Evaluation of the VIA C3 800 Ezra

Tested in an existing passively ventilated system (System II). The background for the evaluation is curiosity as to what possible difference the VIA Ezra C3 800 could make, compared to an Intel Celeron II 700 in an identical system. The usability of the C3 Ezra in audio applications where also examined.

>LUD [13.12.2001]

1. Introduction :

VIA kindly supplied me with a test sample of the C3 800 Ezra and accompanying CoolerMaster Fanless Heatsink. I received my shipment & opened it  with some excitement. The CPU looked esthetically pleasing with it's large golden heat slug. Kind of like my old Intel P-233 MMX.

My first worry was whether my MB ( ABIT BX-133 RAID) would support the chip, as it was not mentioned in VIA's list over compatible boards. It turned out that it worked just fine. Chip detected as : VIA CyrixIII 800B MHz, and Vcore voltage was automatically set correctly to 1.35 V by BIOS/softSETUP. Had some trouble finding info on multiplier settings, turned out
the info was printed on the heatslug of the processor :)

I had literally a hard time with fitting the Cooler Master Fanless Heatsink. Need to be a HE-MAN to fit the retention clip to the CPU. Came to the frustration threshold where I gave up, and decided to fitted the custom made DD element.

Later I got worried about a possible poor thermal connection because I wasn't able to detect the expected amount of heat. So I solved the CM problem by pre-bending the retention clip so that I didn't have to use excessive force to fasten it. I also found that a wide screwdriver was a useful tool for engaging/disengaging the retention clip.

I also decided to remove the entire goopshit/aluminum foil that was underneath it. It totally sucked. Gave the element a quick polish, and applied thermal paste. This improved the heat transfer. How much I don't know as I couldn't measure the temperature, but a change in CPU load was noticed quicker on the surface temperature of the element.

2. Outline of the testing :

The first and most important question I had was whether the C3 Ezra was usable in audio applications, and how well it performed. This was tested using real programs; see [ #A brief testing of audio performance : ] for tests & results.

Secondly, there is a synthetic performance test using different Sandra SiSoft benchmarks, see [ #Synthetic Benchmarks : ]

All the test above compares results between the C3-800, CeleronII-700 & P3-733. Two almost identical systems where used. The first used for the C3 & Celeron is descibed in note(1) ; the other that where used for the P3 is decribed in note(2).

Thirdly I move on to the most interesting part, which is the testing and comparison of the C3 in system 2, versus the CeleronII-700. See [ #Testing the VIA Ezra C3 800 vs. Intel Celeron II 700 in System 2. ]

Finally I will sum up all of my dealings in [ #Summary : ]

Notes :

note(1) : Measurement done on a ABIT BX-133 MB (BX 440 chipset), with a AOpen soundcard., OS Win98SE. Most stable hardware settings/defaults.

note(2) : Measurement done on  an ABIT BE-II MB (BX 440 Chipset), with a Creative soundcard, OS WinME

note(3) : With optimized hardware acceleration settings in BIOS. Min cache latency etc. Less stable.

2.1 A brief testing of audio performance :

FruityLoops v3.3 :

As a basis for the testing I used my favorite application, FruityLoops (v3.3) and a random project I just happened to be  working on at the time. (RP-14-TESTPATTERN.flp). I used the visual observations of the applications CPU meter as basis for judgement.

The project contains : A total of 13 different instruments, of which 9 are sample based, 3 are generators (Simsynth, Buzzgenerator & TS404) & 1 MIDI. A total of 4 effects, where internal FL effects are : compressor, reverb & filter; and the external DX effect is the Ohm'Force Frohmage filter plug-in. A number of patterns are organized together in the playlist; and looped. Average polyphony, or number of notes played at the same time, fluctuate around 50 (if my memory serves me right).

Used the default audio settings, 200 ms DirectSound buffer, polling & hardware buffer of the soundcards.

! Note that the Celeron & the C3 ran under the exact same conditions, but the PIII measurement where taken on a different system with slightly different settings (OS/Soundcard)..

Conclusion : As can be seen from the above table, the VIA C3 performed significantly worse than the Intel processors. In my subjective judgement, I would rate the performance of the C3 at about half that of the Celeron (at its default settings). It's hard to tell. Anyway, despite the weak performance the C3 was fully usable for medium sized project with an optimized configuration (buffers etc.). 

mp3 playback performance :

Test Program : Winamp (v2.76) : Used this program to play a standard 128 kBit/s, 44.100 Hz, Stereo .mp3 file.

Had some difficulties getting a reliable reading of CPU usage, but the most reliable reading I got was a 12-13 % usage reading from the system monitors in Ontracks Fix-It program (v.3.0). The strange reading gotten when no processes where running was a 50 % reading. Same reading as reported by the Windows system monitoring tool, reporting kernels CPU usage in %.

Switching to the Cel 700 : The readings I got where around 9-10 % CPU usage.

Anyway, the readings I got where so unreliable that they are hardly worth mentioning.

Conclusion : The C3 played mp3's without effort, and I didn't notice any significant difference between the C3-800 and the Celeron-700. Even a P-233 MMX plays mp3's without a hitch so this didn't excactly come as a big surprise.

mp3 compression :

To test the mp3 performance the time used to compress a CD quality .wav file to .mp3 format was measured.

Test Program : .mp3 Producer (professional) v2.1 (Build 47)

Source file : Prodigy's "Poison", 69 340 kB, 44100Hz Stereo PCM, wav file, 6:42 playtime.
Output file : 6 276 kB, Quality : 128 kBit/s, 44.100 Hz, Stereo

Conclusion : Not much to say. The above numbers speak for themselves. They suit the impression that I've got from the FL 3.30 testing.

Entries in row 3 & 4 of the table is really an irrelevant side step, but I got carried away in this bencmarking thang. What surprises me is the how close the performance of a overclocked Celeron system with optimized HW (for speed); is to a non optimized P3 system in this case.

2.2 Synthetic Benchmarks :

Test Program : SiSoft Sandra Professional, Version 2001.5.8.11

These are synthetic benchmarks, and of lesser value than the real application testing. Something the estimated PR mark clearly shows. The P3 is obviously a much more powerful processor than the C3, yet the C3 scores higher on the PR mark.

Conclusion : The results for the Celeron & the PIII are very similar. The main difference is in the Memory benchmark. Here the C3 outperforms the Celeron, as expected with it's double FSB speed.

2.3 Testing the VIA Ezra C3 800 vs. Intel Celeron II 700 in System 2.

# Intro

I've been curious to see how the difference between the Celeron 700 & the C3 800 would be in terms of temperatures in System 3 (Fanless system PSU 5). Unfortunately I've not been able to get a thermal readout of the CPU, and I'm yet not certain if I can get an efficient heat maximizing to work with the same utilities that I've used before (CPUIdle). The comparisons won't be perfect, but will give an indication.

There are other differences as well. As I understand it the C3 regulates it powerconsumption by it self, I'm not certain if the ACPI have any effect. Anyway, it's enabled on MB and in windows for these tests.

# Note on measurement limitations & trouble :

Getting a proper CPU temperature readouts for the C3.

Idling of CPU. Tcpu(air) temperature off 35 C @ 21.0 Camb, even in
suspended state.

Accuracy of the measurments.

2.3.1 Definition of measurement states : ( Limited to the scope of this document ).

It is distinguished between the powerconsumption of three different system states. ACPI is enabled for all states. Theses three states are [IDLE], [NORMAL] & [POWER].

1. The [IDLE] state is the state of the system with no load, powerdown on HDD & screensaver is active (i.e. video off).
2. The [NORMAL] state is when there is no load, HDD is spinning idle & screensaver is inactive (i.e. video on)
3. The [POWER] state is when the system is consuming as much power as possible by running different test programs. HDD is spinning idle & screensaver is inactive (i.e. video on)

2.3.2 Test methods & tools :

To maximize the powerconsumption of the system, the program CPU STABTEST6.0 (build: 154) is used. This program have two modes of maximizing powerconsumption :

1. The Normal mode (referred to as test method 1) runs different test algorithms on CPU and also accessing peripheral equipment such as drives.
2. The CPU warming mode (referred to as test method 2) only seeks to maximize CPU temperature/powerconsumption.

2.3.3 Observation points & location :

Two probe locations where available and of particular interest; Tpsu & Tmb.

1. The psu probe (Tpsu), an external sensor located inside PSU(5). When viewing the cabinet with x-ray eyes from above, it is located at the highest point, and as close as it can get to the cabinet center in that plane.
2. The second probe (Tmb), is an onboard sensor that measures the temperature of the motherboard. This measurement typically correlates to the ambient temperature with a fixed difference.
3. Ambient temperature (Tamb) is measured at the base-level of the cabinet, located on a desk in a small room.
4. The measurement of (Tcpu(air)) is made with a probe located in the air stream at the center of the Cooler Master element. Note ! It's the temperature of the escaping air that is being measured, and not the temperature of the CPU itself.
5. Tcpu(diode), refers to the temperature mearured by the internal CPU sensor. Unfortunately I was unable to get valid readout for the C3 from the MB/program that I where using. The measured results are noted however, as they carry some information ... seems to be a scaling problem somewhere.

From the Celeron testing it was observed that the Tcpu(air) tracked the Tcpu(diode) with only a 1 degree C difference ( This was around 60 C however, and at 40 C the difference has grown to 3 C). This can be used to say something about the internal temperature of the C3, even though I was unable to measure this parameter properly by software.

All temperature measurements given in C, Centigrades.

2.3.4 Environment :

System 2 (or system II) consists of a cabinet highly optimized for passive ventilation. The PSU(5) is a 200 W custom fanless unit. ABIT BX-133 Motherboard, 256 MB PC-133 CAS-2 Memory, IBM 5400 RPM 15 GB Deskstar, HP-8100 CDRW, Floppy, AOpen Soundcard, Matrox G-200 AGP card (8MB). Running Win98 SE. Unless otherwise indicated.

2.3.1 Measurement of the VIA Ezra C3 800 :

Settings : The VIA Ezra C3 800 @ 133 MHz FSB,  1.35 Vcore, with CoolerMaster Fanless Element .

* assumed temperature (update when difference at around 30 C is known)

[IDLE]

Testing with method X for 3 hours, gives the following results :

Raw : Tamb 20.0 Tcpu(air) 32.0 Tcpu(diode) 33.0* Tpsu 42.0 Tmb 22.5

[NORMAL]

Testing with method X2 for 3 hours, gives the following results :

Raw : Tamb 20.7 Tcpu(air) 36.7 Tcpu(diode) 37.7* Tpsu 46 Tmb 25

Normalized : Tamb 20.0 Tcpu(air) 36.0 Tcpu(diode)* 35.0 Tpsu 45.3 Tmb 24.3

[POWER ]

Testing with method 1 for 7 hours, gives the following results :

Raw : Tamb 17.9 Tcpu(air) 58.2 Tcpu(diode) 70 Tpsu 48.0 Tmb 22.5

Normalized (20C amb) : Tamb 20.0 Tcpu(air) 60.3 Tcpu(diode) 72.1 Tpsu 50.1 Tmb 24.6

?? Ignore this measure as it is to different.

Testing with method 1 for 8 hours, gives the following results :

Ambient temperature range for test (min/max) [20.0 .. 22.1]C

Raw : Tamb 22.1 Tcpu(air) 59.3 Tcpu(diode) 71 Tpsu 51.0 Tmb 27.5
Normalized (20 C amb) : Tamb 20.0 Tcpu(air) 57.2 Tcpu(diode) 68.9 Tpsu 48.9Tmb 25.4 

Testing with method 2 for 2 hours, gives the following results :

Raw : Tamb 21.1 Tcpu(air) 59.1 Tcpu(diode) 71 Tpsu 49.0 Tmb 26.0

Normalized (20C amb) : Tamb 20.0 Tcpu(air) 58.0 Tcpu(diode) 69.9 Tpsu 47.9 Tmb 24.9

Testing with method 2 for 10 hours, gives the following results :

Ambient temperature range for test (min/max) [19.9 .. 22.5]C

Raw : Tamb 20.0 Tcpu(air) 58.9 Tcpu(diode) 71 Tpsu 48.0 Tmb 25.0

oo It seems that a stable state is reached after two hours. Insignificant changes in Tpsu & Tmb after this.

2.3.2 Measurement of the Intel Celeron II 700 :

Settings : The Intel Celeron II 700 @ 66 MHz FSB, 1.70 Vcore, CoolerMaster Fanless Element (exceptions explicitly  stated).

[IDLE]

Testing with method X for 3 hours, gives the following results :

Note ! This test is performed with the DD element on CPU.

Raw : Tamb 22.0 Tcpu(diode) 28.0 Tpsu 44.0 Tmb 25.5 (assumed values after 3 hrs.

Normalized : Tamb 20.0 Tcpu(diode) 26.0 Tpsu 42.0 Tmb 23.5 (assumed values after 3 hrs.

Environment : HDD powerdown, screen saver on, ACPI on

[NORMAL]

Testing with method X2 for 3/5 hours, gives the following results :

Raw : Tamb 22.3 Tcpu(air) 39.5 Tcpu(diode) 36.5 Tpsu 47.0 Tmb 27.5 (after 5 hrs )

Normalized : Tamb 20.0 Tcpu(air) 37.2 Tcpu(diode) 34.2 Tpsu 44.7 Tmb  25.2

Environment : HDD poweron, no screen saver, ACPI on, Idle in OS.

[POWER ]

Testing with method 1 for 8 hours, gives the following results :

Ambient temperature range for test (min/max) [21.9.. 23.0]C

Raw : Tamb 22.9 Tcpu(air) 68.0 Tcpu(diode) [69..74] Tpsu 52.0 Tmb 28.0
Normalized (20 C amb) : Tamb 20.0 Tcpu(air) 65.1  Tcpu(diode) [66.1 .. 71.1] Tpsu 49.1 Tmb 25.1

Testing with method 2 for 2 hours, gives the following results :

Note ! This test is performed with the DD element on CPU.

Raw : Tamb 22.8 Tcpu(diode) 66° Tpsu 53° Tmb 28° C

Normalized : Tamb 20.0 Tcpu(diode) 63.2° Tpsu 50.2° Tmb 25.2° C 

Testing with method 2 for 10 hours, gives the following results :

Note ! This test is performed with the DD element on CPU.

Raw : Tamb 22.9 Tcpu(diode) 67.0° Tpsu 54.0° Tmb 28.5° C

Normalized : Tamb 20.0 Tcpu(diode) 64.1° Tpsu 51.1° Tmb 25.6° C

Note ! Prolonged testing based on continuation of the first run of this test.  The changes in relevant temperatures after 2 hours are insignificant.

Commenting on the accuracy of the above test results : A linear normalization of the measured temperatures to an ambient temperature of 20 C gives some error. There is also inaccuracy in the measurement results due to limited resolution on sensors. The maximal error for MB sensors are is in the range +/- [0.25 .. 0.5] C, for the external thermometer (used for Tcpu(air)) neglectable.

The measured differences are so small that they are insignificant when taking measurement errors into consideration.

2.3.3 Analyzing the results of the measurements :

A table for comparison :

Conclusion : The difference in measurments are surpassingly small. When taking the error due to measurement limitations into account, the only result that standout is the Tpsu measurement for the POWER state using method 2. This method targets the CPU processor specifically, seeking to maximize the powerconsumption. A 3 degree C difference here, is the same as the difference between the system in it's idle and normal state.

On the whole I can't see any significant difference between the C3 & Celeron in terms of temperature under normal use. Only under extreme worst case conditions is there a noticeable difference. These results probably says more about the efficiency of the system 2's passive cooling, than the thermal properties of the C3. In a smaller standard cabinet, the magnitude of differences would probably be much more visible and critical.

3. Summary :

# Initial Impression :

Noticeable speed increase in standard windows applications compared to a Celeron 700.

About half the performance in Fruity-loops, particularly on effects that rely heavily on floating point operation, such as reverbs.

Troublefree CPU change from Celeron 700 to C3 800 in a Win98SE system.

Surprisingly little difference in temperatures.
PSU temperature, just a couple of C's down. MB temps unchanged, same as normal.

# Pro's'n Con's :

Trying to sum up pro's & con's of the VIA C3 Ezra 800.

+ve Good things about the Ezra C3 :

• Low TDP (Thermal Design Power =  maximal power consumption). With a TDP of 8.5 Watts (unconfirmed) it compared well to Intel's Celeron2/PIII-800  with TDP = 20.8 Watts & or AMD's Duron-800 with TDP = 35.4 Watts. ( On the side. AMD is a nightmare when it comes to powerconsumption; and Intel seem to have caught up with them on their P4 CPU series.)
• Robust & solid packaging.
• Large thermal interface area.
• Independent of ACPI / Thermal Management SW, as CPU implements it's own energy saving strategy.
• 133 MHz FSB Speed. Matches Intel P3 CPU/Memory transfers.
• Overheating handeling. Apparantly the CPU handles temporarily overheating without failiure of operation. This can be an advantage in situations where the system becomes temporarily overheated, but not a viable long term solution; as eventhough it operates it still "burns" of the expected lifetime of the unit. Failiure to "fail", can make it more difficult to detect problems in the heatsink junction.
  Intel CPU's have a built-in overheating protection mechanism that halts the processor once this treshold is reached, which is "overprotective" if the overload is only temporarily.
  AMD as I understands it operates much like the VIA chip, the main difference beeing that AMD chips consumes so much power that some of their chips fail destructivly in smoke without a heatsink.
  The VIA chip consumes so little power that it can handle this more gracefully.
• Fanless cooling for dummies :) When bundled with the right heatsink there are few things that can go wrong. Not even any wires to connect at the right place :)

- ve Downers :

• Weak Floating Point performance.
• A hard-to-fit retaining mechanism on the CoolerMaster heatsink caused some trouble.

I've been twisting my brain to make up an opinion of the Ezra C3. It's a strange bird. It represents little new in  terms of 'hardcore' silencing from my point of view. The main benefit I can see with this CPU is that, when it is bundled with a fanless heatsink it represents a "Fanless cooling for dummies" kit. Very few things can go wrong, it's just plug'n'go.

The built in power management takes care off the temperature issues, and there is no need to be concerned with thermal issues. This could be an advantage on motherboards without ACPI, or on systems where it's not working properly and no other thermal management software is effective. The low powerconsumption also sets a hardware roof for how much power that can be consumed by the system, making it easier to make systems with fanless PSU's. The integer application performance is also good compared to the Celeron, although the Celeron blows the socks of the Ezra when it comes down to floating-point operations.

For audio processing purposes the downside is that floating point performance is about half that of a Celeron. It still gives sufficient power for normal use though, and is powerfull enough to drive a (MIDI/HDD recording o.s.) setup, where most of the sound processing is done externally.

4. Links :

http://www.viatech.com/en/viac3/C3_silent.jsp - VIA's silence related page.

Note ! Lack a link to the C3 800 Ezra Data sheet.

http://www.dansdata.com/c3.htm - Link to a more extensive C3 800 Ezra test.