NVIDIA Performance On Linux, Solaris, & Vista

AtomBOB suggests a Phoronix review comparing the performance of a Quadro graphics card on Windows Vista Ultimate, Solaris Express Developer, and Ubuntu Linux. The graphics card used was a NVIDIA Quadro FX 1700 mid-range workstation part. The cross-platform benchmark used was SPECViewPerf 9.0 from SPEC. Quoting Phoronix: "Using the Quadro FX1700 512MB and the latest display drivers, Windows Vista wasn't the decisive winner, but the loser... Ubuntu 8.04 Alpha 5 with the 169.12 driver had overall produced the fastest results within SPECViewPerf. In only three benchmarks had Solaris Express Developer 1/08 outpaced Ubuntu Linux, but with two of these tests the results were almost identical.""

Re:What is the difference?

[Anonymous Coward]

The difference between the Quadros and the consumer cards used to come down to hardware OpenGL overlay support, if I remember right.

Re:What is the difference?

[sxeraverx]

They have different priorities. Gaming cards try to keep the framerate up by degrading image (not showing every single texture, e.g.), if need be, while cards for stuff like CAD and the like lower the framerate to show every detail requested of them.

Re:OpenGL?

[glob]

http://www.opengl.org/pipeline/article/vol003_9/ [opengl.org]

"Some have suggested that OpenGL performance on Windows Vista is poor compared to Windows XP. This is not the case."

Re:Surprised..

[Alexx K]

No. I have read it here [schneier.com] and here [auckland.ac.nz], but I'm not certain whether Vista actually does this or if it's just a massive fud campaign. From what I've read, it seems to be true. But as I said, I'm not 100% sure.

Re:Surprised..

[recoiledsnake]

It's mostly FUD. See here (read all three parts) [zdnet.com]

Re:broken

[darkjedi521]

If you're running Maya, would should be running the drivers/distro that Autodesk blesses. Last I checked, that was 2-3 year old drivers on RHEL 4/SLES 9/Fedora Core 5. I run the blessed packages for a small animation studio and only have problems when people out of memory their system (8GB RAM should be enough for anybody). http://usa.autodesk.com/adsk/servlet/item?siteID=123112&id=9683256 [autodesk.com] has the list of blessed stuff.

Re:ws cards

[andersbergh]

There's definitely a different between say, 30 and 100 fps: http://100fps.com/how_many_frames_can_humans_see.htm [100fps.com]

Re:the difference does not matter.

[recoiledsnake]

What's significant here is that Windows has lost it's graphics crown. DRM and bloat or industry defection for the same reasons, we all know the root cause. Free software is simply cleaner and works better. If the ability to run DirectX 9 under Wine was not enough to move gamers to Linux, this is. Things can only go downhill for Microsoft now. Free drivers will be even cleaner and the performance gap will widen.

From TFA:

Then in September, we had looked at NVIDIA's multi-GPU performance under Linux and Windows when running two GeForce 8600GT 256MB graphics cards in SLI (Scalable Link Interface). Windows XP and the ForceWare driver had outpaced Linux in every gaming test we conducted.

The drivers have a lot more influence than you give it credit for.

Re:Headline is misleading!

[kc8apf]

That's interesting. Considering that I am a developer for the CHUD Tool (no quotes) and I do performance analysis and benchmarking for a living, I don't think they did anything wrong. Things that aren't running on a system rarely affect run-time performance. Going from a distribution like Ubuntu to Debian just removes a bunch of things from disk, but those things have zero impact on the metric being measured. For Vista, it might make a difference if the version used was shown to have less idle activity, but in practice, you want to compare what a typical user would be running. So, since the OSs chosen reflect typical users, the data is perfectly valid for a comparison between them. If you want absolute performance numbers, then you need to start tuning the OSs before you run the tests. Things like disabling daemons or services and unplugging network cables can cause measurable differences in some benchmarks.

As for the CHUD Tools, they are completely inert unless you happen to be running one of the tools and even then, it isn't likely to cause any significant difference. The kernel extensions used by the CHUD Tools are designed to do absolutely nothing until they are asked to. If you are running a Time Profile in Shark, it will have some impact, but it will be limited to 1-2%.

Re:Surprised..

[Foofoobar]

That's actually an old article filled with half truths that only show one particular point of view of the DRM on Vista. It was released shortly after beta and now that Vista has been on the market for months, it has actually shown that 75% of the DRM problems were actually true (while the rest were exaggerations).

Re:What is the difference?

[alex4u2nv]

I had the very same question, and this article from Nvidia turned out to be very enlightening.
Quadro vs FX -- http://www.nvidia.com/object/quadro_geforce.html [nvidia.com]

According to the article, there are some major differences between the two architectures. Where features are programmed either at the hardware layer (quadro), or at the driver layer.

Re:True, but...

[tapehands]

If you really want to blow your mind, try typing in:

man woman

back on topic...nvidia pdf [nvidia.com] from september of 2003 explaining the differences. Yeah, old, but it's the only document on nvidia's website that I could find that would explain the differences.

Re:Surprised..

[yanyan]

About the author, Ed Bott:

Ed Bott is a freelance technical journalist and book author. All work that Ed does is on a contractual basis.

Since 1994, Ed has written more than 25 books about Microsoft Windows and Office. Along with various co-authors, Ed is completely responsible for the content of the books he writes. As a key part of his contractual relationship with publishers, he gives them permission to print and distribute the content he writes and to pay him a royalty based on the actual sales of those books. Ed's books are currently distributed by Que Publishing (a division of Pearson Education) and by Microsoft Press.

Ed's personal website at edbott.com includes advertising. All display advertising is managed by Federated Media. Ed reserves the right to reject ads for any reason. In addition, Ed allows advertisers to purchase links in a sidebar through Text-Link-Ads. Advertisers receive no special treatment at his personal website, at ZDNet, or in books. Ed makes a small amount of money selling books (his own, primarily) through an affiliate account at Amazon.com.

On occasion, Ed accepts consulting assignments. In recent years, he has worked as an expert witness in cases where his experience and knowledge of Microsoft and Microsoft Windows have been useful. In each such case, his compensation is on an hourly basis, and he is hired as a witness, not an advocate.

Who do i trust, an independent researcher or an M$ lackey?

Re:OpenGL?

[Anonymous Coward]

They don't need to degrade OpenGL. OpenGL being slow (it isn't) isn't the reason most developers who pick DirectX pick DirectX. All the developers I know that use DX use it because they like the tool-kits more than what's out there for OpenGL.

I'm going to have to call BS

[Tsaot]

Free software is simply cleaner and works better.

I have to call BS on that. If I have to choose between the latest versions of Open Office and Microsoft Office, I will take M$'s closed solution hands down. The interface on 2007 is vastly improved over other office offerings out there. Making something free and open source does not make it good. I can think of many free applications that don't make the grade in cleanness and usability when compared to commercial offerings.

Re:Video on Linux

[dbIII]

There has been hardware rescaling to TV modes on their cards for a few years so you'll find the cheapest models with TV-out do a good job. Other features have improved a lot in the linux and other drivers - look at the README on the nvidia download site for the long list and how to turn some on or off.

Re:What is the difference?

[LoRdTAW]

Gaming cards try to keep the framerate up by degrading image (not showing every single texture, e.g.), if need be

Thats called culling and it is implemented in software, not hardware.

If I remember correctly there was a simple hack posted on Toms Hardware a while back for converting a Radeon to a FireGL. You simply solder an SMT resistor to a certain trace on the chip package and it pulls a line low. That line actually signals the BIOS to report the card as a Radeon or a FireGL. So in essence the Radeon and FireGL are the EXACT SAME CARD! The only difference is the FireGL drivers look for a Radeon reporting itself as a FireGL. This keeps production simple and even the video card BIOS versions the same.

The FireGL and Quattro cards come with optimized drivers for specific 3D programs like AutoCAD, Maya, 3DSMax, Light Wave etc. There is a drop down box that lets you select the program your using and it loads the finely tuned driver for that program.