Exploring SSD Performance In Battlefield 3, F1 2011, And Rift
-
Page 1:Profiling Storage I/O In Three New Games
-
Page 2:Test Setup And Benchmarks
-
Page 3:Launching Battlefield 3
-
Page 4:Loading Levels In Battlefield 3
-
Page 5:Gameplay In Battlefield 3
-
Page 6:Launching F1 2011
-
Page 7:Loading A Race In F1 2011
-
Page 8:Gameplay In F1 2011
-
Page 9:Launching Rift
-
Page 10:Loading A Shard In Rift
-
Page 11:Gamplay In Rift
-
Page 12:Hard Drive Performance Comparison
-
Page 13:Frame Rates Examined
-
Page 14:Solid State Won't Improve All Gameplay
Gameplay In F1 2011
The two-minute trace captured in F1 2011 represents playing through a single Grand Prix race.
| Overall Statistics | F1 2011: Gameplay |
|---|---|
| Elapsed Time | 02:00 |
| Read Operations | 296 |
| Write Operations | 2 229 |
| Data Read | 35.93 MB |
| Data Written | 252.19 MB |
| Disk Busy Time | 0.63 s |
| Average Data Rate | 459.51 MB/s |
Gameplay in F1 2011 is interestingly similar to our last set of games, where a majority of operations are sequential writes. The I/O pattern is more similar to Crysis 2 than WoW, as there's little variety in F1 2011's file structure. Almost half of the disk accesses occur at a queue depth of two. Compare this to Crysis 2, where a majority of disk accesses are evenly distributed between queue depths of two and eight.
I/O Trends:
- 91% of all operations are 128 KB in transfer size
- 98% of all operations are sequential
- 50% of all operations occur at a queue depth of one
- 44% of all operations occur at a queue depth of two
Summary
- Profiling Storage I/O In Three New Games
- Test Setup And Benchmarks
- Launching Battlefield 3
- Loading Levels In Battlefield 3
- Gameplay In Battlefield 3
- Launching F1 2011
- Loading A Race In F1 2011
- Gameplay In F1 2011
- Launching Rift
- Loading A Shard In Rift
- Gamplay In Rift
- Hard Drive Performance Comparison
- Frame Rates Examined
- Solid State Won't Improve All Gameplay
And who the hell would buy an expensive SSD but cannot even put a decent amount of RAM on his system?
Good read. Thanks for being so thorough.
I'm a little confused why you would want to see that comparison. We established there was no diff between a slow HDD and a fast SSD. And you expect a difference between a fast HDD and fast SSD?
Cheers,
Andrew Ku
TomsHardware.com
I see what you did there. :P
He probably want to see that comparison on the "Hard Drive Performance Comparison" page, where is a big difference between SSD and HDD performance (the HDD up to 477% slower)
The point of using a slower drive than those is to emphasize the effects of using SSD vs HDD when it comes to gameplay/FPS. Of course, if the slower drive is quite on par with SSD on FPS ratings then the faster HDDs will do well too.
And who the hell would buy an expensive SSD but cannot even put a decent amount of RAM on his system?
Notably, my laptop with a Intel 510 250GB, my in-game load times are about 2 seconds in skyrim compared to the desktop running 1TB spinpoint at raid takes 4-5 seconds. The same desktop with a crucial M4 takes around 2 seconds. Moreover I never had any spikes/lags with SSDs while gaming.
Nobody would of course, but the point is it can make a difference.
If you are just looking at frame rate performance, you are of course correct Andrew; what's the point?. However, what about the launching of the games and level loading? In some games there is a marked difference.
The comparisons I'd like to see would be the load times for game launching and levels with faster hard drive solutions. As an MMORPG player, the zoning times would be of interest - how long does it take to zone into another instance, and does that significantly improve with an SSD?
I'd like to see 2 hard drives in RAID 0 compared. Other single hard drives would be interesting (being a great fan of the Velociraptor series, I'm always interested in those, even if they are kind of an odd duck at their price in today's world) but to a lesser degree for me.
The other test comparisons I'd like to see would be SSDs used in caching systems, such as what the Z68 chipset allows, (a whole article could be done on the effects of size of SSD on caching, I think) and solutions like the OCZ Synapse, and RAM drives. (RAM is so cheap now popping an extra 12GB into a 24GB capable mainboard and setting up a 16GB RAM drive looks like a very attractive way for me to boost performance in my aging X58 system.
Please don't take any of my suggestions as complaints or that I'm saying your tests were incomplete; I'm just throwing some ideas out. I'm well aware of how much effort goes into producing these articles, and am grateful for all that you do.
He did WoW in a previous article.
I would like to have seen Skyrim too, but really Andrew's test suite should be about displaying different facets of storage solution performance before it's about pleasing the readers with games they want to see - and the games he chose are extremely popular. In my opinion, Rift isn't "the latest craze" in MMOGs, it's kind of old news, but it is still a good test to run at least once in that we see how it runs and how it is different than, say, WoW, or the other non-MMOG games.
One thing is very clear, games aren't all the same; not even games of the same type. One of the reasons I kept asking for WoW to be used as a bench in the past was because it was so CPU intensive, and most game weren't. In Cataclysm, that changed, and it became a game that could stress both CPU AND GPU, so I think it is a very good game to use for benches. However, MMOGs are harder to bench with than other games.
You can only use so many games for benchmarks, and some people are going to wish other games were used so an article relates to them more directly. I think it's good to make suggestions, but have trouble understanding why people get surprised when a game isn't used, talk in terms of something being "left out", or seem to take things personally.
It matters a lot when using a helicopter or plane as trees are notoriously appearing at very short range all of a sudden. While it doesn't affect the frame rate the game works a whole lot better with enough IO performance.