 |
|
 |
 |
|
Stress Testing PC Components
Objectives
This test procedure is intended for the most thorough analysis of operating stability of separate PC components (in a certain combination with other components) as well as of assembled computers, mostly desktops. Version One of the test procedure (1.x) focuses on testing the following PC components (separately or in a package):
- CPU
- RAM
- 3D video accelerator
- Power supply unit.
Each of them is tested in a number of synthetic tests and real applications for stability under the heaviest load, maybe even too heavy for real situations.
Additional objectives of our tests:
- Compatibility of an object under review with a certain infrastructure (provided or specified by a customer)
- Only for a processor and a video accelerator - temperature conditions and efficiency of a cooling system, operation of the integrated thermal protection
- Maximum possible overclocking level of the PC component under review (to evaluate its safety margin).
Results
Result of testing a given PC component or a whole PC is a conclusion about serviceability degree of this object. If any failures are detected, a conclusion is drawn on the degree of compatibility between the object tested and the test platforms provided. Reasons for such problems are pointed out or assumed, if possible. A client may order extended tests to determine a reason for these problems.
Results of the additional tests (in compliance with the above-mentioned possible objectives):
- Accented points from conclusions on compatibility with a given infrastructure
- Report on maximum temperature readings (with comments whether these temperatures are acceptable), facts on operation/nonoperation of the built-in thermal protection
- Depending on a flexible procedure for overclocking tests (it will come as a separate document), report on the best (for a given sample) and the worst (for the whole shipment) overclocking results.
Test Procedure
A testbed for testing a ready PC is this computer itself. When one of PC components is tested, a platform is installed into the testbed, a necessary batch of tests is executed, readings are taken, and then the testbed configuration is changed for another iteration. The set of platforms is chosen in accordance with testlab capacities and client's requirements.
Here are the main rules for selecting tests:
- Maximum (or near maximum) efficiency
- Combination of synthetic tests and real applications
- No redundancy.
Synthetic tests generate maximum load on the component tested, while real applications allow to determine performance of a given sample in the heaviest conditions that may happen in real life.
1. Testing a processor
- Synthetic applications for stress testing processors are characterized by maximum speed of executing instructions that heat computing units of a processor. An excellent example of such an application is S&M in "FPU test" mode (two instants of this subtest for dual core models and Intel Pentium 4 with Hyper-Threading). In this case a processor actually dissipates maximum heat. CPU test time can be changed in the range from "quick" to "long" for shorter or longer test duration.
- Real applications can heavily load a processor, depending on an optimization and computation algorithm, though they are outperformed by special synthetic tests for obvious reasons. One of the most efficient benchmarks for stress testing processors is CPU RightMark and its Stability Test with FPU and Formula Optimization parameters (two computation threads for dual core models and Intel Pentium 4 with Hyper-Threading). Test duration is specified directly in settings, the output window demonstrates a relative performance level, which helps track down operation of built-in thermal protection of a processor.
Minimum recommended test configuration (to provide relatively credible results): 1 hour for Test 1.a + 2 hours for Test 1.b.
Maximum recommended test configuration (to provide reliable results): 22 hours for Test 1.a + 2 hours for Test 1.b with performance level control.
2. Testing memory
- Complex test S&M in the memory check mode thoroughly checks for hardware defects in memory modules/chips as well as tests data exchange interface with the processor - four subtests with different algorithms. Additional cycle of this test with a variable sample ("another sample" option) increases chances for detecting hardware failures. Complexity of this test can be controlled within the quick/long range for shorter or longer test duration.
- Memory Stability Test from RightMark Memory Analyzer can be used as an additional memory test, it's a fast and thorough test of memory stability at the specified frequency and timings. Test duration is determined by a memory walk stride size (it must be maximum to increase the test quality), it's also specified directly by a number of passes.
Minimum recommended test configuration (to provide relatively credible results): 1 hours for Test 2.a (Subtests 2 and 3) + 2 hours for Test 2.b.
Maximum recommended test configuration (to provide reliable results): 22 hours for Test 2.a (all subtests, both samples) + 2 hours for Test 2.b.
3. Testing a video accelerator
- Synthetic tests of 3D image quality (in Scan for Artifacts mode), built into ATI Tray Tools and ATITool, allow to detect problems in a given video card by providing maximum load on a GPU. Test duration for ATITool is specified directly in settings ("Artifact Scanning'). What concerns ATI Tray Tools, the test process can be interrupted only manually.
- Running a long demo in modern games (Doom 3, FarCry, F.E.A.R.) at 1024x768 and 1600x1200 with maximum video quality is a good tool to detect operation failures, generating maximum real load on a video card. It's up to a testlab engineer to detect visual artifacts in the process of testing. Depending on the type of a video card, this or that game may prove to be more or less problematic, so we recommend using all the three games (or even an extended list of games). Test duration is determined by your choice of a demo and a number of its cycles.
Minimum recommended test configuration (to provide relatively credible results): 1 hour for Test 3.a (0.5 hour for each subtest) + 2 hours for Test 3.b (1 hour for each of two selected subtests depending on a GPU type).
Maximum recommended test configuration (to provide reliable results): a total of 18 hours for Test 3.a (both subtests should be run for more or less equal periods of time) + 6 hours for Test 3.b (2 hours for each subtest), periodic visual control of the graphics quality.
4. Testing a power supply unit
- The most power consuming components of a state-of-the-art PC are a processor and a video card, so a PSU stress test must use them to full extent. S&M in the PSU test mode is the most suitable program for stress testing (the second subtest that renders a scene), as it generates maximum load on a processor (Test 1.a). Subtest One may also be useful, it provides variable load on all PC components. Test duration is specified by a number of test cycles.
- Running tests for a processor and a video card simultaneously will be a good practical test for a power supply unit. Test 1.b in combination with all Subtests 3.b (in turn) may give a reliable picture. Test duration is specified separately in each test.
Minimum recommended test configuration (to provide relatively credible results): 1 hour for 4.a + 2 hours for Test 4.b (choose one of Subtests 3.b depending on a video card).
Maximum recommended test configuration (to provide reliable results): a total of 18 hours for Test 4.a (2 hours for Subtest 1 + a total of 16 hours for Subtest 2) + 6 hours for Test 4.b (2 hours for each Subtest 3.b).
You can choose any tests for any PC component (or a whole PC), this document contains recommendations for cases, when tests are strictly limited in time.
|
|
|
|
|
