Slashdot Mirror
Anand Tours ATI and NVIDIA
logicalstack writes "The folks over at AnandTech have written an
expose on their visits
to both ATI and NVIDIA. Interestingly enough ATI's facility shrouded in secrecy and NVIDIA's is quite open, Including full color pictures of their server farm, and a pic of the NV30 test machine the 'Ikos.' The CEO even showed off the old school NV1 with 1MB of ram!"
Here
I've had enough abrasive sigs. Kittens are cute and fuzzy.
NVIDIA programs their GPUs into the IKOS boxes and they run what is effectively a very large NV30 at very slow speeds. Very cool and very expensive.
When I was there a few years ago, they would sometimes hijack all the desktops in the company for more power. If I remember correctly, they would boot them into linux at night and make (slashdot cliche imminent!) a beowulf cluster!
-Greg Daly, formerly of riva extreme, aka
Try reading the article:
ATI imposed very strict restrictions on photographs during our visit to their offices in Thornhill, Ontario; we saw a lot of interesting things at ATI's offices (including the foundation for their fountain of fire in the lobby of their main building) but we weren't able to take pictures of most of them. On the other hand, ATI sat us down with one of their chip architects and we were able to get a wealth of information about how their GPUs were made.
NVIDIA wasn't able to set us up with any engineers for an extended period of time (although lunch with Chief Scientist, David Kirk is always informative) but they were much more lax on the picture front so we were able to bring you more of the behind the scenes from NVIDIA.
ATI just didn't want anybody taking pictures, but they were the one sharing the real information.
Never underestimate the bandwidth of a 747 filled with CD-ROMs.
Hear hear! I refuse to use Nvidia based on that fact. I have an ATI Radeon 7200. It may not be as fast, nor may it be quite as good, but at least I'm not putting proprietary software on my machine when there is an alternative.
They're binary-only because they don't own all the code used in them so it would break other licenses to publish it.
I use those binary-only drivers myself with a GF3 and have had no problems with X crashing.
ATI was the company that provided the in depth talk with a chip engineer. With NVIDIA, Andtech had to settle to having lunch with their lead architect. NVIDIA was okay with pictures, but ATI was the one that provided real information.
"Old man yells at systemd"
Well, I suppose we could go over the reasons for the billionth time on /., but there's no reason to believe that you'll listen this time either.
Oh well. Here it goes anyway.
The primary reason is that they cannot. They do not own all of the code that is in the drivers. There are extensive cross-licensing agreements between nVidia and SGI, dating back to the creation of nVidia from a bunch of ex-SGI engineers and the ensuing lawsuits. A good bit of the core code in nVidia drivers is essentially owned by SGI. nVidia cannot release that code. Period. End of story.
The secondary reason is that there is reason to believe that there are trade secrets in the drivers. Why do most people still favor nVidia over ATI? Because of the drivers. They work damn well most of the time, and the drivers you download today still support the original TNT. Additionally look at the GeForce4 and the Radeon 8500. On paper the Radeon 8500 was superior, and yet the GF4 beat it in benchmarks consistently. Why? The drivers. They were more mature, better written, and streamlined.
Don't like the situation? Fine, don't buy an nVidia card. What? Nobody else has 3d acceleration worth a crap? All the other drivers are just as unstable and slower too? Well, gee, maybe there's more proof that nVidia knows what the hell it's doing. Yes, it sucks if you're a *BSD fan or something else such that the binary-only drivers aren't usable, but, again, nobody made you buy nVidia.
Frankly, nVidia has spectacular Linux support. They release the Linux drivers within weeks of the Windows drivers and they're pretty damn stable (frankly, I suspect that if you have continual issues here that it's some other piece of hardware being marginal and pushed over the limit by running the card at full functionality). Oh yeah, and they're fully functional... don't forget that little bit.
It's really sad to see people whining for Linux support, getting pretty damn spectacular support, and then whining that it's not good enough. No wonder most manufacturers don't bother - damned if you do, damned if you don't. So why spend the time and money on a marginal market if you're just going to get roasted anyway?
> (and buggy) Linux drivers?
You might want to try switching from nvagp to agpgart, or vice versa, depending on your mobo.
I went from agpgart->nvagp a few months ago, and suddenly most of my stability problems with the drivers were gone.
But they work find for the majority of the people that use them. I'm not saying their drivers are perfect, but I'd be willing to bet that the nearly all the people complaining of them crashing are having problems with something else.
I'm glad you could publish your drivers under the GPL, but not everyone is and I'd rather have closed drivers that work well for me than no drivers at all.
I fully agree. Last year I was at GDCE doing research at tying to get info from both ATI and nVidia for articles I was writing. all the people from ATI I met were fantastic, I spoke to a few of the heavy engineers (huge kudos to Alex Vlachos and Jason Mitchell) along with the PR and Product Management people. They all went massively went out of their way to help me and inform me, answering any questions I had, burning me cd's of demos, pics, info, etc and following up further technical questions via e-mails and phone calls after the conference was over. This was the overall attitude of ATI at the conference.
Next we come to nVidia, I repeatedly came up against a brick wall, the case was the same for other developers, with David Kirk doing a fine politician-style non-answering of questions after his presentation. You generally got the impression that there were a select few that may be lucky enough to be given certain information, but it was very much on nVidia's terms.
Fair enough, companies have secrets which they need to keep, but from my experiences with the companies, ATI are far far more open. If anything this article backs that up. Would you prefer a bundle of photos or a chance to talk with a variety of the actual engineers?
I don't think SGI's old IP is the reason NVIDIA won't release source code. The real reason is that the drivers deliberately cripple certain advanced features on NVIDIA's low-end cards, to force "professional" users to buy their high-end cards.
Remember how the $600 Quadro2 hardware was exactly the same as a $200 GeForce2, except for a tiny little resistor? I'm sure there are a few places in the NVIDIA driver like:
if(user_paid_for_quadro()) {
make_antialiasing_fast();
enable_overlay_planes();
} else {
make_antialiasing_slow();
disable_overlay_planes();
}
So naturally a few days after they release the driver souce, somebody would provide a "magic" version of the driver that makes all of NVIDIA's low-end cards perform just like their high-end cards. Then they wouldn't be able to charge $600 for "pro" video cards anymore...
Hear hear! And why walk with *two* legs when we clearly only need one!
I don't get it. I'm all for Open Source, but I'm even *more* for a company taking an active interest in supporting their hardware under Linux. I've got a GEForce2 on my system and the drivers are *sweet*. Full support of *all* the hardwares features. How often do you get that under Linux? Not to mention the fact that the drivers compare nicely with their Windows counterparts.
Why spend the same amount of money for hardware that has less support and will effectively run slower because of it? I just don't get it...
If every hardware company were like NVidia we would have far less trouble buying a new printer/modem/videocard/etc.
"Ignorance more frequently begets confidence than does knowledge"
- Charles Darwin
It's not nearly that simple.
NVIDIA cards are unlike anything you've ever seen on the inside. It's not a simple matter of register banging like most hardware. And yes, there is quite a bit of proprietary/trade secret stuff in there, such that publishing the driver source or opening the hardware interface would be detrimental to NVIDIA.
As much as we all hate it, the tech industry is largely driven by trade secrets, patents, and lawsuits. I don't think anyone at NVIDIA really likes that, but it's the only way to survive given the broken state of IP laws in the world.
-John (having contributed to the driver)