SSD Performance Impact On A-LIST Games

[Computurd]

well, finally!
A coherent presentation that shows the tremendous amount of benefits from SSD in-game performance. and loading too, of course.
we know that it loads faster, duh!
but what about in game performance?
one area that many have questioned is whether or not the SSD effects gameplay once the game is running. well, this should help to answer these questions! sorry guys you will have to hit the link, the PDF is larger than the site allows

https://intel.wingateweb.com/us10/sc...EC6F845E486D57

a thing of note here...this is with nothing else running on the systems, what if you have an AV running, or teamspeak, or other things, like FPS monitors, keyboard programs (g15, etc,) mouse programs, all of these add-ons that add value to your gaming experience?
these disk usage figures that you see during these gaming traces of hdd activity are not conducive to these types of things running well. on the ssd though, there is tons of room for anything, the disk utilization with ssd is orders of magnitude lower, mainly because the requests are served so fast! ultra low latency. awesome

[Computurd]

this shows something here...there is a tremendous impact of ingame performance, after the game load. i have noticed this for quite awhile myself....nothing beats an ssd for gaming!
People place far too much relevance on merely looking at ssd game load speeds. the performance benefits are much more than that. you spend thousands of times more time actually playing the game, than you do loading it. so the gameplay performance should be a much more important measurement than anything!

[xeon_1]

i do take a little bit of salt with these resuls as it is a manufactur trining to sell its product's.
And without knowing the test setup (os drive,..) it kinda muts the results.
And if you trow raid into the mix i interferce's again.
On a single raptor i had to waith upto a minute for GTA4 to load the first time.
I could literlay get something to drink go tot the bathroom and come back and still be greeted by a loading screen.
Know 15 seconds with 3 raptors in raid.

[ripken204]

that link doesn't seem to work for me.

i have never really had an issue with load times, i have been running RAID0 on HDDs for a long time now.
even GTA4 I never felt that I had to wait that long on.

occasionally when i do have to wait, it's a good thing b/c it means i can actually take a piss instead of waiting hrs haha

[Computurd]

they did HDD raid comparisons as well. link works for me. it is a download of a pdf.

the most, and i mean most, of single hdd V raid hdd from the hdd raid 0 figures is 10 percent, average around five percent.


also, the main thing of note here is the in game performance, not so much the game loads.





also a thing of note, i have done testing in the past of caviar black array v single caviar black v ssd that mirrors these results.

[Ao1]

Disappointing if you ask me. They show that both HDD & SSD are underutilised when it comes to games. (HDD 90% SSD 20%). Obviously loading is going to be faster but they fail to demonstrate any improvement in game quality experience. If they have shown frame rate improvements or some measure of enhanced experience it would have been a lot more credible.

[Computurd]

well Ao1 on this one i think you might have missed something...the charts are clearly marked with "lag" points. "arrow indicates point of character hitching"
there is also game trace data there that shows conclusively lost performance during hitching. the big gap in the gaming trace behind each lag is performance loss. see "demigod interface analysis"
yes, the ssd is underutilized, however the HDD is being over utilized.
frankly, i do not understand how this is not clear and concise.
they are not the first to publish data that conclusively shows this happening within the game. Anandtech has referred to this in his articles on ssd testing as well, as well as provided graphs.

[Tiltevros]

well, finally!
A coherent presentation that shows the tremendous amount of benefits from SSD in-game performance. and loading too, of course.
we know that it loads faster, duh!
but what about in game performance?
one area that many have questioned is whether or not the SSD effects gameplay once the game is running. well, this should help to answer these questions! sorry guys you will have to hit the link, the PDF is larger than the site allows

https://intel.wingateweb.com/us10/sc...EC6F845E486D57

that link doesnt work for me

[Ao1]

Comp, it’s really hard to know what Intel is trying to say without listening to the whole presentation. (Which I haven't had a chance to do).

Looking at Slide 4 the (single) WD VelociRaptor graph has a Y axis measuring Cumulative I/O. Rather confusingly the Intel SSD graph directly below it is showing a Y axis measuring Trace Time. Putting two graphs together like that can be very misleading.

Looking at the character hitching graphs, first I’m confused as to what that actually means. Are they saying it results in dropped frames? Slower frame rates?

Assassin Creed shows a character hitch point at ~49% disk utilisation but not at ~55% so it does not appear to be directly related to disk utilisation. It also does not seem to be related to a sudden jump in disk utilisation. For example in Assassin Creed there is a big jump in utilisation at 56 to 59 seconds from ~ 3% to ~ 23%.

Regardless of the graphs however we all know that a single SSD is going to perform better than a HDD, but what Intel seem to be showing is that 2 WD VelociRaptor in raid 0 are comparable to a single SSD when it comes to game loading times and using more than one SSD is a waste of time, because a single SSD is not utilised more than ~22% at peak and is typically well below 10%.

[Computurd]

i do not understand how you can say that the raid 0 is comparable to the SSD? surely you are reading the graph incorrectly. look again! the HDD is not on there. the two bars next to each other are BOTH SSD results. it is the comparison of the time it takes for
1. SSD V single hdd (dark blue)
2. SSD V raid 0 hdd (light blue)
there is no HDD bar on that graph. neither are very close to SSD in terms of performance.

looking at the character hitching graphs, first I’m confused as to what that actually means. Are they saying it results in dropped frames? Slower frame rates?

hitching is the lags and stutters that accompany loading during a game, they result in low frame rates while your system is literally hanging waiting for i/o. lets say you are walking from one room or area of a game, into another, and you get that slight hang, or lag, before you go into the next area. that is hitching.
anad says it much better than i do:




now you can compare that information and make your conclusions regarding performance. i mean compare the gaming trace results from anand. the hdd arent even in the ballpark to ssd performance. not saying anand is the end-all word on things, however, he is widely respected, tests about every ssd known to man, and his results mirrors intels.
i did take the liberty of taking references he made from two of his articles, and placing them together to try to make it coherent.

Assassin Creed shows a character hitch point at ~49% disk utilisation but not at ~55% so it does not appear to be directly related to disk utilisation. It also does not seem to be related to a sudden jump in disk utilisation. For example in Assassin Creed there is a big jump in utilisation at 56 to 59 seconds from ~ 3% to ~ 23%.

correct. disk utilization is not n3eccesarily a requirement for hitching, lagging, stuttering. whatever you wish to call it. if you have a 90 percent load of 100 percent seqential large blocks, your hdd will perform better at that high utilization than it will at the relatively low 50 percent utilization of 4k random access files. it isnt the amount of utilization, it is the nature of the utilization that will bring hdd to their knees.

however, that is not how they are measuring this on the demigod trace. it is not disk utilization it is pixel count. the y axis is the number of pixels that change frame-to-frame. you are correct you have to watch it to really get it. they admit that the FPS isnt a good way to quantify the issue, so they have a good methodology to illustrate the point.

if you look at the demigod chart the x axis is time. the game actually progresses faster on the SSD! it takes less time to reach certain points, as there is zero hanging. a few hangs here and there and you see a huge difference, it is the fluidity of game play that is way better.

[Computurd]

here is the link to the audio presentation with slides...it is SSDS004

https://intel.wingateweb.com/us10/sc...og/catalog.jsp

whats cool too is that they have actual third party game developers who come in and back up what they are saying. the insight into how they are designing and utilizing these things.

[eth0s]

Hey, that is some great info, thx. Now i gtg buy me an OCZ Vertex 2 Pro. And correct me if I'm wrong, but putting SSD's in a RAID still makes them slower right? So ur still better off with just one, right?

[hlonipha]

Hey, that is some great info, thx. Now i gtg buy me an OCZ Vertex 2 Pro. And correct me if I'm wrong, but putting SSD's in a RAID still makes them slower right? So ur still better off with just one, right?

Not at all my friend. Hopefully computurd can chime in and answer that as well. In the mean time look at a world record PC Mark Vantage thread by steveRo. It should show you the performance increase that raid brings to the table.

[saint-francis]

It'll show you that RAID brings benchmarks to the table. I think, but don't know as I have never used a system like the ones in that thread, that the returns are highly diminishing. I have run 3 SSD's in RAID 0 and there was a little improvement in responsiveness but nothing to justify the extravagant expense. At least not for me. For benching though it obviously makes an entire world of difference.

[Mungri]

I noticed much quicker loading times in games after raiding my F3s.

Its also a lot cheaper than an SSD.

You can get all of this for the cost of a Crucial 128 Gb C300:



Two 1 Tb F3s, and two 2 Tb F4s = £250. The second F4 is slower because it has lots of backups on it.

I really dont think I need to get an SSD for games over my Raid 0, just the 64 Gb one for Windows, drivers and apps is fine.

[Computurd]

mixed bag there. i think personally the biggest take-away here is there is a very low threshold for the 'diminishing point of return"
outside of uber benching and running hardware raids and being a storage geek there really isnt a need for it outside of two devices.

two devices gives you a fast, flexible response to high Queue Depth situations, but mainly gives you increased write speed. it is an inherent weakness of all ssd's right now that while writing they take a performance hit. hdd is the same of course. so if you have more writing bandwidth you are limiting the amount of reading/writing simultaneously.

above two devices you will not notice any significant increases in normal usage patterns, gaming included. however, two is better than one.

There are increases, but the point where you are getting small increases in performance per device added begins after two.
big uber raids of devices can be faster than anything out there. Steve can attest, and so can these guys in here. but there has to be an entire system approach in order to benefit. your computer has to be able to handle this throughput, and has to be tuned for it. CPU/GPU/RAM all of it, have to be scorching fast to utilize the uber bandwidths. it isnt worth it for the average person.
but you will always get naysayers who have a massive bottleneck in their system who proclaim "there is no benefit to more than two!"
but i can load Left For Dead levels in under 3 seconds, and Crysis levels in under 8 seconds, with no cache. these are speeds faster than a fusion i/o.

a big issue with this is also covered in the presentation. games arent coded well for ssd's yet, but once they are, you will see more benefits to the raided devices.

there is no way that anyone can set through the first 30 minutes of that speech and come away and tell me that ssd arent better for gaming, entirely.
especially considering that one of the presenters was a guy who has been making games for over 20 years. from load times, to actually running the game, and fluidity, there is just no substitute.
the game performance is something i have noticed for a few years now, tbh.
when i game on other peoples machines every single performance issue that they mentioned there is painfully obvious to me.

@bhav...it isnt about what you are seeing there. it is the small file random and latency that are hundreds, if not thousands, of times faster with SSD. they will run 'just fine' of course though...but if you do have an ssd do this..install your favorite game, first to ssd, then to hdd. now play the game for an hour on the ssd, then go and play the exact same part of that game on the hdd, immediately. without a break between.
you will be back on that ssd in five minutes. its not about load times, my friend. it is about the gameplay.

[Mungri]

I really dont think that playing on the SSD is going to improve gameplay, but I will try it with Civ 5 once its released next week.

For now I did notice much faster load times after running my games on the Raid 0 setup. I needed to add a second 2 Tb because my single one got filled bast with backups, so I just went and got two F4s that were recently released and am getting rid of the Seagate I bought a few weeks ago.

But the thing is that a 128 Gb SSD costs around £200 - £250. I can fit a second one in, but I am waiting for them to become much cheaper before buying one for games.

[Computurd]

you should watch the presentation. it makes alot of it (performance improvements) very clear. they are expensive and not for everyone. trust me i only have one larger than a 30gb. i have ten of those though...well three died recently. but they are all generation 1 ssds, cheapies! but if utilized correctly you can do some awesome stuff with them
but the point is you can buy some smaller cheaper ones and raid 'em up. gives uber results~
the largest one i own is a 60GB LE (expanded firmware) and it seems so roomy after the 30GB in my lappy...once you use the smaller ones you will be amazed with how much you can fit in small spaces! there are ways around it too..I use my WHS to stream content to my laptop, and store all of my pics, etc on the network. so i just access them over the network, and i still get great functionality!

[Ao1]

Comp, game developers want their games to be graphically rich but they also need to develop games that the majority of people can play without spending huge sums of money on hardware. If they made games that were too demanding on hardware they would sell fewer games so they have to find a balance.

The agenda of that presentation is a call to arms for developers to optimise games to take full advantage of SSD. That, by default, implies that games do not currently take full advantage of a SSD. A single SSD is clearly underutilised if you go by Intel‘s analysis.

The message to the developers is that they can further push games if SSD is taken into consideration.

I totally agree that game loading will be significantly faster between a single SSD and a HDD. I’d even agree that game play would be smoother, however with 2 HDD drives in raid 0 I think the difference would be marginal and 2 HDD’s are cheaper than a single SSD and you get a lot more storage. (Percentages can be misleading. 3 seconds to 6 seconds is a 100% increase. A 100% increase sounds impressive, 3 seconds less so).

If I look at a trace of reads/ writes playing MW2 SP for 21 minutes I get this:

Read < 1MB/s – 271MB – Over 19.15s (AVG = 0.23MB/s. 4K Read spec of Intel 160G2 = 22MB/s)
Read > 1MB/s – 2.2GB – Over 1.40s (AVG = 23MB/s. Sequential Read spec of Intel 160G2 = 255MB/s)
Write < 1MB/s – 9.6MB – Over 19.58s (AVG = 0.008MB/s. 4K Write spec of Intel 160G2 = 52MB/s)
Write > 1MB/s –10MB – Over 4s (AVG = 2.5MB/s. Sequential Write spec of Intel 160G2 = 100MB/s)

The most significant work load is reads that are over 1MB/s and these occur over a short time frame.

Obviously averages only tell part of the story but they give an idea of how hard the drive is working and clearly game loading can not utilise the available speed of my SSD.

I have a sequential read speed of 250MB/s so I can read 14GB in a minute, but I only needed to read 2.2GB and that was over 1minute 40s.

The “hitching” Intel refers to is frame to frame coherency. To measure this they needed an Interframe Analysis tool to detect the number of changed pixels per frame. A hitch is defined when <0.1 of pixels change for a duration of at least 5 frames and there was a 7% difference between a single HDD and a single SSD. My guess is that that difference would be non-existent if they had compared 2 HDD’s in raid 0 with a single SSD.

The Anandtech Crysis benchmark states it is exaggerated because it is streaming data in a way that does not often occur in a game. This only illustrates that SSD have a lot more potential and ties in with what Intel are trying to push developers to take advantage of. The other benchmark is showing IOPS and again this highlights the potential of SSD over HDD but the issue is that games don’t take advantage of it and again this ties in with Intel’s call to arms to developers.

To cut a long story short SSD’s offer huge potential but currently the way games are developed today I don’t think there is a huge difference between a single SSD and 2 HDD’s in raid 0.

[Computurd]

LOL then clearly you are seeing what you want to. the fact that they have independent third party developers who verify what they are saying is apparently lost upon you. Also, the fact that other respected sources also have documented these issues in the past means nothing.
Raid 0 on hdd does not offer the type of advances in performance that you are under the impression that they do. the increases are ten percent at the maximum. even reading the sticky in the forum will point you in that direction. however, the ssd still dominates them.
two hdds in raid 0 give little, very little, in performance increases with small random access. that is the nature of the reads that are occurring while the game is 'in flight'. the main gain is in the sequential performance with raid hdd, and that is why there is the ten percent increase (best case scenario) in the load times. loading is roughly seventy percent sequential access. the streaming during game play is much more random access, which is not the HDD's forte.

The ssd are underutilized by the current games, but they still blow the performance of the HDD out of the water. the minimal amount of seven percent? dude people spend thousands to get seven more percent out of their systems. if i told you sandy bridge 'only' gave a seven percent increase in game performance just about everyone in this forum would be clamoring for one. over the course of an hour that seven percent adds up to roughly five minutes in gaming time. and that isnt including the loading time you save, that is just game play. but also, who cares about that? the HDD hitches seven percent of the time, yet the ssd hitches ZERO percent of time. i dont care about the five minutes as much as the quality of the gameplay.

The current games play better, faster, and smoother with ssd. the next crops of games the differences will become even larger as the developers work to utilize it better. and again, you will not have to have an ssd. as they talked about , they are making tools for developers that will sense an SSD, then they will utilize, for a much richer game experience, increased bandwidth. the difference between the performance of ssd and hdd is only going to be more exaggerated, exponentially, than it is now.

Frankly, the fact that you are stating this, for our dear readers, pertains to the fact that you have argued for several months in this forum that ssd makes no difference in gaming. Now that there is conclusive data, from several sources, yet you are still 'sticking to your guns'.
Surely, all the fine people who actually design games, and hardware, and test hardware professionally (i.e. for a living) have absolutely no idea what the they are talking about, eh?

[Ao1]

Morning Comp. I only try to give a different perspective. My interest is more about understanding the minimum hardware (cost) requirements necessary to achieve the best real life performance. Somewhere between our different views is the right answer for whoever reads these posts.

[PnoT]

I just received my OCZ Vertex 2 and I can tell you from a gaming perspective I'm not really that impressed.

I will say that I've been using 2 Raptor 150's in R0 forever now and that's what I'm upgrading from but the performance in game load time and map loading is about equal.

[xeon_1]

another matter.
What difference does it make that ssd is faster for games when most games out there are still single threaded and can only access 2Gb to 4Gb of memory depending on os and tweaks.

The 70% beter load time is only a statistic
Say ssd 3 second and hdd 10 seconds
I am not going to spend over 300€ for a decent sized ssd to get a 7s shoter loadtime.
A rather spend that money on a new gfx card that would give me better overal gameplay.

Intel shouldn't have a call to arms for ssd's with the developers but for multi threaded support and utilisation instead.

[lowfat]

If I look at a trace of reads/ writes playing MW2 SP for 21 minutes I get this:

Read < 1MB/s – 271MB – Over 19.15s (AVG = 0.23MB/s. 4K Read spec of Intel 160G2 = 22MB/s)
Read > 1MB/s – 2.2GB – Over 1.40s (AVG = 23MB/s. Sequential Read spec of Intel 160G2 = 255MB/s)
Write < 1MB/s – 9.6MB – Over 19.58s (AVG = 0.008MB/s. 4K Write spec of Intel 160G2 = 52MB/s)
Write > 1MB/s –10MB – Over 4s (AVG = 2.5MB/s. Sequential Write spec of Intel 160G2 = 100MB/s)

What software are you using to do this?

[Computurd]

@zeon and Pnot...sorry guys but you are missing the point. it is in-game performance. loading the game is actually the least important thing you do in gaming. if you want to understand what we are referring to please watch the demo

@A01...sorry didnt mean to come off as irritated as i have. i just feel very passionately about the subject.

months ago i did testing and there were threads dedicated to the difference of in-game performance with SSD, and at the time many dismissed it, saying it would be impossible to prove basically, because fps wasnt a reliable enough measurement. I have been pounding the pulpit about this for a long time....well now they have come up with a metric that can quantify the difference that i have noticed. It is just that there seemed to be a dearth of information about the subject of hitching (as intel calls it). they need to release a version of this for us to use, so that we can compare hitching rates over several different solutions.
now, regardless of whether or not you feel that there is a difference currently (but seriously in the face of the testing/results, that is hard to see for me) there definitely will be a major difference shortly. the game developers that were lecturing were giving instructions to the game developers in the audience of where to find this new tool they have developed to sense ssd, and thus adjust accordingly. the difference is about to be totally irrefutable, as soon as these types of improvements come about.
another point of interest in the presentation was the filler material. during a game if the storage solution does not retrieve certain graphical elements quickly enough, they substitute generic textures/images. this feature is integrated into every major game engine. that alone should illustrate the point that the HDD simply cannot serve these types of files quickly. there is, and has, been a major problem with them streaming images off of the storage solution. the developers also discuss when this becomes too much, and the games lag/stutter/hitch, with ssd this is totally negated.