve done that you can finally upgrade to an i7-hexa core to get even faster texture loading. Who said that FSX is a cheap hobby? =)

 

The SSD is also absolutely outstanding when it comes to load times

 

Some people might have seen that I’ve already looked into this topic previously. There part of my conclusion was: As you can see, texture loading is affected by the different drives, but only a little. The SSD only option actually performs as when everything is already cached in RAM (remarked as Hot). Load times change more with the different drives. The impact of different drives is also gets greater as we use more cores/threads.”.

But in the end I also stated: “I’ve also noticed that even though the HDDs are capable of almost the same measurable texture loading, in reality the actual texture loading can be inconsistent and a bit irregular when looking out of the window during a flight. It can be as fast as with the SSD, but some times, texture loading lags behind a bit using the HDDs.”

I’ve also seen several reports here where people say they have poor texture loading with photo scenery. 

 

That’s why I decided to re-visit the subject once more with this slightly revised test to see if I could get some data that mimic what I’ve actually been observed flying. The drives tested this time are 4TB WD Green HDD, 2TB WD Black HDD, 1TB Velociraptor HDD, 2TB Seagate SSHD and 1TB Samsung 840EVO SSD.

 

Worth to note is that the main FSX installation is always installed on a 160GB Intel X25-M G2 SSD together with Windows. It’s just the scenery that I’ve had on the different drives.

 

The new test uses almost the same procedure as I’ve used in my previous tests where I quickly slew the aircraft away and then measure how long it takes for the ground textures to sharpen. What I’ve changed is where I conduct the test. It’s no longer done over a few islands of high res photo scenery. It’s now done over LA using Mega Scenery Earth 2.0 and Ultra Res cities San Diego and Los Angeles. The flight is loaded over the Camp Pendleton area just north of San Diego and quickly slewed to overhead the San Marino area of LA. That means that there are miles and miles of high-res photo scenery textures to be loaded around the aircraft in all directions this time. I also use external spot view now. The shown result is an average from 3 different measurements. LOD setting is 9 and I have a minimally tweaked .cfg with just highmemfix, BP=0 and affinitymask added. FPS is locked to 30 in game.

 

As I’ve acquired more and more Mega Scenery Earth 2.0 states I’ve now pretty much filled my 1TB SSD. For the test I have the Horizon UK volume 1-8 photo scenery, Mega Scenery Earth 2.0 states CA, CT, DE, FL, ME, MD, MA, NH, NV, NY, NC, OK, OR, RI, TX, VT, VA, WA, Ultra Res Cities San Diego and Los Angeles photo scenery, and FSGlobal Ultimate Europe and the Americas terrain mesh active in the scenery.cfg and stored on the tested drives. It uses 797GB of space.

 

I also tested what happens if I remove all add-on sceneries from the scenery.cfg apart from the CA, NV, OR, WA, San Diego and LA photo scenery, with and without the FSGlobal Americas terrain mesh.

 

ReducedScenery.gif

 

There is an improvement in texture loading to have less scenery active. Especially with a high access time mechanical HDD. The SSD however shows minimal improvement from deactivating scenery. Using the FSGlobal mesh causes a minimal slowdown of the texture loading.

 

Load times are however significantly improved on all drives by deactivating scenery. The regular HDDs sees the load time cut to a 3rd. But that is still longer than the SSD with everything active. The SSHD and the SSD sees “only” a cut in half but the SSD is still very much superior to any other drive.

 

When I use Process Monitor to see what files that are actually accessed during the test I can see why you see a noticeable texture loading improvement on the harddrives by deselecting scenery. One might think that FSX only accesses data for what’s really close to you. It’s not. FSX actually accesses FSGlobal mesh files as far away as southern Turkey and photo scenery tiles as far as parts of Virginia and North Carolina when flying in California with LOD9. So if you fly in central US it will basically access photo scenery textures from all of your active US states so the HDD arms will get a good workout indeed.

 

As you can see, the storage does affect texture loading quite a bit more than what I previously managed to measure. This improved testing procedure shows results that are in line with what I’ve actually seen as I fly along. The SSHD is as I’ve said before a very mixed bag and produces results that are very variable. It’s been a real challenge to create a testing procedure that measures how it performs during normal FSX use when most data would be loaded from the HDD part of the drive and not from the small 8GB SSD part. The reason the SSHD shorten the load times significantly compared to regular hard drives is that during initial load FSX ‘checks’ all the active scenery files. This checking only reads a small section of each file and these LBAs will be put in to the 8GB SSD cache after you’ve used the SSHD for a few flights. If I repeat my test with the SSHD it will also start to cache the actual scenery accessed during the test on its SSD part. In order for me not to measure texture loading from the SSD part I repeatedly load a flight over the UK in order to fill the rest of the SSD cache with that data before returning to my texture loading test. The SSHD can perform a lot worse as well. If you have something else apart from FSX accessing all files it will start caching LBAs that are not helpful for FSX at all. For example, to run a virus scan. When that happens load times gets a lot longer.

 

This is a chart that shows how much the SSHD can vary in performance.

 

SSHD.gif

 

The top result shows how the SSHD performs when all the wrong sectors are in the SSD cache.

The middle result is what I’ve used for the all the previous charts. Initial load data is in the cache.

The bottom result shows how the SSHD performs when the SSD cache is filled with test data.

 

I hope this information can be of help when you chose what storage to put your photo scenery on.

 

Lars

 

Congratulations for this excelent post, SAAB340. I think it should be important to describe the methods, how you came to these results. There´s a software called HDTune PRO, which has a "Disk Monitor" function. This function measure ALL access to all disks on the system, or the access ( load ) on a particular disk. When using FSX or Prepar3Dv2  I run HDTune, on the disk which contains MSE v2 and for my surprise, the data loading is minimal, here and there, a peak no more than 2Mb, I can show some pictures later today. Perhaps what you´re talking about has more to be with access time than to bandwidth from HDD versus SSD. On normal use, with Affinity mask 84 and others you suggested, i didn´t notice much difference between HDD or SSD, sometimes, I think that processor is loading more, but FSX performance is worst. This is a good theme for discussion.

Share this post


Link to post
Share on other sites

Congratulations for this excelent post, SAAB340. I think it should be important to describe the methods, how you came to these results. There´s a software called HDTune PRO, which has a "Disk Monitor" function. This function measure ALL access to all disks on the system, or the access ( load ) on a particular disk. When using FSX or Prepar3Dv2  I run HDTune, on the disk which contains MSE v2 and for my surprise, the data loading is minimal, here and there, a peak no more than 2Mb, I can show some pictures later today. Perhaps what you´re talking about has more to be with access time than to bandwidth from HDD versus SSD. On normal use, with Affinity mask 84 and others you suggested, i didn´t notice much difference between HDD or SSD, sometimes, I think that processor is loading more, but FSX performance is worst. This is a good theme for discussion.

The test was conducted by first of all rebooting the computer, then loading FSX and loading the saved test flight. I time with a stopwatch from pressing "fly now" until the flight is loaded. That's how I come up with the load time.

I have a macro programmed on my HOTAS Cougar joystick so that by pressing just one button the aircraft is automatically slewed away at maximum slew speed and stops in the same place 7 seconds after I pressed the button. There are delays built in so the actual slew is only a few seconds. I time with a stopwatch from pressing the button until all the ground textures have sharpened fully and subtract 7. That's how I measure the texture loading.

The most important thing is to then reboot the computer before running yet another test as Windows keeps all that data cached in RAM and if you don't reboot there will be almost no data read from the disk on subsequent tests and the texture loading will be as quick as the SSD result. The SSD texture loading is as quick as when all the data is cached in RAM.

 

You are right that its all about access time and not bandwidth. The SSD let us be fully CPU limited while the regular drives will be bottlenecking the CPU.

 

The test rig is a Intel i7 3930K @ 4.3Ghz with 8GB 2133Mhz CL9 RAM.

 

I have also monitored with Process Explorer, Process Monitor and DiskMon from Windows Sysinternal. I am by no means any expert on how Windows handles its file system and disk access. Process Explorer says there has been 11.2 GB of disk reads (from all disks) by FSX once the test finishes but I believe that includes reads that remained cached by Windows. Process Explorer gives tones and tones of data you can filter through and will show exactly what files and what part of that file that were accessed. For example, during the initial part of the loading of the flight when FSX 'checks' every active scenery file it looks like it reads the first 4KB of every file, this first 4KB in each file is usually accessed several times during the flight when other parts of the file is needed. So it makes sense to get that part pre-loaded in to FSX for quick access. It also loads the last part of each file and sometimes a middle part. These accesses are a lot of the time also 4KB but they can also be a lot longer. This pre-loading would happen fight at the start the simulator if you don't have disablepreload=1 in the cfg but it renders the UI almost unresponsive during the pre-load and it will take several minutes on regular HDDs but is a lot quicker on the SSD. Thats why we postpone it to the loading of the flight instead. I did write in my previous topic that you won't have to disable pre-load if you use an SSD and therefore can enjoy even quicker load times. That's not true when you fill it with even more scenery though. You have to use disable preload on an SSD as well but you can have a lot more add-ons compared to a regular HDD before it becomes necessary.

With DiskMon I believe I see only the actual disk reads and read sizes. During the pre-load I can see actual reads between 0.5 and 432KB in size on the scenery disk alone, but the absolute majority of reads are either 4KB (62%) or 188KB (29%) totalling a whooping 4GB of reads. A lot of this data will be re-used during a flight.

Then it comes to the textureloading part of the test, so what's loaded during a flight, the DiskMon shows scenery disk access sizes ranging between 0.5 and 428KB. Still with 62% being 4KB but other notable sizes are 8KB (1%), 12K (3%), 16KB(3%), 20KB (3%), 24KB(2%), 28KB (2%), 32KB(1%), 116KB(2%), 120KB(3%) and 124KB (8%). All in all 0.5 GB of disk access.

That shows why access time and not bandwidth is so important for the load time as most disk reads are very small so the actual transfer time is only a fraction of the access time.

When you did your disk read measurements had you already flown where you were flying since your last reboot causing the data to be read from the RAM and not the disk?

Share this post


Link to post
Share on other sites

Hi, SAAB340, we still have to consider the size of the textures on loading. 

 

Here, the differences between SSD Samsung 84 PRO - 256Gb and Samsung 84 EVO 512 

 

14751817922_8a0e84f371_o.jpg

 

 

14565685467_42e50e1139_o.jpg

 

Here, FSX running a MSE v2 scenery while monitoring HDD drive where the scenery is located. This scenery was never used before, so, I believe there´sno caching yet:

 

14771998973_d9c5f5e919_b.jpg

 

 

 

If you can, please test with SAMSUNG 840 PRO 256Gb, it looks 10 times faster on loading small files

Share this post


Link to post
Share on other sites

If you can, please test with SAMSUNG 840 PRO 256Gb, it looks 10 times faster on loading small files

Sorry, but I can't.

Firstly because the scenery I use for the test is 797GB so it wont fit, secondly because I don't have any 840Pro drive and thirdly because I rather fly than test for a while now. =) The SSHD testing alone took about 20 hours to complete.

 

I don't expect a faster or slower SSD to make any difference. The most important thing is to use a SSD as its fast access time makes the CPU the bottleneck again. Pretty much any SSD will do. The 'SSD' part in the SSHD drive is a way worse performer than any SSD on the market. It uses a low performing controller and just two channels to its 2 die 8GB NAND package. But at the best case (when the data is stored in the SSD part) it still give almost as fast texture loading as the way better performing 840EVO.

Share this post


Link to post
Share on other sites

Sorry, but I can't.

Firstly because the scenery I use for the test is 797GB so it wont fit, secondly because I don't have any 840Pro drive and thirdly because I rather fly than test for a while now. =) The SSHD testing alone took about 20 hours to complete.

 

I don't expect a faster or slower SSD to make any difference. The most important thing is to use a SSD as its fast access time makes the CPU the bottleneck again. Pretty much any SSD will do. The 'SSD' part in the SSHD drive is a way worse performer than any SSD on the market. It uses a low performing controller and just two channels to its 2 die 8GB NAND package. But at the best case (when the data is stored in the SSD part) it still give almost as fast texture loading as the way better performing 840EVO.

 

Well, then we got a conclusion of all of this, as you pointed that: "Pretty much and SSD will do". Even as the 840 EVO is not as fast as the 840 PRO, in other softwares, I notice a great improvement overall and no comparison to any HDD. Thanks for the testings. We have plenty of data to analise yet...

Share this post


Link to post
Share on other sites

On a non-OC I5, the bottleneck for PR textures is still the CPU by far (have compared SSD to regular drive). Haven't tested an i7 yet. I have used WD Blacks and Samsung Evo's, the Evo is faster at loading the scenery, but once flying it's all about caching and the CPU gets behind regardless.

 

CPU main thing to upgrade first, then video card, SSD last as it makes the least difference of the 3.

Share this post


Link to post
Share on other sites

On a non-OC I5, the bottleneck for PR textures is still the CPU by far (have compared SSD to regular drive). Haven't tested an i7 yet. I have used WD Blacks and Samsung Evo's, the Evo is faster at loading the scenery, but once flying it's all about caching and the CPU gets behind regardless.

 

CPU main thing to upgrade first, then video card, SSD last as it makes the least difference of the 3.

 

Even on my i7-2600k the bottleneck is the CPU for FSX, all I need to do is run a Full New York City Scenery and look that core 1 is 100% and i can´t achieve 30fps under these conditions. All other cores are far less. Perhaps I would need a 10Ghz CPU, ajajajaj

Share this post


Link to post
Share on other sites

Single core performance is obviously the key for FPS as only one thread is responsible for the FPS.  When you get close to not being able to maintain the locked FPS you have set, the FiberFrameTimeFraction value comes in to play. It decides whether to sacrifice texture loading or FPS. If you have a low FFTF value set, you sacrifice texture loading for FPS, and if you have a FFTF value up to 0.66 you sacrifice FPS but maintain texture loading. FFTF above 0.66 start to give you both lower texture loading and lower FPS.

 

With the affinitymask you tell where the main thread and texture loaders end up on the CPU. FSX also has several other threads that constantly get moved around on different cores as windows finds it best. There are 3 other threads that produce any significant load. One is a sound thread, another a debug thread, and the third one that can produce a lot of CPU load is a thread that is responsible for almost all disk access, it could also possible load the textures for aircraft and custom buildings, or it just loads the files off the storage and pass them on to the main thread to load just as it does with the texture loaders. The load on this thread is very bursty and if you have a lot of add-ons where you fly, the load for this thread can get very, very high. That's when you really need to leave a vacant core for it with the affinitymask, as it otherwise can get stalled by a texture loader or the main thread reducing the texture loading as a result. Neither of the texture loaders are touching any files on the disks ever.

 

The main thread itself is responsible for loading only a few type of files of the disk. For example sound files and airport vehicles. That's why the loading of a sound file with a high access time disk can cause stutters just before a sound is to play.

Share this post


Link to post
Share on other sites

Single core performance is obviously the key for FPS as only one thread is responsible for the FPS.  When you get close to not being able to maintain the locked FPS you have set, the FiberFrameTimeFraction value comes in to play. It decides whether to sacrifice texture loading or FPS. If you have a low FFTF value set, you sacrifice texture loading for FPS, and if you have a FFTF value up to 0.66 you sacrifice FPS but maintain texture loading. FFTF above 0.66 start to give you both lower texture loading and lower FPS.

 

With the affinitymask you tell where the main thread and texture loaders end up on the CPU. FSX also has several other threads that constantly get moved around on different cores as windows finds it best. There are 3 other threads that produce any significant load. One is a sound thread, another a debug thread, and the third one that can produce a lot of CPU load is a thread that is responsible for almost all disk access, it could also possible load the textures for aircraft and custom buildings, or it just loads the files off the storage and pass them on to the main thread to load just as it does with the texture loaders. The load on this thread is very bursty and if you have a lot of add-ons where you fly, the load for this thread can get very, very high. That's when you really need to leave a vacant core for it with the affinitymask, as it otherwise can get stalled by a texture loader or the main thread reducing the texture loading as a result. Neither of the texture loaders are touching any files on the disks ever.

 

The main thread itself is responsible for loading only a few type of files of the disk. For example sound files and airport vehicles. That's why the loading of a sound file with a high access time disk can cause stutters just before a sound is to play.

 

Thank you for this nice explanation about how it works, SAAB340. 

Share this post


Link to post
Share on other sites

I've made one final round of testing at a lower LOD setting just to see how it compares to the maybe somewhat extreme 9.0 LOD that I use with my photo scenery. Using a high LOD together with complex aircrafts is normally to ask for OOMs. I only used the SSD and Velociraptor this time as I've already returned the other disks to their normal usage, but we already know that the WD Black is only slightly slower then the Velociraptor. I also added affinitymasks equivalent to using the Hexa-core with HyperThreading off.

 

This is what the textureloading looks like when you compare LOD=5.5 to LOD=9.0.

LODchange.gif?dl=0

 

 

To start with I see a big difference in sharpness of the fully loaded view between 5.5 and 9.0 but the textureloading at the lower LOD is indeed a lot quicker as there's way less textures to load.

 

Using a hexa-core AM=4084 LOD=9 gives slightly better texture loading than quad-core HToff AM=14 LOD=5.5.

 

The SSD still improves texture loading compared to the HDD in the LOD=5.5 case.

 

Using more cores and/or SSDs will either let us fly faster without getting blurries, or lets us increase the LOD without getting blurries at the same speed. Watch out for OOMs though.

 

I better go and get on with renovating my kitchen now as my wife is ready to throw my FSX computer out of the window...

Share this post


Link to post
Share on other sites

I've now updated my post above with LOD=4.5 and LOD=6.5 as well.

 

Some conclusions I can make is that an SSD does help a lot with texture loading even at LOD=4.5.

 

LOD=4.5 on either type of storage is indeed a lot faster than higher LOD settings, especially with only a few texture loaders. It might be that you find it fast enough already, in that case having your photo scenery on an SSD is just a massive expense apart from initial load time. Activating hyperthreading and using AM=244 still yields a large textureloading improvement with a HDD compared to AM=84/14. Given that I find LOD=9.0 on an SSD with Affinitymask=4089 just good enough in the area where I've done the test but I'm still wanting more I'd expect people using LOD=4.5 and AM=84/14 to feel the same given that the texture loading I get is the same when using SSD in both cases.

 

Not to mention that the LOD=9.0 view looks soo much better on the screen compared to LOD 4.5. One way not to see the blurry ground textures at lower LOD levels is to zoom out. It's the same effect as how looking at a low resolution image on a small, low-res screen can look great (early mobile phones with cameras comes to mind) while it looks horribly blurry if you try to print it. Zooming out is just cheating. If you're happy to look at the printed image at long distance that's fine but personally I prefer the higher resolution cameras of later mobile phones (higher LOD setting).

 

We can also see how AM=85/15 vs AM=244 is very similar with the HDD but if you use SSD the 3 texture loaders, each with their own physical core (AM=85/15), gets a higher boost in texture loading compared to Hyperthreading 4 textureloaders together (AM=244) on just 2 physical cores.

 

As bump up the LOD and add more texture loaders you can see that the advantage of using SSD instead of HDD grows larger and larger.

 

I also tested with the reduced scenery that I used earlier at the extremes of LOD4.5 and LOD9.0 @ AM=84/14 and AM=4089 and saw the same small improvements as before apart from when using an SSD with only two texture loaders (AM84/14) were the improvement is basically non existent @ LOD9 and not measurable @ LOD4.5. An SSD really doesn't care if you fill it up with loads and loads of active photo scenery when just using just two texture loaders. Sure your initial load time still takes a hit when you fill it, but far from what using a HDD would give you anyway.

Share this post


Link to post
Share on other sites
I've just updated the 840EVO numbers in the image of the original post as my 840EVO disk, unknown to me at the time of testing, was already suffering from the performance degradation that plauges the 840EVO disks. I've just posted my experience of how this degradation manifests itself in FSX here. Basically, it makes loading a flight take longer and when it gets really bad it also makes texture loading slower. (It wasn't really bad during my previous testing.) The numbers now reflect how the 840EVO performs at full performance when it's not degraded.

 

I've also added 850PRO numbers as well. As I'd already copied over all the FSX scenery data on to it as a backup when I was restoring my 840EVO performance I couldn't resist testing it.

 

TextureloadTest.gif?dl=0

 

As you can see, the 840EVO and the 850PRO perform exactly the same. As will pretty much any other SSD. Any modern fully performing SSD will fully move the bottleneck away from the storage in FSX.

 

I also took the chance and measured what happens when I connect the different SSDs to the different types of SATA ports I have available on my motherboard. I have SATAII and SATAIII off the X79 chipset and SATAIII off an ASM1061 controller.

 

SATAII%20and%20III.gif?dl=0

 

Using the slower ports only affects the load time of the flight by a few seconds. Textureloading remains the same. The native SATAIII ports off the X79 chipset were unsurprisingly the fastest followed by the SATAIII ports of the ASM1061 controller and the slowest were the native SATAII ports off the X79 chipset.

 

It's actually only the initial part of the first load time, when FSX is accessing all the active scenery files that I've mentioned about in a comment above, that is affected. So if the drive has less active scenery on it (my 1TB drive is filled with active scenery), the slowdown between the different SATA ports will be smaller. It doesn't matter what AffinityMask or LOD setting you're using, the slowdown will still be the same.

 

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this