My take on SSD Performance.

[Swivelguy2]

Never used eclipse, but Dragon age initial load is identical to 7200rpm and each ingame loads are slightly quicker. Other games like, L4D, Demigod, CODMW2, and Batman, all run the same. I was expecting much faster load than this.

Perfect example would be like going from 5400rpm to 7200rpm.

You're probably observing the effect of Superfetch. You're launching those games from RAM with either hard disk, whether you know it or not.

 

[garritynet]

Don't forget the obvious performance benefit of not having your boot drive suddenly die on you one day. I personally have never had a drive die on me. Ever. But I know many who have.

 

[bryanW1995]

Going from a 7200RPM drive to a 10,000RPM Raptor is very noticable to you but going from a Raptor to an SSD is not?

yes, i call shens on that.

Yeah, but the average user won't notice the difference if he compared a 7200RPM drive with a Raptor either.

exactly. if he noticed a sizeable difference going from a notmal hd to a raptor then he should have seen a much greater difference when going to an ssd.

 

[dawza]

yes, i call shens on that.



exactly. if he noticed a sizeable difference going from a notmal hd to a raptor then he should have seen a much greater difference when going to an ssd.

I am not sure what shens there are to call. I am simply reporting my experience. I have performed fresh Vista 64 installs on nearly identical systems twice, both times initially with 7.2K RPM HDDs, and the second time with VRs. I also have a third similar system that is still on a 7.2K RPM, so I am constantly able to compare. These are all C2D systems with the same OS and > 4GB RAM.

To add to this, I have had two Vertex SSDs. I sold the first one because after running two fresh OS installs for several days on my two VR systems, and even letting my co-workers run blind tests, neither I nor they could discern anything for normal usage. I purchased another Vertex to use in a fanless build, which I also use on a normal basis.

One thing I never understood is why there is a general trend on this forum to discount anyone who does not bow down to the mighty SSD.

 

[PUN]

I am not sure what shens there are to call. I am simply reporting my experience. I have performed fresh Vista 64 installs on nearly identical systems twice, both times initially with 7.2K RPM HDDs, and the second time with VRs. I also have a third similar system that is still on a 7.2K RPM, so I am constantly able to compare. These are all C2D systems with the same OS and > 4GB RAM.

To add to this, I have had two Vertex SSDs. I sold the first one because after running two fresh OS installs for several days on my two VR systems, and even letting my co-workers run blind tests, neither I nor they could discern anything for normal usage. I purchased another Vertex to use in a fanless build, which I also use on a normal basis.

One thing I never understood is why there is a general trend on this forum to discount anyone who does not bow down to the mighty SSD.

Because alot of them has no clue of what they are talking about. Most don't even own the right hardwares, and they are one sided. I had originated this thread to inform average users like myself about little performance gain they'd see.

Some members had given an excellent input on what SSD is good for, which I'd appreciate.
The reason you are seeing more speed diff from 7.2 to 10k is because, for your usuage, performance is faded from 10k to SSD.

 

[bryanW1995]

so you performed the exact same blind test comparing 7200 rpm drives to the raptor that you did comparing the raptor to the ocz drives and concluded that there was a bigger difference between the raptor/7200 than there was between the ssd/raptor??? sorry, but that seems highly unlikely.

did you actually use a stopwatch? did you repeat the tests 5 times and take the middle number? the average number? or did you just "go with the gut"? I want to give you the benefit of the doubt here, but, again, if you saw a significant performance increase by doing absolutely nothing other than swapping a 7200 drive for a raptor then you should have seen a far larger change going from a raptor to a decent ssd.


there are 2 possible scenarios, however:

1. crappy ssd, samsung controllers
2. 7200 rpm drive was not working properly or was extremely old and just plain sucked to begin with. as I'm sure you know, not all 7200 rpm drives are created equal, my wd6400aaks is much faster than my old 36gb raptor for example.

 

[Yuriman]

SSD performance is exactly what it is. They are generally similar in write speeds to the fastest mechanical drives, several times faster in sequential reads, and 50-80x faster in small random reads and writes.

Windows has been heavily optimized to make up for the shortcomings of mechanical drives with features such as superfetch and search indexes. They are what they are. Without them, an SSD would be unquestionably many, many times faster than a mechanical drive.

If you are not seeing a performance improvement, you are simply not using your hard drive very much in ways which cannot be predicted and worked around. In all cases where you use the drive heavily, you will see a performance improvement. Otherwise, you are comparing loading a program from ram against loading a program from ram.

 

[dawza]

so you performed the exact same blind test comparing 7200 rpm drives to the raptor that you did comparing the raptor to the ocz drives and concluded that there was a bigger difference between the raptor/7200 than there was between the ssd/raptor??? sorry, but that seems highly unlikely.

did you actually use a stopwatch? did you repeat the tests 5 times and take the middle number? the average number? or did you just "go with the gut"? I want to give you the benefit of the doubt here, but, again, if you saw a significant performance increase by doing absolutely nothing other than swapping a 7200 drive for a raptor then you should have seen a far larger change going from a raptor to a decent ssd.


there are 2 possible scenarios, however:

1. crappy ssd, samsung controllers
2. 7200 rpm drive was not working properly or was extremely old and just plain sucked to begin with. as I'm sure you know, not all 7200 rpm drives are created equal, my wd6400aaks is much faster than my old 36gb raptor for example.

I understand quite well how to conduct a scientific experiment, thank you very much, and specifically chose not to, because review sites have covered that thoroughly and better than I would ever care to.

The answer to your first question is already in my posts, and to answer your second- these were brand-new WD 6400AAKS HDDs.

I propose a third possible scenario, which Pun and Yuriman's posts address- my usage patterns do not allow an SSD to really stretch its legs. I am not trying to argue about RAM caching or indexing, or about whether SSDs are superior. All I am trying to say is that my experience has been that the Vertex is not, in most scenarios (again, for me), noticeably faster than the VR. My regular usage is browsing, email, office apps, and some gaming. I leave it up to the readers to decide whether it would be proper for them to extrapolate based upon my opinions. Is the SSD faster if I try to open several apps at a time while archiving files and installing a program? Absolutely. Do I actually sit in front of my computer and do this on a day-to-day basis and/or consider this normal desktop usage (for me)? No. Will I buy an Intel/Indilinx SSD for my workstations so that I could do this on a day-to-day basis, just for bragging rights? Probably.

 

[AlgaeEater]

I'm making sure the older post related to the same exact subject got its mention again:

http://forums.anandtech.com/showthread.php?t=2023939

Summarizing that post and this one; I'll again say that those of us who aren't really impressed are the ones who had good hardware to begin with.

Everyone in both threads seem to be throwing out names like "Raptors, WD Black/Blue, Raid, SCSI" like it's commonplace. You can see why the performance increase was not as realized to be night and day.

To the average PC user who buys their computer pre-built from a store or someone trying to revive a 3 year old laptop... SSD will be the second coming for them. But for the rest of us, it's just another step on top of a already tall ladder.

Besides since all of us here in this thread pretty much own an SSD, I can safely assume our pockets are deep enough and our minds that detached from practicality for it to be a justifed purchase regardless of the case.

 

[skid00skid00]

There is ZERO lag with WD Black drives as well. I could open over 300 IE browsers without any lags as they are not HD dependent once launched. It's sad that most people have misconception regarding this..

That's absolutely not true for my overclocked C2D. With the HD's, ONE AV running slowed my browser down-alot.

 

[GrumpyMan]

Very good points on both sides of the argument in this thread. For me and most of the members here, do I need a SSD? No. Will it be the second coming in terms of daily performance? Probably not. Am I going to get one as soon as I can afford it? Oh yes, of course I will.
Most of us are hardware enthusiasts after all. It's what this whole web site is built on.

 

[zuffy]

I have 5 Intel SSDs, including home and work. I experience both side of the arugments. All the SSDs are replacing VelociRaptor drives in RAID 0 or single configuration so yeah the performance gain is not as great as some folks are seeing from their experiences. I noticed the speed advantage with my work PCs more because I tend to do more I/O's while at home, it's idling more. If I didn't get my SSDs for free I wouldn't go out and spend $500 on a 160GB SSD. I would wait for the price to drop.

 

[AyashiKaibutsu]

yes, i call shens on that.



exactly. if he noticed a sizeable difference going from a notmal hd to a raptor then he should have seen a much greater difference when going to an ssd.

I noticed a difference going from a 74 gig raptor to an intel g2 80 gig, but it wasn't as great as going from a 40 gig 8 meg cache 7200 rpm drive (I think that's what I was using atleast) to the 74 gig raptor. SSDs crush raptors in certain benchmarks and are better in real world stuff (atleast readwise my raptor about matches it writewise), but they aren't the godsend I see people make them out to be when compared to raptors although I assume people using lesser spindles see that huge jump.

 

[SolMiester]

I have the following setup on my i7 920 @4ghz Asus P6T

80GB Intel X25M G2 with TRIM
2TB RAID 0 WD Black
500GB WD Black

Upon running AS Benchmark, I've scored 410 which is outstanding compared to my WD Black. My Windows Index score was increased from 5.9 to 7.7

My boot time remained the same...maybe even longer than before. I suspect this due to my RAID 0 setup in addition to my SSD (loading drivers, etc)

Startup programs are instantly there...no more waiting for programs to start
But once the window is up and running, I do not notice any other changes from WD Black other than opening 3 or more programs at once.

IMO, it wasn't worth the cost and time to move onto SSD. I made the same mistake of moving up from Q6600 to i7 as I did not notice much difference.

How do you run TRIM and RAID 0 on the same controller, I thought you couldnt?

 

[PUN]

How do you run TRIM and RAID 0 on the same controller, I thought you couldnt?

TRIM on ICH10R
RAID 0 on Jmicron (i hate this controller)

Do you guys think intel will pass TRIM on raid setup? I really hate this jmicron controller and I was thinking of getting a PCI Raid card for my 7.2k

 

[Voo]

Do you guys think intel will pass TRIM on raid setup?

At the moment no raid controller passes the TRIM command to the SSDs, so no..

 

[PUN]

At the moment no raid controller passes the TRIM command to the SSDs, so no..

My question was, will they ever support TRIM command on RAID?

 

[jimhsu]

My performance test will be, how does it affect entering Dalaran in World of Warcraft. I have years of pent up frustration with apparent WoW disk bottlenecks.

WoW and generally most mmorpgs benefit obscenely from SSDs. The workload is perfect - small random reads, high randomness (people wearing 100 different types of armor/weapons, etc), low need for sequential speeds (because textures/models are generally low detail). I see differences of 15 seconds without ssd, <5 with. That kind of difference.

Most modern games, specifically RTSes and FPSes do NOT benefit from SSD random read speeds; rather sequential reads are key (because textures these days are >2K by 2K). Most importantly, the data load is predictable - the game knows exactly what content to load for each level/map.

Also much less noticeable is stalling on texture/model loads. It's been a dirty secret of sorts that minimum FPS is much more important than average or max FPSes (making most game benchmarks close to useless). An SSD dramatically, but does not remove, the HDD-memory bottleneck where new textures have to get shuffled into memory. Again for modern games this is less of a problem because textures/models are preloaded anyways, but for RPGs this remains a problem.

PS App launch times are a relatively poor benchmark. A more relevant one for multitasking scenarios is "&#37; time that disk queue length remains above 1". Most reports of computer stalling other than CPU load can be traced back to queue lengths significantly above 1 (4-5 or more is not unusual). Windows 7 can easily show you the disk queue length in the resource monitor.

 

[jimhsu]

For some math:

Avg mechanical disk seek = 15ms
Avg SSD read seek = 0.1 ms
HDD sequential read speed = 100MB/s
SSD sequential read speed = 200MB/s
We'll assume 3 bytes/pixel (uncompressed)

Typical wow texture size: 256x256x3 = 196KB
Typical "modern game" texture size: 1024x1024x3 = 3.1MB * 3(diffuse, normals, glow) = 9.4MB. The diffuse, normal map, and glow map are typically contiguous in the packed file, so random access is insignificant here.

An SSD fetches a wow texture in 0.1 + (196KB/200MB/s) = 0.1 + 0.98 = 1.08 ms
A hard disk does it in 15 + (196KB/100MB/s) = 16.96 ms. Notice that even with a huge RAID 0 array you can't get this below 15 ms.

Performance advantage = 15.7x

For the modern game texture, an SSD does it in 9.1MB/200MB/s + 0.1 = 45.6 ms
The hard disk does it in 9.1MB/100MB/s + 15 ms = 106 ms. The SSD lead here is not so impressive anymore. With a 2 disk RAID 0 array, this becomes 60.5 ms. With enough cheap disks, we can easily beat the SSD in cost.

Performance advantage = 2.32x

This all goes back to the latency vs bandwidth argument. Here's a really old but decent review on that:
http://www.stuartcheshire.org/rants/Latency.html

For perspective, accessing data via ethernet on a remote computer with SSD (0.3+0.1 ms) is many times faster than accessing data locally via a standard hard drive.

 

[ScorcherDarkly]

For some math:

Avg mechanical disk seek = 15ms
Avg SSD read seek = 0.1 ms
HDD sequential read speed = 100MB/s
SSD sequential read speed = 200MB/s
We'll assume 3 bytes/pixel (uncompressed)

Typical wow texture size: 256x256x3 = 196KB
Typical "modern game" texture size: 1024x1024x3 = 3.1MB * 3(diffuse, normals, glow) = 9.4MB

An SSD fetches a wow texture in 0.1 + (196KB/200MB/s) = 0.1 + 0.98 = 1.08 ms
A hard disk does it in 15 + (196KB/100MB/s) = 16.96 ms. Notice that even with a huge RAID 0 array you can't get this below 15 ms.

Performance advantage = 15.7x

For the modern game texture, an SSD does it in 9.1MB/200MB/s + 0.1 = 45.6 ms
The hard disk does it in 9.1MB/100MB/s + 15 ms = 106 ms. Not so impressive anymore. With a 2 disk RAID 0 array, this becomes 60.5 ms.

Performance advantage = 2.32x

For someone still fuzzy on the exact advantages of an SSD, that made things a lot clearer. Thank you!

 

[PUN]

For some math:

Avg mechanical disk seek = 15ms
Avg SSD read seek = 0.1 ms
HDD sequential read speed = 100MB/s
SSD sequential read speed = 200MB/s
We'll assume 3 bytes/pixel (uncompressed)

Typical wow texture size: 256x256x3 = 196KB
Typical "modern game" texture size: 1024x1024x3 = 3.1MB * 3(diffuse, normals, glow) = 9.4MB. The diffuse, normal map, and glow map are typically contiguous in the packed file, so random access is insignificant here.

An SSD fetches a wow texture in 0.1 + (196KB/200MB/s) = 0.1 + 0.98 = 1.08 ms
A hard disk does it in 15 + (196KB/100MB/s) = 16.96 ms. Notice that even with a huge RAID 0 array you can't get this below 15 ms.

Performance advantage = 15.7x

For the modern game texture, an SSD does it in 9.1MB/200MB/s + 0.1 = 45.6 ms
The hard disk does it in 9.1MB/100MB/s + 15 ms = 106 ms. The SSD lead here is not so impressive anymore. With a 2 disk RAID 0 array, this becomes 60.5 ms. With enough cheap disks, we can easily beat the SSD in cost.

Performance advantage = 2.32x

This all goes back to the latency vs bandwidth argument. Here's a really old but decent review on that:
http://www.stuartcheshire.org/rants/Latency.html

For perspective, accessing data via ethernet on a remote computer with SSD (0.3+0.1 ms) is many times faster than accessing data locally via a standard hard drive.

Thanks for clearing things up.
That's probably the reason why I dont see much difference from my 7.2k
I play L4D, CODMW2, Demigod, etc...never played mmorpg, but now dying to try WoW

 

[SolMiester]

I thought SSD'd were all about random read\write where they are 40x faster, and the typical i/o access of a operating system....spindle drives have always been good at sequential

 

[bryanW1995]

great explanation!

I wonder what will happen to my load times in nwn 2? I'm going to start using a stopwatch, though before/after times won't be exactly comparable since I'm also going from a q9450@3.3 to an i7 920 at (hopefully) around 4.0.

 

[Yuriman]

I thought SSD'd were all about random read\write where they are 40x faster, and the typical i/o access of a operating system....spindle drives have always been good at sequential

Benchmarked with CrystalDisk
2x Intel x25-M G2 in raid0 (well worn-in, no TRIM) vs 1.5tb WD Green (very new drive) vs 300gb Seagate (several years old)

Sequential Read - 459.8MB/s - 127.9MB/s - 59MB/s
Sequential Write - 169.1MB/s - 110.5MB/s - 57.7MB/s

Read (512k) - 324.6MB/s - 30.5MB/s - 35.7MB/s
Write (512k) - 153.5MB/s - 94.4MB/s - 33.4MB/s

Read (4k) - 26.9MB/s - 0.8MB/s - 0.7MB/s
Write (4k) - 104.0MB/s - 0.8MB/s - 0.9MB/s

----

Seq
Read - ~4x faster
Write - ~50&#37; faster

Small files
Read - ~10x faster
Write - ~50% faster

Tiny files
Read - 35x faster
Write - 130x faster

 

[Corsairs]

jimhsu, that was a brilliant explanation that really helped me. One of my (surprisingly time-intensive) hobbies is running a fictional Out of the Park Baseball sim league (the PEBA). OOTP sounds like exactly the kind of application that would benefit from an SSD. The game loads and archives tons and tons of very little files. Based on your overview, I think I'll stick with my plans of targeting a SSD for my new rig.