Friday, November 06, 2009

HP 5735, Disappoints

I get email and know that some of you are thinking about thin clients in your organization. I have always tried to be honest and describe what works and what doesn't.

After a good amount of testing, we have been unable to get the HP 5735 to work well enough to put into production. They may work in other areas perfectly well, but our goal was to add ATI 9250 video cards to them and use them for the 3D desktop. Even though they have higher specs than the HP 5725, under no conditions have they ever run faster. 3D performance is very sluggish and certainly not fast enough for day to day use.

I know this isn't a perfect test, but the results seemed to correlate with Compiz "feel". It also could easily be replicated by anyone else. For my test, I booted an X server with just a xterm window and then connected to our remote Compiz servers and ran glxgears. The 5725 was the clear winner with 653 FPS. I also tested the onboard X1200 and then 2250 video cards.



We only purchased a few 5735s for testing, and will be moving them into our cyber cafes which don't use 3D and for that purpose they will work well.

We will continue to use the HP 5725 (which works great) until we have another device that performs as well.

12 comments:

Infenwe said...

How many times does this have to be repeated? glxgears is not, was never intented to be, and never will be a benchmark application.

GIYF.

Dave Richards said...

infenwe: Understand about glxgears. I tested over multiple days and with multiple samples. I had to use something that HP could replicate on their side as well. They don't have Compiz and it's hard to explain "too slow" with no numbers. The FPSs that I got seemed to correlate with the speed of Compiz. The 5735 felt like it was running at half speed compared to the 5725, and these numbers reflected that.

Anonymous said...

sudo apt-get install phoronix-test-suite or go here
http://global.phoronix-test-suite.com/ to get reliable, repeatable benchmarks.

anholt said...

If you want to benchmark compiz, benchmark it using the benchmark plugin. Don't use something else, that runs faster than refresh rate (and thus is sufficiently fast in all your numbers), and *doesn't use any of the paths that are problematic for compiz performance*

Anonymous said...
This comment has been removed by a blog administrator.
AdamW said...

In addition to all the 'glxgears is not a benchmark', which is entirely true, I forget what distro you're using, but if it enables kernel modesetting for the radeon driver by default you might try testing with it disabled. Performance is currently known to be worse in the KMS case than the UMS case, on most Radeon hardware, with current driver releases.

AdamW said...

damn, of course it's Debian, d'oh. Don't think they have modesetting on yet. Never mind.

Jorge E. Gómez said...

I believe HP don't understand that the thinclient business is very different from their laptop or desktop units.

They keep revving the models more often than thinclient customers would like them to.

We value stability above new features, and would probably keep buying the same model for years instead of upgrading an installed base of changing the OS to keep up with newer video and network devices.

ombzzz said...

Sorry Dave let me ask this question?

Why do you need Compiz? I understand that Compiz is only eye candy for the (desktop) home user; i can't imagine its use in a work environment.

Anonymous said...

It'd be great if you commentators could either praise Dave for his Open Source application in business or provide constructive criticism / ideas for making his job easier, rather than attack him for doing a simple non-scientific test just because it was quick and easy. Ombzzz, follow his blog: Compiz isn't for everyone, the users get to choose if they want it or not. Some people / power users can be more efficient with such a 3D desktop if they know how to use it properly.

DAVE: KEEP UP THE AWESOME WORK

I completely agree with Jorge E. Gómez, the end of life on thin clients comes too soon with HP.

Dave Richards said...

I knew glxgears was not a benchmark going into the project. In this specific case it was a fit for showing HP identical hardware running different operating systems. I was able to test Lenny and Etch on the 5735 and compare it against the 5725. It also was the only thing that I could easily get to HP for their own testing.

The Compiz benchmark did not help me at all in this case, HP has no way to run it. That is especially true when you consider that I'm running a stable version of Compiz from last year on the production machine. No way to replicate it outside of this building; which we wanted them to do.

Jorge: We actually would have been content buying the 5725 for many years to come. It's smoking fast, stable and reliable. So your statement is completely true. What is expensive in enterprise use is having *different* hardware. I'd rather have 100% of the same model than some people being on the latest and greatest.

ombzzz: The 3D desktop has helped us greatly believe it or not. The issue of running and navigating through multiple windows and using workspaces is a difficult concept for many 'regular users'. About 70% of our users have enabled 3D effects, and it's entirely optional; and off by default. The users love the cube, and use it all the time.

anonymous: Thanks for the words. I just kind of post the things that are working here, and if they help other people that's great.

ombzzz said...

Thanks for your reply Dave.

( i don't want to criticize , just was wondering about that ).

Thanks for sharing your expertise in deploying solutions in large-scale.

Greetings from argentina.