Slashdot Mirror
nVidia Posts First Linux Graphics Drivers for Opteron
Brian Stretch writes "nVidia posted the first publically available Linux graphics drivers for the Athlon 64 (aka Hammer series) on their website today. There are updates for the lesser x86 and IA64 architectures as well. Now, if only the Athlon 64 and Opteron boards and CPUs themselves were publically available, or is AMD's developer program sending out more of these things than I know? (If so, gimme!) I guess I'll have to tough it out with my mere dual Athlon 2400+ workstation for now (heh heh heh)."
In related news, an anonymous reader writes "The new AMD Opteron servers designed by Newisys are using embedded Linux for system management. This allows remote management via web browser or ssh to examine processor state, switch power on/off, regulate processor power states and fan speeds, update BIOS firmware, etc. See the docs for more info!"
Are there any of these available(especially OSS) for the more common motherboards? I've never heard any.
:)
Granted, I'd hate to be a beta tester.
What's this Submit thingy do?
Are these really the first Linux graphics drivers for Hammer? Surely XFree86 and other graphics systems like the kernel framebuffer are already being ported?
It sounds like this is just the first Hammer release of Nvidia's proprietary, binary-only drivers for cards they won't release specs to. Useful, but hardly any more significant than some other random piece of proprietary software being ported.
-- Ed Avis ed@membled.com
Come on y'all--there's been more dicussion about this guy's rig than about the drivers. I'm sorry if you're bitter that your box is less than state of the art, but dems da breaks, kids. In the meantime, kudos to NVidia for getting these out before anyone needed them.
Don't Panic!
What's next? ATI doing Natalie Portman/hot grits drivers?
oh do shut up
Invoicing, Time Tracking, Reporting
RTFA. Linux drivers for AMD64 (Hammer, Opteron, what-have-you).
Linux x86 drivers have been available forever, and last month FreeBSD drivers came out (I know, I'm running them right now *g*)
TODO: Something witty here...
I guess I'll have to tough it out with my mere dual Athlon 2400+ workstation for now (heh heh heh).
I don't think anyone here cares how big your e-penis is, and no... I won't stroke it.
Not to be a zealot, this is an honest question: Why dont they open-source their linux drivers? They are giving them away for free anyway, and i would think it could only increase the attraction to their products for some ppl. I can't think of any reason why not to.
These are drivers for the 64 bit processor.. The ones for 32 bit (what we all have) have been out for years.
"tough it out with my mere dual AMD 2400 (heh heh heh)"...
I havent gotten laid since October of '67. (sob sob sob)
This is my sig. Its pathetic.
There are Linux drivers for 3 different platforms intel based platforms now. There are OS X drivers for PowerPC. Why can't Nvidia merge some of that code to give us Linux/PPC Nvidia drivers. A lack of a good graphics system for Linux/PPC is the major factor holding it back. Hopefully these things will change once IBM's GPuL hits the shelves.
I guess the rest of us will have to make due with the mere thousands of dollars we saved (heh heh heh...)
Nvidia - Random Video Card Driver?!
That's like saying that Microsoft is porting some random operating system to the Hammer architecture.
The first "officially" released drivers by nVidia may pragmaticlly seem insignificant, but in the big picture this is a big success for the linux community as a whole
I have great faith in fools; My friends call it self-confidence. Edgar Allan Poe 1809-1845
You had sex with a mere dual Athlon 2400+ workstation?
However, if you want to possibly commit libel while accusing someone else of it, go right ahead, I won't stop you.
And I think there's some value in being first to do something, regardless of the licensing status or any other factors.
"Yeah, whatever, first space satellite... but they were COMMIE BASTARDS!" <-- sour grapes
pb Reply or e-mail; don't vaguely moderate.
ouch, watch out for those sawder points
It was me, I did it, I moved your cheese
With the new drivers I didn't see any performance gains with Quake3 timedemos or Ut2k3, with a ti4200.
-- taking over the world, we are.
I guess I'll have to tough it out with my mere dual Athlon 2400+ workstation for now (heh heh heh).
[sigh] You're bragging about a single dual Athlon machine when I've got a beowulf cluster of those??..
"Truth is not decided by majority vote" consensus gentium -- Norman Geisler
To clarify AMD's processor naming scheme for those who haven't been keeping up with the Joneses, the Athlon 64 is AMD's 64-bit desktop and mobile CPU, while the Opteron is AMD's 64-bit server and workstation CPU. Both utilize the x86-64 architecture, which is essentially an extension of the existing x86 instruction set for 64 bits.
A few key differences between the two are that the Opteron will be multiprocessor-enabled and have three HyperTransport pipes (each providing a theoretical 6.4GB/s of throughput) versus one in the Athlon 64. The Opteron will also have more on-die L2 cache (1MB and 2MB are being talked about right now), and will draw quite a bit more power (90W+ vs. ~65W for the Athlon 64).
Since almost no comments here are about the actual post, I've got a cultural question...
Where/How did the "In Soviet Russia..." posts come from? I remember where things like 1.2.3.Profit and the now-passe Mastercard and All Your Base jokes started, but what brought on this spur of neo-Marxist-Lenninist thought?
Cue The Sun...
He didn't say Athlon XP, I just finished building a dual Athlon MP 2400+ myself. The mother board is a Tyan Thunder K7X Pro, power supply is an Antec True550 EPS12V, and memory is 4 sticks of Corsair's 1GB 266MHz ECC registered memory. I also have Adaptec's 0 channel RAID controller, and 8 18GB Seagate 15K.3 RPM SCA drives. Using built in ATI RageXL video and gigabit LAN. All in a Lian Li PC-626SCA case.
The machine is rock solid from what I can tell, and blazingly fast, and I'm glad I didn't have to pay for it.
They are not the first. I saw at least one Opteron-based 1U system running XFree in the AMD booth at SC2002, just a few weeks ago. No idea what the video/driver subsystems were like (maybe fbdev?).
Wouldn't be surprised if this was the first x86-64 driver to support hardware accelleration though.
-Isaac
I am not a lawyer, and this is not legal advice. For Entertainment Purposes Only.
quake3, rtcw, civ:ctp, tribes2, alpha centauri, rt2, simcity3000. with wine(x), the sims, diablo2, sof2, halflife/cs,
:o
baldur's gate 2, alice, gta3, wc3.
And I'm just listing games that I regularily play, although I'm mainly a q3 junkie
And ya just know he does no 'work' on his workstation, or even have a clue what it's capable of - because he somehow thinks he needs an Opteron on release day.
*Clue stick* Opteron is not going to make either of the 2 games for linux any faster.
I don't need no instructions to know how to rock!!!!
I missed the 3rdly in my post, it wasn't chrisd that has the dual 2400+, it is Brian Stretch.
I've had absolutely no lockups in X over the past six or so months. The drivers have come a looong way in stablity from the past. I remember about a year and a half ago, I used to get random lockups all the time, even after messing with many things like AGP 4x->2x, sidebanding off, fastwrites off, nvagpagpgart, etc.. But recently, it works fine pretty much out of the box.
NvAGP "0" in XF86Config in the Driver section
It saves them time fixing bugs, proting etc... potentially giving them a higher usebase.
thank God the internet isn't a human right.
binary-only drivers for cards they won't release specs to
Whinge.
They can't. There is IP involved they don't own, both from S3 and SGI.
Not that this hasn't been said everytime some jackass whined about the same thing. And not that it'll help morons like you who don't have a clue how business works understand this little concept.
If you really want them to fix it you could report the bug to them, rather than Slashdot... ;)
Wah!
You missed Unreal Tournament, both the 1999 and 2003 versions. It took more than a bit of tweakage (tar.gzipping a working install and then untarring/ungzipping the working install into the new computer) to get UT(1999) installed because I couldn't make the Loki installer work under Linux-Mandrake 9, but now that it is installed it works beautifully. Monster Kill, baby!
Knowledge is power. Knowledge shared is power multiplied.
I don't work for Nvidia, but if they're like most high end competitive products, there are probably trade secrets involved in their designs. Trade secrets are generally things that might not be patentable, but nonetheless are critical to the technical lead your product might have. The fact that they may not be patentable is why companies try to keep them secret for as long as possible.
In the case of NVidia, it's entirely possible that their driver code would necessarily reveal some of their hardware's trade secrets.
The irony here is that most Slashdotters probably don't have anything big against the need for hardware companies to keep trade secrets in general, but when this necessitates closed-sourcing some of their driver code, everybody screams foul.
I'm all in favor of OSS, and I use OSS for everything I do unless there's no option, but put yourself in their shoes for a moment - if you happened to make the world's fastest consumer video card at some point in time, would you be in a hurry to release details that would likely help your competitors to catch up faster?
You might ask "then how come company X can release open source drivers or specs and NVidia can't?", and this would be a valid question. I don't know the answer, but there are several possibilites. One is that the specs they release to the OSS community don't really have *all* of the details (which would mean their proprietary drivers would always be a little bit faster). Another possibility is that their design is such that the driver code or programming specs don't reveal as many trade secrets.
UK-Smiley (ironicon) £:-)
Hmm, on my screen that looks like a smiley with a bad toupee. Would that be the Marv Alberticon, or the Bill Shatnericon? Or would it be a Ted Koppelicon, being quite obvious about it being there, but insisting it's not a pound symbol on your head and just having a bad haircut for years?
Heh, maybe NVidia should open-source main.cc, whose only function is to call closed-source library functions. :)
There is IP involved they don't own, both from S3 and SGI.
And your proof to back up this statement would be what?
Dinivin
They said the first public release of thier drivers. Past instances of SC have generally shown future work (as in beta stuff), not current tech. I'm willing to bet that there are other graphics drivers from other vendors that are still internal.
Plus if you don't know what video/driver subsystem how do you know it was not nvidia (and hence, still the first)?
------- Sorry about the spelling, I suffer from two problems. Dyslexia makes it difficult to spell well, lazy makes it
Oops.. I was thinking about UT in my head, but I guess seeing ut2003 in the grandparent post made me subconciously not put it in the list.
Is a video card manufacturer's hardware driver a trade secret?
How is this possible? The only way that we can consider software to be a trade secret is if we allow ourselves to believe that the closed source model is good for all. Is the layout of what's under the hood of a Toyota a trade secret? No? Oh. I guess software is more special. We should be happy sticking stuff in one side and getting stuff out the other without having a clue as to what's going on in the middle, even though there is absolutely no reason (in an ideal world) not to share that information.
All that open source advocates are asking for is a chance to better use the equipment that they have purchased. It's still possible to charge however much you want for the driver+card or whatever your business model requires, but, in the long run, the open source dogma says that it's better to let people play with the stuff they've bought.
The only reason why I say that you may be missing the point about open source is because I read a thread where you kept saying, "What's the point? I mean, what's the point? Tell me, what's the point?"
I gave you an example of a direct consequence of "the point" and you say you already know this? Why ask, then?
As for your quip about open-sourcing the plans for building jets, I would think that if it were possible for a heretofore unknown group of non-jet-designing-as-a-full-time-job US citizens (say, graduate students in aerospace engineering, for example) to improve on current plans then we would benefit from those people working on it in their free time, don't you? The DOD could then spend their money on actually building the equipment to make these fancy new designs. This would be the new race.
All open source people generally want is to get as many capable people as possible to attack any and all current, and sometimes unforeseen, problems. This is a good thing. There's not really much bad to say about it. If a couple stupid paradigms have to be changed on the way, so be it.
As my father lik@(munch munch)...
He didn't say Athlon XP, I just finished building a dual Athlon MP 2400+ myself. The mother board is a Tyan Thunder K7X Pro, power supply is an Antec True550 EPS12V, and memory is 4 sticks of Corsair's 1GB 266MHz ECC registered memory. I also have Adaptec's 0 channel RAID controller, and 8 18GB Seagate 15K.3 RPM SCA drives. Using built in ATI RageXL video and gigabit LAN. All in a Lian Li PC-626SCA case.
The machine is rock solid from what I can tell, and blazingly fast, and I'm glad I didn't have to pay for it.
So I'm guessing you're either a beta-tester for iD's upcoming Doom 3 game, or a java developer?
Code, Hardware, stuff like that.
I don't care how many times this is mentioned. Every time something comes up about the nVidia cards, I hope we get a whole flood of people posting about this problem. I don't really care what excuse they have for not having Open Source drivers. I only care that they aren't.
It makes my kernel unsafe and insecure whenever I load the non-Open Source drivers into it. I end up suspecting them first in every case of strange system behavior because I know they haven't undergone peer review. I am extremely distrustful of them, and if I had any other choice that was within 75% of the performance and used Open Source drivers, I'd jump to it in an instant, even if it was %20 more in price.
Need a Python, C++, Unix, Linux develop
You would have preferred it if nvidia hadn't published linux drivers? Thats called cutting off your nose to spite your face.
no sig.
There are about 20-30 games for Linux. Go see http://www.tuxgames.com/. I will admit though, that there are only about 10 pretty good ones. :-)
Need a Python, C++, Unix, Linux develop
How about Nvidia takes ATI's stance and writes crap drivers or no drivers at all for other OS's besides Windows? Now THAT would make everyone happy!
good post. This also means that not only is Linux the only OS that is Opteron-ready, including video drivers, it is also Doom-III ready.
---
Except that ATI's linux drivers are a far cry from crap (and actually work for me, unlike nVidia's most recent drivers which just segfault).
In addition, ATI makes the specs available so that users who want to can develop their own drivers.
As a result, under Linux, Radeon 8500 (and newer) users have 3 drivers to choose from, depending on their needs, ethics, and desires.
Dinivin
Mine used to crash all the time, but it was due to a bug in the Athlon Thunderbird.
I had this very same problem after I upgraded from a voodoo3 to a geforce2 mx. I was (and still am) dual booting win2k professional and redhat 7.3. The strange thing was that I only experienced crashes during 3d acceleration in windows - linux worked fine, even without the mem=nopentium option. I researched the problem quite a bit before I found the solution, and I got two contrary explanations. One places the blame on the cpu and one points at the OS.
The first explanation I got was that the Athlon Thunderbirds and two steppings of the Palomino core don't properly execute the invlpg instruction under certain circumstances (i.e. when 4MB pages are used), so some TLB entries can be left behind even though they shouldn't be there. This can cause the modification of areas of memory (AGP memory in use by the GART in this case) to be written to when they shouldn't be. Either the graphics card is getting confused because its data is being modified by code that thinks it's modifying its own data, or a program is tweaking out because its data is being modified by the graphics card.
The other explanation was that the instability was caused by an interaction between a "feature" of Athlon processors and OS page allocation code. Athlon CPUs allocate cache lines for speculative writes (reading memory into the cache so that cache lines can be mapped to other areas of memory, expecting that they'll be written to later, without causing problems if the cpu needs to read from that memory instead) and writes the data back to main memory whether the data is actually changed or not. This explanation says that if the OS allocates 4MB pages and marks them as cacheable and if those areas of memory are in use by the AGP GART, the cpu will "guess" that the cpu may be instructed to perform a write to that area of memory (which would never be done, but the athlon doesn't know this) and will read that area into the cache in preparation for a write. Later the cpu will write it back. Normally this wouldn't cause a problem, but since that area of memory is in use by the GART, the data may have changed since the time that it had been read by the cpu. When the athlon writes that data back to memory (even though it wasn't changed by the processor) it inadvertently writes stale data into memory used by the agp card. The graphics card gets confused and the machine then locks up.
The problem is that both explanations say that the problem only occurs when 4MB pages are used, and it goes away if the machine only allocates 4KB pages. Also, all the fixes I've seen disable 4MB pages (mem=nopentium in linux and a registry patch in win2k), so there's no way for me to be absolutely sure which is the real culprit.
I'm inclined to lean towards the first explanation because:
-It provides a reason why changing to 4KB pages fixes the problem. The CPU treats 4MB pages differently from 4KB pages, providing a distinct set of circumstances in which the INVLPG instruction won't work properly.
-The problem disappeared with later palominos and all cores after that. I don't think AMD decided to disable speculative writes, so by my logic explanation 2 predicts that the same thing should happen on all athlons.
It seems like explanation 1 is more likely to be correct, except that I have no problems under linux, even without the mem=nopentium option. Since explanation 1 relies purely on the hardware in use and does not involve the OS, it seems as if linux should crash as well when 4MB pages are not disabled. I've read AMD's processor errata and anything I could skim off of google on the subject, butI can't seem to find a really good, complete explanation.
Does anyone out there know what's really going on?
A few key differences between the two are that the Opteron will be multiprocessor-enabled and have three HyperTransport pipes (each providing a theoretical 6.4GB/s of throughput) versus one in the Athlon 64.
The last time I looked at any Hammer-related information on AMD's web site, it said that hypertransport would be used for interprocessor communication, but they would only operate at 6.4GB/sec on quad-cpu setups. Dual-cpu hypertransport pipes would operate at 3.2GB/sec, but considering the hammers will have onboard PC2100 memory controllers only a maximum of 2.1GB/sec of that could be eaten up by memory transfers (mulitprocessor athlon systems use a NUMA) so 3.2GB/sec is a reasonable number. I think the hypertransport pipe connecting the cpu to the chip that's something like a stripped-down north bridge may be even slower (1.6GB/sec?) but I'm not sure about that.
check here.
this assumes you are already using accelerated nvidia drivers and not the SVGA-nv driver that comes with XFree86-4.1xx. If that's not true, read the Nvidia readme on changes you'll need to make to your /etc/X11/XF86.config-4.
/sbin/lsmod
/sbin/rmmmod NVdriver /sbin/rmmod nvidia
/sbin/rmmod agpgart /sbin/rmmod nvgart
/etc/X11/XF86config-4 by adding the following under the "Device" section:
First, download the tag.gz source files, not the rpm. Even if they have rpms/debs for you rdistro and kernel, don't do it. You dont' save any time by using sources, assuming you have a config that compiles.
Uncompress the archive, save it somewhere logical. You can be in X-vindows if it's easier to use gedit/kedit instead of vi (but learning to use vi even minimally is a life saver; do it now b4 you really need it).
Okay, in the NVIDIA_kernel-1.xxxx directory, you'll find a file called "os-registry.c" This is the make configuration file. Read up and turn features on or off depending on your chipset (like AGP 4X, FastWrites, and sideband addressing; the first two are the better enhancements). Save your changes. I would recommend being conservative to start-- leave the defaults alone and if it works, turn on AGP4X,then Fastwrites, and then SBA in that order.
It's a good idea before you complile to a backup. Brush up on what you would need to do to install the older drivers that are already working for you.
Quit Xwindows, and become su if you aren't already. Change to to NVIDIA_GLX-1.xxxx directory first and do:
make install
this takes care of the 3D rendering backend, particulary if you already have mesa installed.
change (cd) to the NVIDIA_kernel-1.xxxx directory, and do:
make install
it should say somthing like "module nvidia.o complied successfully." If you get any errors, be ready to implement your backup plan (revert to known good drivers or edit your os-registry.c.
If you are repetitively trying to get the drivers to work or to change compile options, do a "make clean" before you "make install" in the NVIDIA_kernel directory.
okay, assume it compiled without errors, do a:
you should see the older verion of your NVdriver or nvidia module and the agpgart or nvgart modules still loaded. Just do a :
or
and
or
depending on which modules you have loaded. Okay, let's see if it works:
startx
Obviously, you should have runlevel 3 as your default-- do have a graphical GUI/logon prompt if you are going to be messing with your video drivers.
Some people have complained about a 2D slowdown; you can turn on the accelerated renderer in your
Option "renderAccel"
for reference, here's what that portion of my XF86config-4 looks like:
Section "Device"
# no known options
Identifier "NVIDIA GeForce 2 GTS (generic)"
Driver "nvidia"
VendorName "NVIDIA GeForce 2 GTS (generic)"
BoardName "NVIDIA GeForce 2 GTS (generic)"
Option "NvAgp" "3"
Option "renderAccel"
BusID "PCI:1:0:0"
EndSection
---
OK, I should have said "nVidia posted their first publically available Linux graphics drivers for the Athlon 64". My bad. I should also have highlighted the fact that while Linux Athlon 64 drivers are posted, Windows Athlon 64 drivers aren't, though I'm sure the people who actually have x86-64 hardware are able to get them. It's still nice to see Linux get official driver support first for a change.
It is also abundantly clear that my smartaleck sense of humor doesn't translate well to Slashdot. Geeze...
It seems I've struck a chord. Let's hope the metamods are paying attention.
That may or may not be true, but it doesn't mean that he's obliged to give a shit, you troll. Who're the idiots that modded this up?!?
-1 Uncomfortable Truth
Wow, 20% more in price?
And 20% more than 0 is how much, exactly?
If you don't like it, then quit whining and buy a different card. Matrox has open source drivers. Yeah, they suck. But they're open source, and that appears to be the driving factor in the whining.
One day the open-source-uber-alles zealots will get a clue and realize that the business world does have things like trade secrets and intellectual property that can't just be released. The driver code is part of nVidia's core business, and asking them to open it wide open is equivalent to asking them to just hand money to their competitors.
I'm talking total hardware and software price. It really doesn't make any sense to talk about anything else since the hardware is useless without the software, and the software is useless without the hardware.
If you read my post, you'd realize why I don't get a Matrox card. As you said, their performance is sadly lacking. I clearly laid out the parameters under which I would consider a different card.
Ahh, yes, the business world... Can't find a way to make money without installing secret, possibly trojaned software on my computer. I feel so sorry for them. Next I suppose you'll tell me they need government handouts to survive. If you want to follow along to the endpoint of that stupid road, install Kazaa, Audio Galaxy, Morpheus and Windows XP on your computer.
If you want to be there, quit using your Linux box and Open Source all together. Clearly, scam artists making money is more important to you than your freedom.
Businesses exist to serve me, and my interests. That's why they get my money. They have no intrinsic right to exist or make money. If I were complaining about the quality of their harware (which I'm not, because it's excellent), or that their drivers constantly crashed (which, though I'm very suspicious of them, I don't think they do), you'd have no vitriol to spew. Their drivers not being Open Source is just another missing feature, and one I place an explicit value on.
Need a Python, C++, Unix, Linux develop