A further look at HDDs and SSDs

[SAAB340 28]

I’m long overdue posting this. I’ve had a further look in to HDD and SSD performance in FSX. You can read my earlier post about it here. There I also have the full explanation of what my spreadsheets show.

 

Since then I’ve upgraded my system to a i7 3930K SandyBridge-E. I’ve also acquired a GTX780 and a 1TB Samsung 840 EVO SSD.

 

My choice of the 3930K over a Haswell system was based on the fact that my personal user scenario is high-res photo scenery at high LOD, and that’s where many cores and HyperThreading makes a change in FSX.

 

Firstly I wanted to see the differences between my old i7 860 Lynnfield and my SandyBridge-E. With the new system being capable of running my RAM a lot faster then my old system I initially had a look at how big changes in RAM speed impact FSX and this is what I came up with.

 

As you can RAM doesn’t really change texture loading and load time. But it can impact FPS and stutters quite a lot. No one will however run it at 800MHz so it’s just a reference point. 1333 MHz DDR3 RAM should be a thing of the past, but it’s still available as the cheapest budget option, and some people might still hang on to it. Be warned, and please stay away.

 

I’ve also compared texture loading and load time at the same clock speed (3.8GHz) on the Lynnfield Vs the SB-E.

 

The architecture alone is providing a small 5-10% improvement at the same clock speed. Add to that the higher available overclock and utilise the 2 extra cores on the SB-E (bottom result) and there is well over 50% improvement in both texture loading and load times.

 

Here is how well FSX scale with adding more cores.

 

The load time prefers extra physical cores, and is only slightly helped by HyperThreading, whilst texture loading is helped quite a bit by HyperThreading as well as extra physical cores. Even though the scaling isn't perfect, FSX continues to scale past 4 physical cores the way its been coded to use multi threading past 3 cores. (Improve on texture loading and load time, not FPS.)

 

Now on to the different storage drives.

I’ve tested having both OS and FSX installed on my 160GB Intel X25M G2 SSD, and having only the photo scenery installed on the different HDDs. The photo scenery in question is only a 8GB large part of the UK with 50cm textures and a 5m terrain mesh.

 

The HDDs used are:

A 250GB WD SE16 from 2005. A fairly good performer back when I got it. a very slow drive now by todays standards.

A slightly newer 250GB WD Blue. It’s still an old drive and I think it’s from the first line of the WD blue product line.

A 150GB Raptor. The last of the Raptor line before it turned Velociraptor

A 2TB WD Black of latest model.

A 1TB Velociraptor. The fastest HDD at this moment.

 

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. Thing to remember though is that the main FSX installation is allways installed on the SSD with only the "small" 8GB photo scenery installation on the different HDDs. Had FSX and the scenery both been installed on the same HDD, with only the OS on the SSD, we would have seen an even larger performance difference between the drives.

 

I did these tests shortly after I acquired the SB-E. Since I’ve used my Velociraptor and filled it with more and more photo scenery. As the scenery collection has grown in size, so has the time it takes to load a flight. Once you’ve installed a lot of scenery on a HDD, you pretty much have to use DisablePreload=1 in the .cfg, as FSX becomes very unresponsive without that entry for a long while after you launch the simulator as FSX is checking all the active scenery files directly after innitial launch. With Preload disabled, FSX is responsive when setting up a flight, as it postpones checking the scenery files until you load the first flight after launch, adding a lot of time on this initial load time though. I now have the whole of Britain in high res + a few MegaScenery Earth states as well, so the photo scenery alone totals around 367GB resulting in the initial load time to be several minutes.

 

The increasingly longer initial load time is why I decided to go back to a SSD only setup for FSX. I purchased a 1TB Samsung 840EVO SSD and off course I had to see what difference it made. I tested connecting the SSD with both SATAIII and SATAII and also compared it to my WD Black and Velociraptor. This is how much impact that has on the load times. (This flight is not the same as in the previous spreadsheet so it can’t be compared with the previous numbers. But on this spreadsheet they are all the same. It's loading a flight just east of Edinburgh having loads of high res photo scenery tiles all around it.) On the results without DisablePreload=1 in the cfg., I have waited until the checking of the scenery files is complete before timing the results.

 

Well, there is quite a difference in how long time it takes to load the same flight. Using SATA II vs SATA III on the SSD is having a very small but measurable impact, but the different drives have a lot more impact (The HDDs are both on SATA III). Especially when it comes to the DisablePreload=1 results at the bottom part. There, the SSD is several minutes faster than the HDDs. But even loading new flights after all the scenery files has been checked (top part of the spreadsheet) can be a minute faster to load with the Samsung SSD compared to the WD Black. I’d expect the load time using the HDDs to increase as the drives are filled up with even more scenery, whilst I expect the SSD to maintain its speed. Remember that 367GB is only 1/3^rd full on the Velociraptor, and 1/6^th full on the WD Black.

 

I now use the SSD without DisablePreload=1 and enjoy being in the cockpit ready for a flight 3 minutes quicker than I was using the Velociraptor where I had to use DisablePreload=1. The upgrade had the effect I was hoping for. If only we had overclockable CPUs with even more cores available to decrease load times even more...

 

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. Probably as HDDs have a large range of response times for different requests as it access files from different locations on the drive. Actual texture loading is consistently fast using the SSD.

 

Hopefully this information can be useful for some of you as you make decisions of what storage to use. I’m not going to check FPS this time as I’ve already shown that SSDs make no difference to that. It would have been interesting to measure a bit more about stutters but that would require several more days, if not weeks, of testing to get reliable results. I simply don’t have time for that, sorry. I’ve already previously established that using a high access time HDDs does cause stutters, regardless if you have FSX, the OS, or even worse, both on it.

 

SSDs continue to fall in price, and they are now cheaper per GB than they have ever been. This will continue. If you can afford SSD only for FSX it’s great, but not a game changer. Remember that it's safe to keep FSX and OS on the same SSD without performance penelties. I thought I went down a good route when I acquired the 1TB Velociraptor for photo scenery. Yes, it’s still cheaper than the SSD, but with hind sight, it was probably not the best move. Yes it’s better than the WD black, but also a lot pricier. Paying double the price for an SSD instead (it doesn’t really matter what SSD you use as long as it’s an SSD) does however give you a lot more notable performance difference. And I'd dare to use use the space on the SSD a lot more than I’d dare to fill up the Velociraptor when it’s used for FSX scenery exclusively as its read only. In the end, it all depends on your own budget. As far as hybrid drives are concerned I’m positive, but sceptical of todays implementations. If a lot of NAND is paired with a fast HDD it would be to prefer over a regular fast HDD. But the current implementations are still a very small NAND bit, paired with a slow HDD. I’d stay away from that as they will perform a lot more like a slow HDD than a SSD. This might change in the future though.

[Appliance 51]

 

 

large range of response times for different requests as it access files from different locations on the drive.

 

Surprised here.  If you load photoscenery west to east, reboot after each instal, defrag then instal next etc. etc. there should be no stutter, delay in loading, within the VC or without.  I understand completely if installs are haphazard Ie spread across the disc then stutters caused by disc access delays are inevitable, but they needn't be.

 

Once again most valuable research SAAB340.

 

I disagree however with your hybrid comments. I have two Seagate 1tb SSHD hybrids and they load my FS9 install very fast indeed ... maybe not as fast as an SSD but monsterously faster than a HDD.  Certainly worth it to me.

[SAAB340 28]

Surprised here. If you load photoscenery west to east, reboot after each instal, defrag then instal next etc. etc. there should be no stutter, delay in loading, within the VC or without. I understand completely if installs are haphazard Ie spread across the disc then stutters caused by disc access delays are inevitable, but they needn't be.

 

Once again most valuable research SAAB340.

 

I disagree however with your hybrid comments. I have two Seagate 1tb SSHD hybrids and they load my FS9 install very fast indeed ... maybe not as fast as an SSD but monsterously faster than a HDD. Certainly worth it to me.

I'm glad you liked it.

 

High res photo scenery is taking up loads of space even for a small area. Especially when its combined with a high res mesh as well. The 3 main areas of my Scottish scenery are around 50GB each. I can be flying over the south part, see through the whole central part and in to the north part. That's even at fairly low altitude. That means that the HDD will be accessing sectors that are at least 50GB apart. Most likely double that. Mega scenery earth have a few very large states as well. Texas is way over 100GB. So as I fly along the border to neighbouring states the HDD will have to move its mechanical parts quite a bit, regardless of how well things have been defragged.

In the end photo scenery will be installed in sub optimal locations on the drive as well when new regions are acquired, as I doubt any one will re-install hundreds of gigabytes of scenery for hours and hours, every time they purchase a new region, just to have everything optimally placed on their drive.

 

Just to clarify, I don't see stutters from using the HDDs for my scenery, just irregular/worse texture loading compared to the SSD's consistent texture loading. I expected not to see much difference given how close my measured values are. But my texture loading test uses only a small scenery install isolated to a few small islands. So the mechanical movement of the HDD arms will be very minor

 

Interesting what you said about the SSHD. I wonder if they help with the initial loading where the simulator checks the scenery files? Could it bee that the sectors it accesses during the check are few enough to be cached in flash, but normally very spread out on the disk? Does the SSHD help once the scenery has been checked initially though, or is It just helpful for the first load?

[Ted Striker 338]

Very interesting and informative post SAAB340. Thanks for taking the time to document and post your findings. I always enjoy reading your reports. My new system is all SSD drives. The quicker load times and not having to defrag anymore was worth the extra cost for me.

 

Ted.

[SolRayz 392]

Never having to defrag is enough to never touch mechanical drives ever again.