[OBLIVIONLORD]

I'm still encountering some people that say Superfetch is beneficial in gaming however I beg to differ. It does not load Textures into memory nor does it dynamically load/unload .Dll files pertaining to the game while you are playing on the same map. SuperFetch does NOT cache files - it prioritizes 4kb pages e.g. parts of files - which XP/Vista prefers to deal with... and loads them into memory ahead of time. Files contain 100s of 1000s of pages each.

I have done some extensive benchmarks with 3dMark06 and Forged Alliance to confirm this.

Test System = The Specs In My Sig
Nvidia Driver = 181.22
Forged Alliance Build = 1.5.3599
3DMark06 Build = 1.1.0

I always conduct my tests by doing 5 passes and then determine the total avg at the end. I have nothing running in the background while conducting these tests. The first pass is always after a fresh boot and then I continue on with the rest of the passes after the 1st.

For 3dMark06, I simply clicked on "Run 3dMark". Since it's the basic version, I can't adjust anything which is the way it's supposed to be.

3dMark06
--------

Superfetch Enabled
------------------

1st Pass
--------
Score ----- 15369
Cpu Score - 5737
Sm2 ------- 6012
Sm3 ------- 6441

2nd Pass
--------
Score ----- 15412
Cpu Score - 5823
Sm2 ------- 6018
Sm3 ------- 6440

3rd Pass
--------
Score ----- 15405
Cpu Score - 5802
Sm2 ------- 6026
Sm3 ------- 6435

4th Pass
--------
Score ----- 15359
Cpu Score - 5763
Sm2 ------- 5994
Sm3 ------- 6439

5th Pass
--------
Score ----- 15366
Cpu Score - 5765
Sm2 ------- 5998
Sm3 ------- 6440

Total Avg
---------
Score ----- 15382
Cpu Score - 5778
Sm2 ------- 6009
Sm3 ------- 6439


Superfetch Disabled
-------------------

1st Pass
--------
Score ----- 15371
Cpu Score - 5761
Sm2 ------- 6004
Sm3 ------- 6440

2nd Pass
--------
Score ----- 15370
Cpu Score - 5766
Sm2 ------- 5999
Sm3 ------- 6773

3rd Pass
--------
Score ----- 15422
Cpu Score - 5836
Sm2 ------- 6021
Sm3 ------- 6443

4th Pass
--------
Score ----- 15377
Cpu Score - 5743
Sm2 ------- 6015
Sm3 ------- 6443

5th Pass
--------
Score ----- 15362
Cpu Score - 5721
Sm2 ------- 6009
Sm3 ------- 6443

Total Avg
---------
Score ----- 15380
Cpu Score - 5765
Sm2 ------- 6017
Sm3 ------- 6508

----------------------------------------------------

For Forged Alliance I used its built in benchmark which has some minor inconsistencies in how it plays through but, it is more real time than just a timedemo. To activate it, enter "/perf" at the end of the Target Address in the shortcut.

Forged Alliance
---------------

Superfetch Enabled
------------------

1st Pass
--------
Score ----- 18190

2nd Pass
--------
Score ----- 18192

3rd Pass
--------
Score ----- 18193

4th Pass
--------
Score ----- 18244

5th Pass
--------
Score ----- 18185

Total Avg
---------
Score ----- 18200


Superfetch Disabled
-------------------

1st Pass
--------
Score ----- 18165

2nd Pass
--------
Score ----- 18269

3rd Pass
--------
Score ----- 18258

4th Pass
--------
Score ----- 18274

5th Pass
--------
Score ----- 18162

Total Avg
---------
Score ----- 18225


My memory and HDD are quite fast which is why I don't see much of any difference on my end while even doing regular stuff outside of gaming. If it is any stipulation, I will do the same tests on a slower system later on to show the difference on slower hardware.

[e6600]

i think superfetch enabled on a lower memory (2gb) system should destroy performance

[OBLIVIONLORD]

My other system is an AMD 4600x2 with a HD4650.. 2x1gb 667mhz DDR2 Ram. I think that will be a great system to do the test because the ram in general is slower than the 800mhz DDR2 Internal memory controllers capability. It also has a 320gb HDD which has a higher random access than the 640gb AAKS which will also be a factor.

Ill do it tommorow.

[johnz]

None of those tests bench load time, which is what makes SuperFetch nice for gaming. Anyone expecting higher fps was just confused.

[usp8riot]

I love these thorough benchmarks. But I do believe Johnz's right. I'd believe superfetch may be worth enabling on a slower system with lots of memory. I have it disabled.

[GTengineer]

Originally Posted by johnz

None of those tests bench load time, which is what makes SuperFetch nice for gaming. Anyone expecting higher fps was just confused.

BINGO

[usp8riot]

I can see how Superfetch would work in typical apps but how can it work inside of games? If the game needs gamelevel.pk3 when changing levels, how would Vista know to prefetch it? I just assumed it prefetches standard dll's and such to load the app or game and once loaded, it leaves the rest to the game. Any benchmarks to show any increases in level load time?

[johnz]

Originally Posted by usp8riot

I can see how Superfetch would work in typical apps but how can it work inside of games? If the game needs gamelevel.pk3 when changing levels, how would Vista know to prefetch it? I just assumed it prefetches standard dll's and such to load the app or game and once loaded, it leaves the rest to the game. Any benchmarks to show any increases in level load time?

I don't know about standard sp games(I don't play that many), but the load times are noticeably reduced in UT2k4. I mostly play on my clan server, so the map selection is fixed. In most games there's many recurring textures, and other attributes, so they could be prefetched fairly reliably.

Edit:
I'm first person to complain about empirical performance improvements that people make claims to. After playing UT2k4 for 3 years on XP(avg 4 hours per day), I immediately noticed the change when I went to Vista. It would be a PITA to bench, and I don't care too much tbh, but that's my subjective experience.

Edit2:
The boost in load times isn't so great that I think gamers should rush out and buy Vista. I'd put it under the category of a "nice bonus". If a map loads in 5 seconds as opposed to 10 seconds, it isn't really that big of a deal. You still have to wait for the others to join the game :^D

[OBLIVIONLORD]

The faster HDD makes more of a difference in load times since you are loading maps. Superfetch does not preload maps into memory. It does load .dll files essential to the functionality of the program in pages. That still is very minor. The first pass of these games show virtually identical results to the other passes. The first pass on each test was initialized after the game was loaded and from a warm boot. There weren't any preruns of the game prior to the 1st pass on the tests.

In essence you should get similar results on slower hardware while in game. However the results will be different and in favor for the system that has Superfetch enabled when it first loads since that's the main area where Superfetch comes into play. Games have their own caching systems for textures and such. If enabling superfetch made any game run faster, then it would mean the game coders suck. The game would probably run very slow to the point that it is unplayable. It's simply not the reason Superfetch is used for.

However for the sake of the matter, I'll do some tests on my old system just as another reference. I am conducting these tests only because people think that Superfetch will increase performance. Again this is not about how fast the game loads.

[OBLIVIONLORD]

Ok I finished the tests on the other system.

AMD X2 4600
320gb HDD .. to lazy to look up the model
2x1gb 667mhz DDR2

At the times when Superfetch was enabled, I had to let the system sit for 5 minutes after a warm boot so it could load everything into memory. When Superfetch was disabled, I could just jump right into things without wait. This has alot to do with the HDD having a high random access time even though it's a 7200rpm drive compared to my AAKS which is a good 5ms lower. I'm sure the times would improve if I used the AAKS drive instead but, I was mainly going for an 'older' system comparison to a modern system.

Anyway here are the results....

3dMark06
--------

Superfetch Enabled
------------------

1st Pass
--------
Score ------ 4218
Cpu Score - 1713
Sm2 -------- 1511
Sm3 -------- 1855

2nd Pass
--------
Score ------ 4213
Cpu Score - 1731
Sm2 -------- 1503
Sm3 -------- 1853

3rd Pass
--------
Score ------ 4211
Cpu Score - 1730
Sm2 -------- 1500
Sm3 -------- 1853

4th Pass
--------
Score ------ 4215
Cpu Score - 1720
Sm2 -------- 1505
Sm3 -------- 1856

5th Pass
--------
Score ------ 4221
Cpu Score - 1748
Sm2 -------- 1503
Sm3 -------- 1853

Total Avg
---------
Score ------ 4215
Cpu Score - 1728
Sm2 -------- 1504
Sm3 -------- 1854


Superfetch Disabled
-------------------

1st Pass
--------
Score ------ 4221
Cpu Score - 1726
Sm2 -------- 1511
Sm3 -------- 1853

2nd Pass
--------
Score ------ 4221
Cpu Score - 1733
Sm2 -------- 1506
Sm3 -------- 1855

3rd Pass
--------
Score ------ 4217
Cpu Score - 1726
Sm2 -------- 1508
Sm3 -------- 1853

4th Pass
--------
Score ------ 4223
Cpu Score - 1759
Sm2 -------- 1502
Sm3 -------- 1854

5th Pass
--------
Score ------ 4214
Cpu Score - 1728
Sm2 -------- 1503
Sm3 -------- 1854

Total Avg
---------
Score ------ 4219
Cpu Score - 1734
Sm2 -------- 1506
Sm3 -------- 1853

----------------------------------------------------

Forged Alliance
---------------

Superfetch Enabled
------------------

1st Pass
--------
Score ----- 15481

2nd Pass
--------
Score ----- 15526

3rd Pass
--------
Score ----- 15537

4th Pass
--------
Score ----- 15585

5th Pass
--------
Score ----- 15337

Total Avg
---------
Score ----- 15493


Superfetch Disabled
-------------------

1st Pass
--------
Score ----- 15450

2nd Pass
--------
Score ----- 15556

3rd Pass
--------
Score ----- 15596

4th Pass
--------
Score ----- 15567

5th Pass
--------
Score ----- 15552

Total Avg
---------
Score ----- 15544

These results show that there is no difference while in a game even on this older system.

I will say that I much prefered the load times of these games as well as the other apps while Superfetch was disabled. The only way Superfetch made an impact to providing faster load times was if you opened an app while the system was completely at idle. Problem with this was that alot of the times you have to wait just for it to be in a complete idle state since it's constantly loading pages in memory. Again this is a slowdown do to the HDD. I had plenty of physical memory of the 2gb while doing the testing so the space wasn't an issue.

I see Superfetch to be a great benefit on modern hardware since the performance hit is no longer an issue but, for older systems I would always recommend to disable Superfetch.