NVIDIA GeForce GTX 680 "Kepler" On Linux

Before getting to the Linux test results, very special thanks go out to Hardy Doelfel of NVIDIA Corp. Last November, Hardy Doelfel replaced Andy Ritger as their new manager of NVIDIA Linux Graphics Driver Software. Andy is still with NVIDIA but he moved back into an engineering role on the Linux team. When he introduced me to Hardy, he let me know when new hardware is needed to just let them know. When the GeForce GTX 680 Kepler launch was forthcoming, I took them up on their offer. However, at first there was limited GPU availability. That's understandable and obviously with this high-end graphics card their first priority is with Windows gamers and those enthusiasts, since it's a bigger market and all, so that's fair enough. Several weeks passed, and still nothing.

Hardy kept trying to get one sample out of the press department at his company. In the meantime, there still have not been any GeForce GTX 680 reviews under Linux... There were a few Linux benchmark results of Kepler uploaded to OpenBenchmarking.org, the Phoronix Test Suite platform for collaborative/crowd-sourced benchmarking and analytics, but still no formal Linux review on Phoronix or any other web-site. NVIDIA Linux customers can generally expect that their hardware will "just work" with the binary blob as soon as the hardware ships, but in terms of any feature-limitations or how well the new hardware performs, that is still an open question. In terms of how well the latest NVIDIA hardware/drivers also compares to the latest AMD Radeon hardware and their sometimes-schizophrenic Catalyst Linux driver is a big question. Finally, yet importantly, the open-source NVIDIA support for each new hardware generation is also an interesting story each time with NVIDIA Corp not providing any official open-source support but leaving it up to community developers to clean-room reverse-engineer their blob.

Finally in early April following on one of the status reports from Hardy Doelfel, Bryan Del Rizzo contacted me. Bryan serves as the senior PR manager for NVIDIA and is someone who I have dealt with years ago when trying to (unsuccessfully) get NVIDIA samples for more Linux testing. This time again he was not contacting me to deliver good news, even though Doelfel had been asking for a GTX 680 sample for this Linux testing now for weeks. Bryan contacted me as he wanted to know what platform I would be running the tests on (as if their Linux software manager contacting him wasn't a big enough hint or that 99% of the content on Phoronix is Linux-focused), how I will be doing the testing (another silly question with almost every other word on Phoronix being the Phoronix Test Suite or OpenBenchmarking.org and with NVIDIA's Tegra team relying upon the Phoronix Test Suite for their internal testing and QA, likely with other Linux-focused divisions in their own company too), and the last question was asking about my power supply. In telling him I had many power supplies, he said he was making sure I had a power supply with a dual six-pin connector... As if needing dual six-pin PCI-E connectors on a power supply was something new and rare, but it has only been that way for several generations now on high-end NVIDIA and AMD graphics cards. These questions were a bit silly for someone that has written over 440 articles on Linux graphics drivers, over 120 Linux graphics card reviews, and more than one thousand Linux graphics related news posts in the past eight years. After - politefully - answering these questions, there was no response back from Del Rizzo.

Fortunately, Hardy followed back up with me to see if I had any hardware yet from NVIDIA marketing. After telling him everything has been quiet, he responds, "Ok, we've waited long enough. I'm going to ship you a GTX680 from MSI, I bought for my team. Can you send it back after you are done with the evaluation?" So, in the end, the NVIDIA Linux Graphics Driver Software manager sent out the GeForce GTX 680 that he had personally purchased, so that I could borrow it for Linux Kepler testing after having waited weeks with NVIDIA PR/marketing not being able to provide a sample for Linux testing. It has been about one month since the GTX 680 first shipped and this is still the first in-depth Linux review of the graphics card hitting the web.

This also is not the first time that something like this has happened in NVIDIA PR/marketing being unable to get samples for Linux coverage. This is pretty much a repeat of what happened when trying to get a Fermi sample for Linux benchmarking. In fact, since 2004 I do not remember a single instance where NVIDIA marketing successfully delivered for Linux; it is usually some AIB partner sending out the hardware, me buying the hardware, or the NVIDIA Linux team buying the hardware. The last time hardware was requested from NVIDIA for Linux tests, Andy Ritger was still in charge of the Linux team and he requested a Fermi sample for Phoronix well after the first GeForce 400 hardware was shipping. After having waited a while, he too was fed up in waiting and ended up voluntarily purchasing a graphics card himself to send out to Phoronix. "Since marketing was dragging their feet, I purchased the one I sent you, so you can keep it." This was the GeForce GTX 460 that you still commonly find in Phoronix testing when it comes to NVIDIA Linux graphics driver articles looking at improvements (e.g. when benchmarking every NVIDIA driver release from the past year), testing of the open-source Nouveau driver, and various other NVIDIA Linux articles.

NVIDIA marketing just has no love for Tux?

I think NVIDIA PR/marketing may just fail with Linux; they evidently just do not care. Need I remind Phoronix readers that it was NVIDIA's Andrew Fear who famously said, "It's so hard to write a graphics driver that open-sourcing it would not help." Yet the Nouveau driver is continuing to be worked on to this day, they have been making performance improvements, and they even already managed basic accelerated support for Kepler. This open-source NVIDIA driver is being largely written by unpaid independent Linux developers doing the clean-room reverse-engineering in their spare time (it's really only Ben Skeggs of Red Hat that's now working on it full-time in an official capacity). For doing something that's "so hard" and most of them not even being compensated for it, they are doing a damn fine job. NVIDIA even conceded with their officially supported open-source xf86-video-nv driver when Nouveau took over. Outside of graphics, open-source also showed NVIDIA wrong back in the day when it came to the "Forcedeth" driver, among other occurrences.

For reference though, there really is not any tier-one hardware vendor with a marketing department that is really knowledgeable and too caring about Linux. Most of them at least just send hardware without questions asked or to just surprise me with hardware at the door in advance of a new launch. When questions do arise from the marketing folks at the other vendors, it's usually them asking me about the level of Linux support for their upcoming/current products, asking me to explain to other media/press people about Linux support / driver setup for their products, or similar matters as they know their Linux developers and I are frequently communicating and out on beer-drinking adventures as my form of "media briefings" with the developers on next-gen hardware support under Linux. On the plus side, at least I'm not inundated with shit listening about DirectX, other Windows-only features for new products, or how their latest proprietary API is going to be the best feature ever.

So anyhow, special thanks to NVIDIA's Linux graphics manager, Hardy Doelfel, for letting me borrow this GeForce GTX 680 that he purchased in order to provide Linux results for the community. Kudos as well to Andy Ritger for his support and information pertaining to the reported GTX 680 clocking discrepancies.