Way 2 Cool
Way 2 Cool
This is old stuff -- circa mid-to-late1997 for the initial tests, mid-1998 for the article -- enjoy it, but it doesn't really apply in 2003+!
I have been testing and writing about the CPU cooler programs for about a year now. I that time there have been a number of version changes for these programs. In this archive are the old reviews and tests. This is also where you will find the results for the cooler programs when run on my AMD K6.
The objective of these tests is quite simple. Determine which CPU cooler program will keep my processor running its coolest while adversely affecting performance the least.I have assembled a group of temperature and performance readings for real-world applications and synthetic benchmarks.What I have found is that while all of the programs help to keep the CPU from running as warm, some of the versions of CpuIdle don't seem to be very effective in keeping the processor cool during file transfers.
As is usual for me, the results I get from testing seem to raise more questions than answers. This leads to more tests. I am now at the point that I am questioning the results of using WinBench98 to test the processor when CpuIdle is installed in the system. When running WinBench98 as a suite of tests, the results are predictable in 3 out of 4 tries. For the 4th attempt, the CPUmark32 results dip by a large margin. The frequency of this phenomenon is increased when running CPUmark32 by itself. Whether this is a problem limited to my system or not, remains to be seen. We are now getting set up to mirror these tests on a friends computer and I will update this page when the results are in.
First I needed a way to monitor the CPU's temperature. As my motherboard does not include this function, I needed to add some type of thermometer to my system. I ended up with a Radio Shack Temperature Sensor and Control Module. Part number RSU 10640316 and catalog number 277-123. List price was $19.95 There are good and bad points to this unit. On the good side, the unit comes with a schematic that allows you to add some switches for displaying the highest and lowest temps for each session. You can also add a simple circuit to enable a visual or audible alarm for high and low temps. Pretty neat.
On the down side, the temp readout will not exceed 122° Fahrenheit (50° C). This meant that with the case cover on my computer and stressing the processor, the temperature of the CPU got hotter than I could read. This was initially solved by removing the computer's cover while performing the tests. After I had finished the tests in a uniform environment, I reworked the case's cooling system. This has reduced the temp enough that the 122° limit is no longer reached.
The Cooler Programs
How do these programs work? Basically by executing an "HLT" instruction to the processor to put itself into a suspend mode when it is not actively computing. Windows NT and many non-MS operating systems do this as a normal process. Windows 95 and Windows 98 (at least the most recent beta versions I have seen) do not. These programs offer the benefit of less power consumption when the CPU is idle and no benefit when the CPU is actively processing. However, many tasks do not require the CPU to be computing constantly and this is when these programs go to work.The methods of enabling the suspend mode differ between CpuIdle and Waterfall/Rain. CpuIdle uses a virtual device driver (VxD) and Waterfall and Rain do not. With all three programs, the instructions are carried out in the Ring0 level of the architecture. Ring0 is the most privileged ring of the architecture. While the majority of people using these programs are not experiencing major problems, it is quite possible for programs written to run in this privileged ring to adversely affect how other programs run.
There have been reports of different problems associated with each of these programs, ranging from locked up computers (infrequent) to static coming through the system speakers (not uncommon). Personally, I have not experienced any of these situations over the last few months of use. For the most recent reports of problems, I find that searching DejaNews with "program name problems" yields a good amount of reading material. Because it has been available longer, there is more documented information for the various versions of CpuIdle (Both positive and negative).
To quote Andi's Home Page, the home for CpuIdle, "CpuIdle runs an HLT command in an idle-priority thread under Windows95. This allows modern microprocessors like Intel P5/ P5 MMX/ PPro/ PII, AMD K5/ K6 or Cyrix M1/ M2 to save power and stay cool. It helps to create a more stable system when CPU temperature is an issue with overclocking and help the environment."
Included in the download is DosIdle, which performs the same function in DOS mode as CpuIdle does in the Windows environment. There have been a few versions of CpuIdle over the past few months. I have tested versions 2.4, 2.52, 2.52b (2.53), 3.0, 3.01, 3.02 and 4.0.
Note: CpuIdle2.X also comes as a "Load Only" version that does away with the tray graphic. While it does use less memory, the virtual device driver is the same as in version 2.4. For that matter, it is the same in all versions I tested (except version 3.01), if one is to judge by the date stamp of 2-20-98. The "Load only" version's results were the same as version 2.4.
Note: CpuIdle3.0 also can be run as a control panel plug in. It uses less memory than the graphic version, but, as far as I can tell, benchmarks the same.
To again quote the CpuIdle home page, "v3.01: Minor Update - removed direct95.vxd which is no longer needed." Short and sweet. However, this is not the only virtual driver change. The Cpuidle.vxd file has increased in size from 8 KB to 9 KB. This applies only to version 3.01. The "load only" version still retains the earlier Cpuidle.vxd file. As has always been the case, the program installed easily and with no problems in both the control panel mode or with the icon in the tray.
In past tests, I had experienced some discrepancies in the WinBench98 CPUmark32 scores. This version of CpuIdle's results never varied more than 1% from test to test. However, this version still lacks the ability to cool as well as some earlier versions of CpuIdle and all versions of Waterfall and Rain while transferring files.
The results of CpuIdle 3.01 "icon" mode and control panel mode were virtually identical in the various benchmarks, so I combined both results into one entry for each category. For the "memory used" section, I have given results for each interface.
Among the many good things to be said about CpuIdle is that an update always seems to be in the works. Again, Mr. Goetz has listened to the program users and come out with a new version. According to the "what's new" text, the changes include:
Version 3.02 (July 8th, 1998):
- changed the icon to something for friendly for the eye (due to popular
- changed handling of VxD handle (now correctly unloads if CpuIdle is closed)
- finally decided to add -nosplash option (no guess what this does?)
- updated FAQ (please, do read at least the new items)
Both the vxd and dll remain the same as version 3.01. The performance results were also virtually the same as version 3.01. This includes the tendency to run a bit warmer than other versions of CpuIdle, Rain, and Waterfall when accessing the hard disk, CD-ROM, and floppy drives.
The results of CpuIdle 3.02 "icon" mode and control panel mode were virtually identical in the various benchmarks, so I combined both results into one entry for each category. For the "memory used" section, I have given results for each interface.
Because my system has evolved since I started these tests, setting up the system to the original specs used for the earlier tests has become quite a chore. This now involves removing a couple of cards, some RAM, changing BIOS settings, and drivers. This will be the last test using the old configuration. With the introduction of the next update for any of the programs, I will just test the most recent version of each of the three programs. The same format will be followed, but the results will reflect the new hardware I have acquired.
Mr. Goetz has been busy. This is the second update in less than a month. CpuIdle 4 is listed as a major revision of the popular CpuIdle series. The new changes include a "run as a Windows service" mode, a "bug fix" for the control panel mode, added graphic interface for the start up mode, and the readme.txt is now written in html as cpuidle.htm. CpuIdle is now listed as "uncrippled 30 day shareware." Considering the benefits of the program, the $10 price tag is quite reasonable.
Unfortunately, on my system, putting a check mark in the "run as a windows service" box did not enable CpuIdle the next time I rebooted my system. Running CpuIdle from the command line, start up menu, or "control panel mode" worked as they should. I have been in contact with Andi and It's my opinion that we will see a fix for this shortly.
CpuIdle 4.0 Update
The "run as a windows service" feature has been removed from version 4.0. This change is in place with versions of 4.0 with a cpuidle.exe file dated 8-11-98. The previous version had a date of 8-5-98.
One other note concerns the use of CpuIdle in conjunction with Qemm97. Using both programs at the same time resulted in CpuIdle showing that it was active, but not reducing the CPU temperature. I don't know if this phenomenon is limited just to my particular system, but if you run Qemm97, be advised.
Other than these two situations, CpuIdle 4.0 ran well, providing cool idle temperatures with no negative interaction with any of the many programs I use.
As I said in the CpuIdle 3.02 update, my system cooling system has change a bit. I now am using a home made heat sink with a large (70mm) fan and a couple of new thermometers to read CPU temperatures. I have also flipped the power supply fan over to exhaust and the auxiliary fan to intake (article coming soon). This has made a big difference in the ability to keep my system cooler. The charts for this configuration are shown in blue. I have also added some memory and a new network interface card, but these changes had little or no effect on the program's scores.
For those of you who are familiar with the rest of the introduction material and the original tests, please check the updated system information here, before proceeding to the new tests.
To quote an earlier Waterfall readme, "...(a) little utility that works almost exactly like CpuIdle, only it uses less resources and does not require VxDs or DLLs. Because it doesn't need to issue hundreds of VxD calls per millisecond, it runs even cooler than CpuIdle! Windows 95/98 only." Hype aside, Waterfall is another great little program that performs well. No provision is made for DOS mode. I have tested versions 1.2, 1.22, and 1.23. Waterfall is available from Leading Wintech at the Brotherhood of the CPU site.
Waterfall Pro 1.01 - Update
Leading WinTech has introduced a new version of their Waterfall series of cooler programs. Waterfall Pro 1.01. With the superior results I received in testing Waterfall 1.23 with the Celeron A / Abit BH6 combination, I was very interested in how this new offering would benchmark.
I had the opportunity to beta test WFP and observe its evolution as the program was being refined. From the beginning, this program was designed with much more in mind than enabling the HLT statement which reduces power consumption and allows the processor to "sleep" while idle. With the inclusion of LM75 and LM78 / LM79 chips on many newer motherboards, software has been able to monitor these chips for temperature, voltage, and fan speed. WFP was written to include the monitoring of these values. To set up Waterfall Pro, the user is asked to supply the program with name of the processor and motherboard so the proper motherboard monitoring chip can be read. As the support module for my BH6 motherboard has yet to be released, (soon, as I understand it) I used the LM75 / LM78 setting. Because this setting is not written for my board, there were a couple of readings that were displayed incorrectly. Core voltage was listed as 1.48 volts rather than the 2.11 listed in my BIOS. The fan 2 reading was absent as it is in many of the motherboard monitoring programs. (On some other programs, I find the fan 2 RPM readings in the fan 3 box.) However, even without the correct module installed, the 5 and 12 volt readings were very close to matching the BIOS's readings, being within .1 volt.
The ABIT BH6 motherboard temp sensor, (the LM79 chip) has a stated accuracy of ± 3° C (by National Semiconductor). Waterfall Pro's reading of the chip and the chip itself were well within this spec, reading +1° C between 25 and 42° C (77 and 108°F). Alas, as nice as it is to be able to read the motherboard temperature, with the placement of the LM79 chip near the bottom of the board on the BH6, the only information you are able to read is the case's lower interior temperature. This doesn't give you much information on what is happening to the CPU. It would have been nice if Abit would have included a temperature probe for the processor. (Abit, are you listening?)
Along with the CPU optimization feature found in the earlier Waterfall versions, optimizing of the file system and virtual memory is now included. The file system optimization switches the roll of the computer to a Server rather than a Desktop computer when the system has 64 MB of RAM installed. With the computer in Desktop mode, VFAT allocates approximately 10K of memory for finding recently accessed folders and files. In the Server setting, VFAT allocates approximately 40K of memory for this task, which allows for more files to be accessed quickly. This is a well known tweak that is widely used and may be set manually from the system properties dialog box in the control panel.
The memory swap optimization feature will set and place the virtual memory to the fastest disk that is next in line. On my system, with 64 MB of RAM, Waterfall Pro set the VM to a minimum of 64 and a maximum of 128 MB and placed it on the slave drive of the primary IDE controller, which is slightly quicker than my primary master drive. This is also a well known performance enhancer.
There is another new feature called CPU Throttling, which is designed to limit the processor's ability to attain 100% usage to help keep the heat generation to a minimum. The feature does what it is supposed to and does limit the processor in the incremental steps shown in the program's interface (as close as I could tell by my benchmarking). For my purposes, trying to O/C (or is that over-bus?) one step more to 504 MHz from the 464 I'm now running at, heat is not the issue, so the throttling feature was tested for function, but not used for these tests. Whether or not this feature will be useful for overclockers will take more than a few benchmarks to determine, but it is an interesting concept and another tool to try.
The only "feature" of the program I could complain about was the minor annoyance a pop-up screen for the WWC Channel found on the freeware version. The pop-up is not present in the $15 shareware version of the program. However, my feelings on this are the same as with CpuIdle or any other shareware I have found useful. If you're going to do more than just test the software out, buy it. Supporting the people who write the innovative software allows and encourages them to continue producing programs for us.
The system configuration used for the following tests may be found here. The tests start here.
Rain is another offering from the folks at Leading Wintech, who seem to delight in the "late night TV commercial" approach to promoting their product. The following is the introduction from the readme. "Especially designed for overclockers (or for severe heat conditions), Rain is the latest CPU cooler from Leading Wintech. What makes it different from other CPU coolers (such as Waterfall and CPUIdle) is the fact that it has a different concept: 'Extreme Cooling!'" I have spent the least amount of time with this program, but it appears to work well. As for the claims of "extreme cooling," that remains to be seen.
Tested System 1(gray
Abit Ax5 board
AMD K6 233 o/c'd to 250 (3X83 MHz)
Quantum Fireball SE 3.2
Quantum Fireball ST 3.2
Matshita CR-574 CD ROM
Fujitsu SDRAM 32 MB
Inwin A500 case with added 3.5" exhaust fan
CPU Cooler heat sink w/ ball bearing fan
MS Win95 OSR2
Case cover removed
Room temp 72° F (until recently!)
Set Up 1
With my temperature sensor module assembled, I needed a way to attach the probe to the processor's heat sink. This required a slight modification to the cooling fins on the heat sink. This was accomplished with the aid of a Dremel tool with a miniature die grinder bit. The distance between the fins was close enough to the probes diameter that very little modification was necessary. When the "pocket" had been formed, the surface was sanded and buffed to provide as much contact area for the probe as possible. When I achieved a nice, tight fit, I coated the contact areas with a thin coat of thermal compound to ensure good heat transfer between the heat sink and probe.
Tested System Update (blue charts)
Abit Ax5 board
AMD K6 233 o/c'd to 250 (3X83 MHz)
Quantum Fireball SE 3.2
Quantum Fireball ST 3.2
Matshita CR-574 CD ROM
Fujitsu PC 66 SDRAM 64 MB
Inwin A500 Case p/s fan exhausting
with added 108mm intake fan (front lower)
Home made heat sink with 70mm fan
MS Win95 OSR2
Case cover removed
Room temp 80° F plus
See Thermometer article.