Performance regressions with 256.25 driver?

[Lithorus]

Originally Posted by wiresquire

What has that got to do with THIS benchmark on an nvidia driver and nvidia card?

Do you see anything specific to this article/benchmark? If so, point it out. If not.....

ws

For starters he should atleast test on more than just 1 gfx card.

[Sean_W]

Originally Posted by wiresquire

What has that got to do with THIS benchmark on an nvidia driver and nvidia card?

Do you see anything specific to this article/benchmark? If so, point it out. If not.....

ws

As said above and it was done on a laptop mobile GPU, Ideally it would be better on a high and low end PC.

[Licaon]

Originally Posted by mooninite

Example: his filesystem benchmark "showdown" between ext3/4, xfs, etc. was done in such a way that some file systems were not tuned properly or the benchmark itself was not performed correctly.

that's the point: to test file systems at the DEFAULT settings ( provided by the distro ) because a generic user will use the defaults while the specialised ( server, workstation, etc ) user will know and want to optimise the settings as s/he sees fit anyway

[gonzo_s]

Lol, Phoronix...
They always post monstruous test results of 'comparing the uncomparable'.

2davelaser: Stop pulling that sh**, especially here. Better do your own measurements and post complete results (hardware, distribution etc.) for us to see. Never read that crap.

[wiresquire]

Originally Posted by Sean_W

As said above and it was done on a laptop mobile GPU, Ideally it would be better on a high and low end PC.

Please explain why that is so? The phoronix article basically says "We tested 1 laptop, and there is a performance regression on it. To quote from the article:
"At least in select software/hardware configurations the NVIDIA 256.25 Linux display driver does carry some hefty performance regressions".

It's entirely valid for phoronix to post the result from the machine they tested. If you disagree with their methodology, then say why. If you don't see regressions or see improvements, then that's great, and post your specs. Don't bitch at phoronix for posting theirs.

My reply above was specifically to moroninite, who started raving on about filesystem benchmarks when the benchmark in question is about nvidia cards and the 256.x driver.

BTW, Sean_W, why do you say it would be better testing on a high and low end PC vs a laptop mobile GPU?

ws

[weltall]

Originally Posted by Sean_W

As said above and it was done on a laptop mobile GPU, Ideally it would be better on a high and low end PC.

it's not even that I've a mobile gpu way worse than the supposedly tested one (8600MGT 512MB) and I noticed improvements there.

[Sean_W]

Originally Posted by wiresquire

Please explain why that is so? The phoronix article basically says "We tested 1 laptop, and there is a performance regression on it. To quote from the article:
"At least in select software/hardware configurations the NVIDIA 256.25 Linux display driver does carry some hefty performance regressions".

It's entirely valid for phoronix to post the result from the machine they tested. If you disagree with their methodology, then say why. If you don't see regressions or see improvements, then that's great, and post your specs. Don't bitch at phoronix for posting theirs.

My reply above was specifically to moroninite, who started raving on about filesystem benchmarks when the benchmark in question is about nvidia cards and the 256.x driver.

BTW, Sean_W, why do you say it would be better testing on a high and low end PC vs a laptop mobile GPU?

ws

I never said it's invalid but I would have 'preferred' to see a desktop setup test. Linux users generally run lower system specs due to Linux being lighter. I like to see a Fermi test as well with the new betas.

[firewalker]

From what I've seen, at many Phoronix benchmarks there are big differences between the results of the various benchmarks. IMHO either there are not enough "benchmark runs" to get accurate results or he uses the average between the "benchmark runs".

There are dozens process running or sleeping in the background especially when running a "Default Desktop". How do you know that something else didn't decide to run or to sleep when benchmark was running.

But it is hard to tell without some profiling/perfomance information which would make the benchmarks actually helpful to developers or to distro maintainers.

As said above and it was done on a laptop mobile GPU, Ideally it would be better on a high and low end PC.

Remember hardware is not free and high end hardware costs a lot.

[Sean_W]

I don't trust Phoronix benchmarking for one, based on a number of past benchmarking usage and these guys are such noobs, they use Ubuntu pre-compiled PPA kernels.

[wiresquire]

So what are you guys saying?

a) The results for this test of the 256 driver:
[ ] are rubbish. You don't believe the Phoronix results.
[ ] are OK. I believe the Phoronix results.

b) Performance regressions for the 256 driver:
[ ] i) don't exist. On any systems at all.
[ ] ii) exist, but only on mobile systems.
[ ] iii) exist only on the specific laptop that Phoronix tested.

c) That the reason for these results for the 256 driver are:
[ ] i) Phoronix setup is non-standard and cannot be reproduced
[ ] ii) Phoronix benchmarks run on a standard desktop and are subject to spurious background/sleeping/random processes.

Please feel free to check all that apply.


Also I have a question or 2 for you. Please answer each question:
i) Has anyone here ever run the phoronix-test-suite benchmarks for the specific benchies that they ran in this article?
ii) Have you run those benchmarks for the 256 series drivers?
iii) Have you run those benchmarks for any previous driver versions?
iii) If so, report your config, relevant settings and results.

ws

[Sean_W]

Phoronix make too much drama out of this. Regression in a testing release? NO WAY!

Testing pre-release software and then claim there is some kind of regression or "disaster" (as in the 2.6.35 kernel tests) is just plain silly without identifying what it is.

I would agree that people should post their benchmarks on games with both drivers, specifically Openarea.

[Licaon]

Originally Posted by wiresquire

i) Has anyone here ever run the phoronix-test-suite benchmarks for the specific benchies that they ran in this article?
ii) Have you run those benchmarks for the 256 series drivers?
iii) Have you run those benchmarks for any previous driver versions?

i) i ran it once in the past, i see no need to get the whole suite just to test one bench
ii) i did ( as previously mentioned ) as others did here: here and here with mixed results as you can see
iii) yep, see above, the test was specifically to see if those finding are ok since i saw better performance while running the composition comparison tests here: Sanctuary demo benchmark edition v2.3 / Tropics Demo v1.3

maybe Phoronix had PowerMizer issues, maybe that mobile GTS250M has issues... we won't know until he gets some more tests with another card preferably a desktop one

/LE: Phoronix: PowerMizer Is Not Likely The Problem In 256.xx Issue