Cryptic output from lspci

[Yellow Pasque]

All of this sounds easy, but it had some bothersome aspects. Additional Drivers gave me no clue which driver was the better one for this adapter, or whether either one would even work.

It would not offer you drivers not known to work with your GPU.

I wonder what will happen when I insert other NVIDIA adapters in the machine. Will this driver work with all of them, or will I have to install different drivers for different adapters? If this driver works with all adapters (or even with many), will nvidia-smi also work, or will it malfunction if the driver is wrong for the card? If nvidia-smi malfunctions, will it do so obviously or non-obviously? If I need different drivers for different adapters, can I avoid having to install a driver for each adapter I test by grouping adapters into families that use the same driver, and if I can do that, how?

http://www.nvidia.com/object/IO_32667.html

If you've got the correct driver installed, nvidia-smi should work (assuming you're using an nvidia proprietary driver).

I'll conclude by reproducing the information that nvidia-smi gave me. Beyond the obvious (this adapter is at least nominally a GTX 760), I'm not sure what is significant. Any comments, hints, etc. are welcome. I'm particularly puzzled by the value of 0 under "GPU." Is it saying that this card has no GPUs (or that for some reason no GPUs are functioning)? If so, what does that mean?

It's a GTX 760 with 2GB of memory. What more are you looking for?
The 0 just means the GPU numbered 0 (computer geeks like to start counting at 0). If you had run nvidia-smi with sudo/root privs, you would actually see the processes belonging to GPU #0 listed.

[orthoducks]

> What more are you looking for?

I'm looking for the things that I don't know I should look for. I'm trying to avoid the too-common habit of assuming that because I don't know much about a subject, there isn't much to know!

It sounds like there are no precious Easter eggs hidden in this tool, though, so I'll let it be. When I find time I'll read through the man page, and when my eyes uncross again I'll probably be a lot better informed.

[orthoducks]

I'm back, with a somewhat different kind of question. Now that I know how to identify video cards, how reliable are the results?

I'm asking because I just tested a card which nvidia-smi calls a Quadro K5000. In case you aren't on intimate terms with the K5000, it's a double-wide card with two DVI outputs and two DisplayPorts, intended for professional modeling, animation, etc. It is selling new for $600 to $850, and used on eBay for $300 to $400.

But this card looks nothing like the pictures of the K5000 I've found. The bracket layout is right, but that's about all. The K5000 is shown as a full length card with a black plastic shell and a turbine-bladed fan near the front end. This card is about 3/4 length, with no shell, and a 9-blade fan that is closer to the back. (If it were as far forward as the ones in the pictures, it would be hanging off the end of the card.)

I thought I might have a K5000 whose shell had been removed, but the fan makes that impossible; this is really a differently designed card.

I have two theories: nvidia-smi could be misidentifying the card, or it could be some kind of prototype. I don't like either theory much. The first one seems inherently unlikely; its only virtue is that it explains the card. The second one demands an answer to the question: Why would anyone produce a prototype card that looks so different from the final product? (And why would a prototype have various machine printed tags with a part number, a ROM version number, "Made in China," and "EMC Certification Pending"?)

I wanted to run a benchmark to see if the card is about as fast as a K5000, but I haven't found a graphics benchmark program for Linux for which results from a genuine K5000 are available.

Comments on any aspect of this are welcome: either of my theories, your own theory, or where to find suitable benchmarking software (and/or benchmark results).

[Yellow Pasque]

Now that I know how to identify video cards, how reliable are the results?

The PCI ID database is usually accurate, but no one's going to give you a 100% guarantee that it's infallible, and it feels like that's what you're looking for.

I thought I might have a K5000 whose shell had been removed, but the fan makes that impossible

Why? If someone removed the blower ("shell"), they would have to put another fan on it.

But this card looks nothing like the pictures of the K5000 I've found.

Not every manufacturer follows the Nvidia reference design, and of course, someone could have put a different cooler on it (maybe the old one was too long or the fan died).

various machine printed tags with a part number, a ROM version number

Did you try googling them?

or it could be some kind of prototype.

It's not impossible, but it seems like a significant leap logically.

[orthoducks]

>>Now that I know how to identify video cards, how reliable are the results?
>The PCI ID database is usually accurate, but no one's going to give you a 100% guarantee that it's infallible,
>and it feels like that's what you're looking for.

Not really. I've encountered an anomalous situation, and I'm seeking information about probability that one of its elements is misinformation. I do not see how recognizing the possibility of error, and seeking information about risk, implies a demand for infallability.

>> I thought I might have a K5000 whose shell had been removed, but the fan makes that impossible



> Why? If someone removed the blower ("shell"), they would have to put another fan on it.

I've already ruled out the "shell removed" theory on other grounds, so it's moot. But since you ask, I thought that because there is a fan on the card, and (although I didn't say so) it doesn't look like it was patched on after assembly. My impression is that a "real" K5000 also has a fan mounted on the board, and the shell could be removed without affecting it, but I can't verify that without finding a "real" K5000 and taking it apart, which isn't likely to happen.

>> But this card looks nothing like the pictures of the K5000 I've found.



> Not every manufacturer follows the Nvidia reference design, and of course, someone could have put a different cooler on it (maybe the old one was too long or the fan died).

The card is an NVIDIA card, with the NVIDIA name printed on the circuit board, not just on the chips. Virtually all of the cards I'm testing are NVIDIA cards, so if I had a question about some other brand of card it would be the exception, and I'd mention it. I'm sorry I didn't make this explicit in my last message, although I mentioned it earlier in the thread.

>Did you try googling them?

No; that might be informative, and I'll try it tonight. I'll post again after I do.

[Yellow Pasque]

EDIT: nvm

[orthoducks]

Temujin, I use many different forums with different conventions. Most of them have a "quote" feature, but they all work differently, with different icons in different places on the toolbar. I do not have time to keep track of them all. Instead I usually follow Internet conventions, which are inelegant but universally understood.

I may liken myself to a trader visiting a multitude of tiny islands in the South Pacific. Rather than try to learn each island's dialect I use pidgin English, which may sound crude but is understood. If some islanders refuse to deal with me because I do not speak as they do, that will not lead me to master their speech. If they make things too difficult for rne, it I will stop visiting that island.

I need to clarify the question I asked earlier. When I asked "how accurate are the results?" I was asking a broader question than "how accurate is the database?" I meant, "How accurate is the program, the data it acquires, and the database it uses, taken as a whole?" The database might be perfectly accurate and still yield incorrect results because the acquired data is not detailed enough, and/or the program makes inferences from it that are misleadingly specific. If the whole not always accurate (thus explaining why this mystery card is nailed as an NVIDIA K5000), the causes of the inaccuracy are of interest, but are secondary to the fact of inaccuracy.