“At Phoronix we are constantly exploring the different display drivers under Linux, and while we have reviewed Sun’s Check Tool and test motherboards with Solaris in addition to covering a few other areas, we have yet to perform a graphics driver comparison between Linux and Solaris. That is until today. With interest in Solaris on the rise thanks to Project Indiana, we have decided to finally offer our first quantitative graphics comparison between Linux and Solaris with the NVIDIA proprietary drivers.”
I think the Solaris Nvidia drivers are in the same league as Linux and support x86 and x64. I’m sure they work very good. Both are in better shape than the FreeBSD driver.
But I own an r300 ATI card and those dont work good in Solaris.
Edited 2007-06-24 15:09
Not sure performance wise, but I have a laptop with a Go 7600 256mb with FreeBSD-STABLE and Xorg 7 and the nVidia drivers work great!
It’s nice that between all the bickering about Vista, OS X and Linux there is still room for articles about technology itself. I don’t really care about Aero, Quartz or Compiz but I do care about stability, reliability and performance.
At a certain point we have seen all the normal usability and application availability stuff come by and it’s time for high-end stuff like what SGI, Sun and IBM are doing for computing now and into the future.
I have been running Slackware, Solaris (OpenSolaris, Nexenta) and BSD (OpenBSD) over the last few years and I like them all for different reasons. Let the masses have their slow and unstable Windows and OS X operating systems and let us have the high-end mission critical stuff.
That’s why I’d like to see more thorough comparisons between Solaris, Linux and the BSDs. Not only on the graphics level but across all subsystems so we can see where they differ and where each is better or worse than the others and what can be improved in each of them. That will help free operating systems become even better than they already are at the moment.
Not to mention that Nvidia is the only viable option at the moment to get a stable, high-performance workstation setup with free operating systems. Let’s hope that Intel can catch up and AMD finally deliver on their promise to create stable, usable drivers for their newer graphics adapters.
The best thing would of course be that with time the Nouveau and R500 reverse engineering efforts produce functional free drivers. An OpenGraphics adapter would be even nicer :-).
I don’t know what you’re doing to cause it, but I haven’t had a Windows crash in years. The last crash I had was due to faulty hardware.
In my experience whilst it was bad in the past, Windows is actually quite stable. It’s the end users who are buggy in their perpetual ill thought out actions.
I had a desktop with XP on it for 4 years, and it only crashed 3 times, always due to the graphics card driver. Although the desktop did crash occasionally and I would have to manually restart explorer.exe. I also had a laptop with XP on it, and it crashed all the time. It didn’t last long before it was replaced with linux. Clearly your experience with Windows these days relies mostly on the quality of the drivers for your hardware.
I think poor drivers are the cause of a vast majority of instability slack Windows gets.
Completely agree. XP has been very stable for me. I’ve had one blue screen on one of three systems I run XP on since moving from 98. It was a vid card (bad vid card) that gorked the system.
An OS that lets a bad driver bring down the whole machine is poorly designed.
Additionally, Windows is bound to have bad drivers forever as most motherboard manufacturers have little economic incentive to fix poor drivers on hardware that’s over six months old. The push to be first to market and the quick obsolesence of hardware means that most bad drivers stay bad. Since the close-source nature of Windows makes it impossible to fix these drivers, I would say that this is also a design decision that users end up paying dearly for and that the Windows OS and the company behind must be held accountable for the problems created by their design and licensing choices.
To say, oh, but windows works wonderfully if you find the right choice of drivers is to create a moving target, so that no valid critique can ever be leveraged against it, because someone can always claim that if I only had the right driver combination, things would be dandy.
Wake up and smell the coffee.
Edited 2007-06-25 10:19
An OS that lets a bad driver bring down the whole machine is poorly designed.
Then by your standard, Linux is poorly designed.
Installation wise:
1. solaris will take just 10 seconds to install nvidia latest driver “./driver”; and it would be all you have to do, the system will become very fast afterwards, check “gearsglx”
2. Linux in general will take 5 minutes to install the driver if the distro doesn’t nag about missing dependancies
3. solaris kernel updates will not mess up the system
4. linux kernel updates will mess up the system and the OS will ask you to recompile the graphics driver again, starting with previous driver uninstallation.
5. On the benchmark results I got with ViewPerf version 9, solaris vs RHEL (both are supported) I got solaris advantage over linux; and don’t forget the light speed of solaris startup.
1. See below
2. It’s doesn’t take 5 minutes for the driver to install on Linux, what are you on about.
3. See below
4. Nether will they in Ubuntu for example, binary distros have precompiled versions of the driver.
Please stop lying as it only makes you look bad and also , makes you look like your making the OS look better than it is by lying without standing on it’s own merits
I’m glad that Solaris has almost caught up to the best performance and usability of modern Linux distributions but it still has some ways to go.
The problems you highlight are possibly valid for most RPM/DEB based distributions. Slackware is an entirely different beast altogether.
Since I am totally in charge of what’s installed and what not without dependency problems coming up at all it will not take more than a minute to install the driver. I will have to recompile the driver module for the running kernel though.
Solaris has the advantage of a fixed ABI so updated kernels will work with the same driver. That’s why I always build and install kernels myself on Slackware and I recompile drivers and build packages for them, which makes it an integrated experience for the end user.
Solaris is faster than many “enterprise” or “user-friendly” Linux distributions and that’s an advantage it’s never had, at least on x86 hardware. I still have to see that it’s faster than Slackware (or Gentoo), though.
Points 2 and 4 are irrelevant (lies?) as it depends on the packaging system. Ubuntu is an example where it “just works(tm)” and I wouldn’t be surprised if installation of nvidia driver there is faster than on slowaris.
Point 3 just shouts “solaris has stable ABI” which is true, I’ll grant you that. But kernel ABI stability is second to useless with a proper package management. It’s the libs ABI stability which counts in the end, and Solaris distributions use mainly same library base as Linux does, which sucks (ABI wise). So there…
5. “light speed startup” ? Solaris is called “slowaris” for a good reason. Don’t bliss yourself too much it still sucks on x86 hardware. It won’t even boot on this machine and that’s an older piece with such standard components that EVERYTHING works on the 1man made SkyOS! So much for that…
So all in all, next time your rant about something, be at least objective a bit mkay? I could go around shouting how Solaris doesn’t work on my hardware while 1man OSes like SkyOS do (which is a true shame for Sun IMHO). But I don’t because it’s one machine and it only speaks for the drivers…
As for the gfx drivers, from what I know they are very comparable. The base code is abstracted from OS so if the “glue” part is ok they will work almost same. I know that FreeBSD has certain specific memory access problems which resulted in hacky ways the nvidia driver does it stuff and could impair performance. AFAIK solaris and linux both have most or all “required changes” in the kernel for nvidia to work happily so the differences should be minimal. It might even be that the performance difference is caused by other things, like memory bandwidth etc. which are tied to the kernel (scheduler etc.) and drivers.
Edited 2007-06-25 09:59
First, the term “slowaris” was coined by SunOS users when transitioning to Solaris 1, that was a long time ago. And let’s see, it sucks on x86 hardware because it won’t work on the machine you just happen to have.
I have used Solaris x86 on a variety of hardware since 1999, and although its hardware support isn’t as great as Linux is, the amount of hardware that Solaris does supportr is increasing. The performance of Solaris, particularly disk I/O with IDE drives has long been an issue prior to the release of Solaris 10 required some tweaking. Solaris 10 now recognizes IDE/ATA hardware and sets the appropriate DMA mode on boot. And before you bring up that Linux does this automatically, it wasn’t all that long ago that you had to use hdparm to get optimal performance out of Linux with IDE drives.
On decent hardware Solaris performs just as well as Linux does.
Solaris trades kernel flexibility for stable driver api, see her http://www.mjmwired.net/kernel/Documentation/stable_api_nonsense.tx…. The good side is that when something in the kernel changes, all drivers are fixed by linux developers, where on solaris, you would have to rely on vendors for that.
You can see this effect on windows, where the os itself is quite stable, but one small bad writted driver can make it unstable. Linux policy will make vendors release open specifications in the long run and every other os would also benefit from that. I mean i see this as a feature and not as limitation. So i think the need to recompile the kernel is a small inconvenience opposed to bad drivers in the long run. One can try to remedy the stability problem with microkernel architecture, but it’l be slow and won’t fix the bad drivers.
While I am pleased to see a comparison across the two OSs, the benchmark method used was somewhat inappropriate. The ViewPerf benchmark is designed to test workstation class graphics performance, not general consumer hardware. Sure you can run the benchmark on a GeForce card instead of a Quadro, and it will work, but it’s not designed to test that sort of hardware. Many of the key differences in the Quadro and GeForce cards are they very things that the ViewPerf benchmark relies upon.
You wouldn’t use an HPC benchmark designed to test cluster performance to compare a single workstation install of Linux or Solaris. Sure you might get some relative numbers, but the test just isn’t appropriate.
That said, ViewPerf is, however, really the only cross-platform benchmark available for 3D graphics now. If the driver performance is being tested I think it would be much more sensible to re-run this test w/ a Quadro card. You would still be able to get the relative driver performance comparison but the hardware would be much more appropriate for the test.
http://www.3dprographics.com/workstationgraphics.pdf
“NVIDIA Quadro v. GeForce series”
Edit: Added the link to hardware comparison.
Edited 2007-06-25 01:30 UTC
I remember it was just a few years ago, Solaris had a terrible reputation for being slow and unbearable on x86, things have changed big time. I hope that with further development it’ll result in Solaris overtaking Linux in some areas, along with improved hardware support.
I’ve been using Solaris lately, and though I’m typically not interested in doing 3d coding myself, I would like to do some GPGPU coding. Unfortunately, none of the existing libraries (RapidMind, Sh, BrookGPU, or Nvidia CUDA) are available for Solaris at this point.
This isn’t a big hindrance to me, but I like Solaris and don’t like to see it held back.
Microkernel is a myth. There are other ways an OS can crash the system besides drivers; even userspace components can crash the system.
Drivers are the least of Linux’s problem and hardly ever contribute to a crash unless you use the terrible ATI drivers.
As for stable ABI… not a good idea. When Solaris starts seeing the number of drivers in Windows and Linux, it too will be spending years trying not to break compatability just like Microsoft did in Vista.
The advantage of a stable ABI are all mitigated when binary packages of proprietary drivers are available.
Edited 2007-06-25 15:00
Care to explain why?
I agree that microkernel is a myth, since if for example I write a wrong register on a processor/some hardware piece the system will crash regardless of monolithic/microkernel, the point of my post was to contradict the solaris fanboy.
“4. linux kernel updates will mess up the system and the OS will ask you to recompile the graphics driver again, starting with previous driver uninstallation.”
The reason for that are nvida/ati’s drivers and not linux kernel. The other points in his post are also a pure fanboism, since it largely depends on the distro.
I was just wondering why people modded him up since I couldn’t find anything of use in his post.
Is CAD support. I’ve been managing CAD workstations for years, Sun, SGI, IBM, HP-UX. Everything has to run on the SPARC (or MIPS, or PA-RISC, etc) architectures, because that’s what our CAD vendors compile to (unless we want to go to [shudder] Windows.
I’d love to run Catia V5, or NX, or I-deas on Solaris/x86/64. I’d even like to start putting some linux clients out there. We have the hardware. We have the expertise. We have the rationale …. now all we need is the damn software!