Quick, overall system performance suite?

[teotwawki]

Hi all, I'm new to this thread so perhaps you've already thought of this & rejected it for some reason but why not make this as close as possible to the Windows Experience Index reported by Microsoft's Windows System Assessment Tool ?
It gives a user friendly score for each major area & a final over-all score. It's also pretty well documented (see links from http://en.wikipedia.org/wiki/Windows...ssessment_Tool or just search for WSAT) though the exact details of the tests are probably protected. However, as long as the PTS version tested the same aspects (E.g. using the new Unvanquished test for 3D) it could then use a weight to produce a comparable figure to what WSAT would give.
Although the normal user's view of WEI is a pretty GUI summary, it actually has a command line that allows you to run individual tests which would make tuning the PTS version to match it a lot easier.
Microsoft's WEI was introduced in Vista & seems to have abandoned in Windows 8. It was never back-ported to XP and I've never seen an advert for a PC mention it's WEI score. But I'm sure they did a lot of research into what weights to put on each aspect and it does do exactly what this thread aims for, so why not aim for comparability? You could call this the PTS-EI.
As a side note, if Microsoft have abandoned it, they might be willing to disclose more detail.

[mendieta]

Yeah there are a few tests which I have found that are relatively quick and have worked on the few systems I am working with. I'm not sure how well they test GPU though... Renderbench, gtkperf, and jxrendermark. have you ever looked into any of these?

edit: these tests have relatively small download sizes as well.

They all seem to be 2D tests but I'm not sure how well they rep performance

Exactly, these are 2D. AFAICT, qgears is exercising opengl, so it gives an idea of the 3D capabilities. Glmark2 is another simple openGL test, but it is extremely simple and it takes a long time (why does that head need to turn so many times?? One turn would be more than enough).

Overall, I continue to think that if this idea ever works, it's best to have a quickbench-cli and a quickbench-gui. For the former, something like quickbench could give a balanced view: there is a single threaded test (scimark2), a multhreaded (7-zip) and a non-trivial disk stress that seems to capture real world speed.

For the GUI, perhaps some of the tests in gtkperf or similar could do the 2D part, and then I guess a 3D test would be challenging. I am not sure qgears2 is a good test, even in a synthetic sense.

Cheers!

[channon]

Yeah there are a few tests which I have found that are relatively quick and have worked on the few systems I am working with. I'm not sure how well they test GPU though... Renderbench, gtkperf, and jxrendermark. have you ever looked into any of these?

edit: these tests have relatively small download sizes as well.

They all seem to be 2D tests but I'm not sure how well they rep performance

[mendieta]

Im looking for a good graphics test that will install on linux. I like the quicktest v4 but not the v5 that test does not always install on every machine

Yep. I think it's an issue with the dependencies, you need this package

https://launchpad.net/ubuntu/raring/+package/libxrender-dev

And the test installed for qgears2 doesn't seem to be aware of that. It is hard to find a simple test that will at least assess very crudely how 'fast' the IGP is, that also runs fast, and that doesn't involve a huge download. Sigh!

Anyways, many thanks for chiming in!

[channon]

Im looking for a good graphics test that will install on linux. I like the quicktest v4 but not the v5 that test does not always install on every machine

[mendieta]

Mmm ... I am thinking that including graphics might not be great, in that a server may only be CLI. Here is a test with three legs: sinlge thread, multithreaded, disk. The overall scores at the end seem pretty intuitive to me from using all these systems:

http://openbenchmarking.org/result/1306121-MEND-130612516&obr_sgm=y&obr_sor=y&obr_nor=y&obr_hgv=Atom+N270

It might be better to have twp suites: QUICKBENCH-CLI and a QUICKBENCH-GUI.

[mendieta]

Here is one comparison of three systems: an older Vostro V13 with a single core celeron a 1.6 Ghz, a Chromebook with a newer dual core celeron @1Ghz, and an older Phenom 3 core @ 2.5 Ghz.

http://openbenchmarking.org/result/1306118-MEND-130611474&obr_sgm=y&obr_nor=y&obr_hgv=Vostro+V13

The numbers make a lot of sense as of the feel for speed of the machines. The V13 makes up in disk speed (SSD) against the Chromebook (with a regular HD) with more CPU and GPU power. The Phenom is outdated but overall faster, paired with a modest (but faster than a Celeron IGP) Radeon HD 4650, and an SSD.

The scores for the three are 1, 1.13 and 1.71, respectively.

I like the idea, but we need better ideas for some of the tests, as discussed above.

[mendieta]

So, here goes a dry run:

http://openbenchmarking.org/result/1306113-MEND-QUICKBE80

Here is the idea:

• 2 (CPU + RAM) tests (one with emphasis in single threaded + floating point, one in multi-threaded + floating point)
• 2 GPU tests (one 2D, one 3D)
• 2 disk tests (one READ, one WRITE)

Some comments:

- The CPU seems to be decently covered, and the tests download and run fast.
- I am not happy with the Graphics tests: the 2D failed to install in the second machine I tried, and the 3D is very simple AND takes forever. Any suggestions? We need fast and reliable install, and fast runtime.
- These disk tests seem to focus on throughput, but we really care about random IO for a general speed measure (this is what the OS does more often, by far). My problem so far was finding a test that does that _and_ runs in a couple minutes. Any ideas?

Please test against these if you can and report, as well. Cheers!

[mendieta]

Whatever happened to this? I really like this idea, and hopefully no one is offended, but it sounds like making a version of PC Mark or Passmark for linux. A nice, quick, overall system benchmark for memory, disk, cpu, graphics, and you get an overall score. Though geekbench kind of does all this doesn't it? And it's included in pts.

Mmm. No, geekbench actually does CPU + RAM, no disk or GPU.

I think this is more feasible now, with openbenchmarking providing a geometric means in the analysis section.

(Although there is a bug with the geometric mean, see http://phoronix.com/forums/showthrea...664#post335664 )

I'll try to revive this thread soon

[skeetre]

Whatever happened to this? I really like this idea, and hopefully no one is offended, but it sounds like making a version of PC Mark or Passmark for linux. A nice, quick, overall system benchmark for memory, disk, cpu, graphics, and you get an overall score. Though geekbench kind of does all this doesn't it? And it's included in pts.