"Something is bunk, then it's de-bunked." -- Jerry Seinfeld
Black Viper (amongst others) runs a Windows tweaks page offers many changes and customizations that users can make to MS Windows in the name of greater performance. BV recommends that many active-by-default Windows services be set to manual activation or disabled altogether, to save on memory useage and CPU cycles.
I have taken three systems with clean installs of Windows XP and reviewed these tweaks. Well, the third isn't really a clean install, it has a bunch of software running which any typical user would run e.g. firewall, antivirus, speedfan, etc. The first two are lower-end systems that should theoretically benefit most from this treatment, neither has the prefix ?giga? in either its CPU speed or RAM capacity. The last one is a modern system (at the time of writing, anyway).
Part I: Low-end system
First up is the PII-300 system, in fact a PII-233 overclocked to 300MHz, with 160MB of RAM and an nVidia TNT2 PCI graphics card. This system is not really capable of being a useable XP box. We'll see if BV can help make it into one.
First I measure RAM utilization in Task Manager, then I run some components of PCMark 2004 since this system lacks the hardware capability to run portions of the test. As peak commit charge while running this suite is about twice the physical RAM, if performance can be improved it should show here. Each benchmark was run three times and the results averaged, with a reboot in-between to prevent caching skewing the results.
I also ran 3DMark 2001SE in the same average-of-three manner, although not all tests were completed since, again, the computer isn?t capable. Then I ran Quake II timedemos, best-of-three FPS on demo1.dm2. 3DMark and Quake II were run at 640x480x16 to avoid a possible graphics card bottleneck.
Before the results, a note on usability. The "Power User" level is a misnomer since it actually disables many services a power user might want or need, such as the IMAPI CD-burning service, the logical disk manager, the Themes (the pretty interface, which I happen to prefer to the mid-90s look), and the Windows firewall. This also poses a security risk. Since I was not running a 3rd-party firewall (didn?t want to waste the RAM), once I had gone to the Power User tweak level the computer was vulnerable. I simply unplugged the Ethernet cable from this point on, although it wasn?t necessary at the Barebones level because a computer thus configured isn?t actually capable of using any kind of network. Don?t imagine that the services BV disables to achieve his Power User and Barebones configurations are unnecessary or useless for the desktop user, they?re not, and memory savings here definitely come at the expense of functionality.
This shows RAM used on first booting the system and consulting Task Manager. The Manager does consume a little overhead, but it?s the same across the board. I also show the saved RAM, the percentage improvement, and the value of that saving, calculated at about 8.7 cents per MB (Newegg offers 1GB of Corsair value RAM for $89.75).
Fresh installation: 90.6MB Screenshot
Safe tweak level: 85.4MB (5.2MB saved or a 5.74% improvement, 7.5% more available, value: 45 cents) Screenshot
Power User tweak level: 66.1MB (24.5MB saved or a 27.04% improvement, 35.3% more available, value: $2.13) Screenshot
Barebones tweak level: 57.2MB (33.4MB saved or a 36.87% improvement, 48.1% more available, value: ($2.91) Screenshot
This does look good as a percentage. However, bear in mind that on a system with 2GB of RAM the absolute saving will be practically identical. On the 160MB box, getting 33.4MB back amounts to 20% of total physical RAM, but with 2GB, it?s a piddling 1.63%. At current prices, just buy more RAM, especially considering the massive loss in functionality to achieve that 1.63%.
PCMark results (File Compression & File Decompression in MB/s)
These types of operations tend to show performance increases with better latency and speed in benchmarks. Let?s see how the different configurations fare, along with percentile improvement over an untweaked system.
Fresh installation: 0.494/3.036 Screenshot
Safe tweak level: 0.482/3.025 (-2.43%/-0.36%)
Power User tweak level: 0.474/2.990 (-4.22%/-1.54%) Screenshot
Barebones tweak level: 0.486/3.011 (-1.65%/-0.83%) Screenshot
The tweaks are an abject failure here, yielding an extremely slight but across-the-board decrease in performance. It?s not for certain that it is the tweaks which have made the performance worse since the changes are so small and inconsistent, however, we can certainly say that they have made absolutely no improvements.
PCMark results (File Encryption & Virus Scanning in MB/s)
Fairly standard desktop tasks. Let?s see how the tweaks can help performance here.
Fresh installation: 3.521/181.629 Screenshot
Safe tweak level: 3.487/187.717 (-0.98%/+3.35%) Screenshot
Power User tweak level: 3.520/184.549 (-0.03%/+1.61%) Screenshot
Barebones tweak level: 3.486/179.930 (-1%/-0.94%) Screenshot
Again, we see absolutely infinitesimal changes, but not overly favourable to tweaking and unsupportive of the idea that freeing up more RAM automatically increases performance - the configuration with the most free memory was the worst performer.
PCMark results (Grammar Check in KB/s)
Another pretty standard task.
Fresh installation: 0.395 Screenshot
Safe tweak level: 0.385 (-2.6%) Screenshot
Power User tweak level: 0.383 (-3.13%) Screenshot
Barebones tweak level: 0.381 (-3.67%) Screenshot
More infinitesimal numbers that definitely wouldn?t be noticed by the end-user, however, here we actually have a trend: more tweaking makes performance progressively worse, although not in proportion to the RAM saved, so again the results don't really support any conclusion other than that services tweaking does not have a positive effect on performance.
PCMark results (Image Processing in MPixels/s)
Image processing tasks are supposed to like lots of RAM. Let?s see how freeing up a little more helps.
Fresh installation: 1.615 Screenshot
Safe tweak level: 1.616 (+0.06%) Screenshot
Power User tweak level: 1.570 (-2.87%) Screenshot
Barebones tweak level: 1.596 (-1.19%) Screenshot
Again, nothing noticeable. Tweaking proved to be a complete waste of time here as well.
3DMark results (Car Chase, Dragothic, Lobby)
Here are the recorded framerates, along with the average % change in performance (sum of the % changes/3).
Fresh installation: 4.0/3.1/10.1 Screenshot
SafeSafe tweak level: 3.9/2.9/10.0 (-3.49%) Screenshot
Power User tweak level: 3.9/3.0/10.2 (-1.63%) Screenshot
Barebones tweak level: 3.9/3.0/10.0 (-2.3%) Screenshot
Again, the tweaks seem to have a negative effect, but not a consistent one. The only clear messages here are that tweaking does not help, and also that this system seriously sucks at 3D graphics.
Quake II FPS
An old game, but within the capabilities of this system.
Fresh installation: 25.4
Safe tweak level: 25.7 (+1.18%)
Power User tweak level: 25.57 (+0.67%)
Barebones tweak level: 25.47 (+0.28%)
Although this may look like a triumph for tweaking, absolutely nobody, ever, is going to notice 0.3fps (the biggest increase). Nor are the results consistent. In fact, these kinds of results could very easily be caused by something else in the system and not tweaking at all.
The conclusion at this point is obvious: tweaking doesn't help real-world performance at all, at least, not in gaming or any of the common desktop tasks tested here. In fact, it seems to be an overall net loss. Don't bother, if you need more RAM, buy some.
Part II: Mid-range system
OK, so this system isn't exactly mid-range, but it can run XP comfortably. It's a PIII-667 with 640MB of RAM and a GeForce DDR card. It's a brand-new installation of XP SP2, untouched except for the 71.89 nVidia drivers and the installation of the required benchmarking suites and programs.
Fresh installation: 93.7MB
Safe level: 86.7MB (7MB saved, 7.5% reduction or 1.2% more available RAM, value: 61 cents)
Power user level: 70.0MB (13.7MB saved, 14.6% reduction or 2.5% more available, value: $1.19)
Bare-bones level: 61.7MB (22MB saved, 23.5% reduction or 4% more available, value: $1.91)
The absolute and percentile RAM savings here are even less than on the low-end system. The PIII box uses more memory because it has more attached hardware, but even so, it seems the more RAM you have (and the more peripheral hardware), the less these tweaks are worth.
Before I get into the benchmarks, I ran 5 tests on each one (with a reboot in between each) on the same system configured in the exact same way. This is to give you some idea of the margins of normal variation. They are listed as min/max/avg/% variable. Scores after tweaking within these rough boundaries cannot be ascribed to the tweaks, but rather to the anomalous behaviour of a highly complex system such as a modern computer.
PCMark 2004: 956/1023/987 -3.24/+3.65%. The PCMark scores show the greatest variation.
3DMark 2001SE: 2375/2414/2395 -0.84/+0.79%. 3DMark is a lot more consistent.
Quake III: 64.3/64.5/64.4 -0.16/+0.16%. This is the least variable of the benchmarks. A Quake III benchmark run should vary by more than 0.1-0.2 FPS before being ascribed to outside influence.
Fresh installation: 1019
Safe level: 1041 (+2.16%)
Power user level: 1044 (+2.45%)
Bare-bones level: Did not complete (system too crippled at this point), but the individual results are similar.
Tweaking appears to offer a performance increase here, but it is so slight that it falls well within the boundaries of normal variation. There is no conclusive proof that tweaking has helped performance at all.
3DMark 2001 SE
Fresh installation: 2363
Safe level: 2379 (+0.68%)
Power user level: 2414 (+2.16%)
Bare-bones level: 2409 (+1.95%)
The Power-user and Bare-bones configurations appear to offer some performance advantage outside of the normal tolerances. However, the inconsistency of these results suggests that it's not due to saving RAM and CPU cycles, or else the bare-bones system should have performed better than the power-user one, not worse.
Quake III Arena 1.32, OCAU's slayer demo
Fresh installation: 64.4
Safe level: 64.7 (+0.47%)
Power user level: 64.2 (-0.31%)
Bare-bones level: 63.9 (-0.78%)
Again, inconsistent and unnoticeable results. As before, nobody will ever notice 0.3fps, and the more services are disabled, the worse the performance gets, not the better.
In conclusion to Part II, the idea that services tweaking can actually produce useful and noticeable performance gains is proven wrong again. In no way are these tweaks worth the time or the sacrifice in functionality.
Part III: Modern system
This is my rig
. We'll see if this makes any difference. I expect the absolute differences will be even less noticeable than before. Since BV's site is offline at the time of writing, I took the "Safe" and "Gaming" configs from Major Geeks
. All graphics tests were run at 800x600x32 with all eye-candy off.
Default: 163MB Screenshot
Safe: 170MB No, I don't know why this happened. Loss of 4.2% used, 0.34% of total physical RAM. Value: -$0.60 Screenshot
Gaming: 142MB Gain of 14.7% of used, 1% of total physical RAM. Value: $1.82 Screenshot
Default: 26516 3DMarks Screenshot
Safe: 26485 3DMarks Screenshot
. A loss of 0.1%. Unnoticeable and well within the margin of error.
Gaming: 26499 3DMarks Screenshot
. A loss of 0.06%. Again, unnoticeable and within the margin of error. We can't even be certain that it was the tweaks which caused these incredibly infinitesimal variations.
Default: 4882 PCMarks Screenshot
Safe: 4936 PCMarks Screenshot
. A gain of 1.1%. However, we know that PCMark 2004 results can vary by over 3% above or below the baseline, so this doesn't mean anything, especially as the "safe" config somehow ended up using more
Gaming: 4915 PCMarks Screenshot
. A gain of 0.68%. Again, inconsequential, within the margin of error, and inconsistent with the level of tweaking applied.
Quake 3 timedemo
Safe: 405.9fps. A loss of 0.34%, which is totally unnoticeable anyway.
Gaming: 406.5fps. A loss of 0.2%.
This was not a demo, but a FRAPS measurement of framerate while I played 30 seconds of the "Catacombs" level. I attempted to do much the same thing in much the same order, so they should not vary that much, but obviously there will be much greater variation here. I included these because of the criticism of "artificial" benchmarks made against timedemos (which is invalid, but regardless, this is actual gameplay and not pre-recorded).
Safe: 166.7fps. A loss of 4.38%, not especially noticeable (although 5% is not inconsequential), and we can't be certain that the tweaks caused this. Certainly the game felt no different, subjectively.
Gaming: 166.7fps. Coincidence? Maybe! In any case, definitely not an improvement.
This was run with Hardware OC's FarCry Benchmarking Utility
1.4.1. I ran three passes of the included "PC Games Hardware Demo". The results between passes were virtually identical anyway.
Safe: 138.05fps. A loss of 5%, close to what we saw in Painkiller.
Gaming: 145.07fps. A gain of 0.05%. You'd never notice this.
Default: 1m 22.329s. Screenshot
Safe: 1m 22.485s. Screenshot
. A loss of 0.19%. Inconsequential.
Gaming: 1m 22.032s. Screenshot
. A gain of 0.36%. Again, inconsequential.
Again, on a modern system, services tweaking fails to produce any useful or noticeable result.
As Adam and Jamie from Mythbusters
would say, "it's busted."
Part IV: "Swap files" [sic], Themes and other miscellany
Now we come to BV's non-services recommendations.
First off, he discusses "swap files" (by which he means page files). His recommendations, from the worst performance to the best:
1) The Default: A dynamic swap file on the same partition and physical hard drive (usually C as Windows.)
2) A dynamic swap file on a separate partition, but on the same physical hard drive as Windows.
3) A static swap file on a separate partition, but on the same physical hard drive as Windows.
4) A dynamic swap file on a separate hard drive (and preferably, controller) from Windows and frequently accessed data.
5) A static swap file on a separate hard drive (and preferably, controller) from Windows and frequently accessed data.
6) No swap file at all. Some software may fail. You also need "much" memory to do this. Greater than 512 MB, but I recommend 2 GB.
He has actually run some benchmarks himself, and has noticed no difference in FPS. However, his methodology isn't that good. With the load he places on the system he can't even be certain that the pagefile is even being used, and if it's not being used, it's performance won't impact FPS at all.
I went back to System 1, because it's much easier to waste 160MB of RAM than 640MB. On bootup, I loaded Task Manager, 2 instances of Internet Explorer (Google home page), Windows Movie Maker (no file loaded), Outlook Express, and 4 instances of Paint, each with a 1152x864 24-bit BMP file loaded.
Total load is about 185MB, so we know for sure that we're hitting the pagefile. This is borne out in benchmarks. The same system without any extra load gets around 25FPS in a Quake II timedemo, at exactly the same settings with this load, it gets around 20.
On to the benchmarks. Three runs of each were run and the results averaged.
Test 1: Default (dynamic, Windows-managed pagefile on system partition
Quake II: 20.5FPS
3DMark 2001 SE: 279
PCMark 2002: 739 CPU, 416 Memory, 194 HDD
Test 2: Dynamic, Windows-managed pagefile on a different partition of the same disk
Quake II: 20.3 (-1%)
3DMark 2001 SE: 279 (0%)
PCMark 2002: 738/416/193 (-0.1%/0%/-0.5%)
Distinctly underwhelming. Nothing that could be perceived.
Test 3: Static pagefile (1000MB, as recommended by BV) on a different partition of the same disk
Quake II: 20.1 (-2%)
3DMark 2001 SE: 279 (0%)
PCMark 2002: 737/421/189 (-0.3%/+1.2%/-2.6%)
Absolutely nothing noticeable again.
Test 4: Dynamic pagefile on a different hard drive
Quake II: 20.2 (-1.5%)
3DMark 2001 SE: 279 (0%)
PCMark 2002: 737/409/195 (-0.3%/-1.7%/+0.5%)
On this supposedly higher-performance configuration, the memory gain and HDD loss of the last configuration are both gone. This would seem to indicate that we're seeing variations within the benchmark's margins of error rather than actual performance differences. Anyway, there's still nothing noticeable.
Test 5: Static pagefile (1000MB) on a different drive, using a different controller channel
Quake II: 20.1 (-2%)
3DMark 2001 SE: 279 (0%)
PCMark 2002: 737/421/189 (-0.3%/+1.2%/-2.6%)
Results absolutely identical to (3). Nothing worthwhile.
I didn't test configuration (6). Windows is not supposed to run without a pagefile, and even BV acknowledges that he encounters glitches, crashes, sound problems, programs refusing to load and so on. If it's not useable, what's the point?
Anyway, pagefile tuning for performance is useless. There are no gains to be had here. About the only thing it may be worth doing is to move the file to another partition, but that's to lessen fragmentation, not for performance.
To those who don't believe me, maybe you'll believe Anand Lal Shimpi (and you should, since bothering to read his website). He recently reviewed Gigabyte's i-RAM and tested it as a pagefile drive here
Some excerpts that reflect his impressions:
There was no real tangible performance difference between putting more memory in the system and using the hard disk for the pagefile or putting less memory in the system and using the i-RAM for the pagefile.
...for the most part, our system was slower when we had less memory and stored the swapfile in it than when we had more memory and less swap file.
For most people, you're much better off just tossing more memory in your system.
Bear in mind that the i-RAM is a solid-state device that tests
as having 3-6.75x the I/O performance of a WD Raptor, and the results of putting the pagefile on it were so underwhelming compared to just adding more RAM, how big do you possibly think the results of putting the pagefile on another hard drive, or on another partition of the same
hard drive, could be?
If the i-RAM is not really a good replacement for more memory, there is no possible way that pagefile tweaking could be.
From BV's site:
This is only valid for Windows XP Home: Do this NOW!! Everyone on XP Home, by default, has Administrator privileges and the User name is "Owner." If I know that, so does everyone else on the planet. Change the name and / or password your account. If anything, password it. NEVER have an account unprotected! EVER!
From STaSh, AT forum member:
Passwords are fine, but since most people will probably use a password like 'password' this tidbit actually LOWERS the security of the system. Why? Because BY DEFAULT, accounts with no passwords are not allowed to connect over the network. Only to the console. If you are not worried about physical access (like most people who use XP Home), an account with no password is very secure.
Get rid of System Restore Service and Indexing Service.
System Restore Service can be invaluable, if you like to use the latest drivers, change hardware, try out new software and so on. Yes, it will use hard disk space. Hard disk space costs about 50 cents per gigabyte. What do you care? Even if it uses a full gigabyte (which I doubt), I'd pay 50 cents for peace of mind.
Indexing Service is not started automatically anymore. If you don't like it and don't need it, don't use it, and it will never bother you.
From BV himself:
I turn off Automatic Updates... I highly recommend you DO NOT disable this function.
BV recommends disabling them. As Parts I and II show, there's no performance gain to be had from disabling them. When I installed XP on System I, Windows automatically disabled all but three of the visual enhancements. System II ran fully enabled and I did not notice any decrease in speed. Based on the specs of these systems I would say that if you cannot run all of XP's visual enhancements without suffering noticeable slowdowns then the computer is simply not fast enough, and you would be better to stick with 2K. In any case, deviating from the Windows defaults is simply not necessary or worthwhile.