Nouveau vs. NVIDIA GeForce Linux Performance At The End Of 2014

[phoronix]

Phoronix: Nouveau vs. NVIDIA GeForce Linux Performance At The End Of 2014

Complementing yesterday's Radeon Gallium3D vs. Catalyst Linux benchmarks that compared AMD's open and closed-source Linux graphics drivers using the latest code at the end of 2014, here's similar benchmarks done in comparing the open-source Nouveau driver against the closed-source NVIDIA Linux graphics driver when testing several GeForce GPUs.

http://www.phoronix.com/vr.php?view=21286

[imirkin]

You can hit playable frame-rates if you have a higher power NVIDIA Kepler GPU and can hit the 0a performance level, but it's sort of a waste and lower-end NVIDIA GPUs would run at more crippled speeds.

http://www.phoronix.com/vr.php?view=21286

Of course if you were using a lower-end Kepler card, you'd more likely be able to reclock to the maximum speed (GDDR5 being the main sticking point in reclocking failures... if you have DDR3, success is more likely). It's also a little disingenuous to leave out the GTX 650, which in your previous articles, you showed could reclock all the way to the highest level just fine, and performed at something like 60-80% of blob speed, depending on the test.

[duby229]

Is it possible to flash the video cards bios with an image that has the clock speed set to max? Or does it not work that way?

[imirkin]

Is it possible to flash the video cards bios with an image that has the clock speed set to max? Or does it not work that way?

I don't see why not - it would be difficult to change the vbios around, but certainly possible. In fact for a bunch of these cards, you can get them into highest mode by telling nouveau to switch to that mode at boot (nouveau.config=NvClkMode=0f). But not always.

[Michael]

Of course if you were using a lower-end Kepler card, you'd more likely be able to reclock to the maximum speed (GDDR5 being the main sticking point in reclocking failures... if you have DDR3, success is more likely). It's also a little disingenuous to leave out the GTX 650, which in your previous articles, you showed could reclock all the way to the highest level just fine, and performed at something like 60-80% of blob speed, depending on the test.

On all the Kepler cards I've tried, only with the GTX 650 was I ever able to reach the 0f speed. The reason it wasn't tested in this article is that it's part of the systems running daily benchmarks via LinuxBenchmarking.com, as the card was used since it can reclock and is power efficient for daily testing.

[Xaero_Vincent]

Yeah, it's generally possible to replace the VBIOS. I did this with my AMD Radeon 5850 so that it could be overclocked beyond what AMD Overdrive generally permits with MSI Afterburner.

I'm planning to try AMDOverDriveCtrl, which is a GUI program almost exactly like MSI Afterburner but for Linux and AMD Radeon cards using the proprietary Catalyst driver. Overclocking the card should offset any slight performance loss experienced by the lesser optimized Linux Catalyst driver and make it run OpenGL and Wine+CMST games at least as fast as on Windows at stock GPU and Memory clock speeds and maybe even slightly faster.

Some of the newer Nvidia Maxwell cards now require signed BIOS firmware or else not all low-level functionality gets exposed so the future trend, at least for Nvidia, is to prevent you from changing the firmware.

[Luke]

Scam alleged source of the signed firmware on Maxwell, but camel's nose under tent

Yeah, it's generally possible to replace the VBIOS. I did this with my AMD Radeon 5850 so that it could be overclocked beyond what AMD Overdrive generally permits with MSI Afterburner.

I'm planning to try AMDOverDriveCtrl, which is a GUI program almost exactly like MSI Afterburner but for Linux and AMD Radeon cards using the proprietary Catalyst driver. Overclocking the card should offset any slight performance loss experienced by the lesser optimized Linux Catalyst driver and make it run OpenGL and Wine+CMST games at least as fast as on Windows at stock GPU and Memory clock speeds and maybe even slightly faster.

Some of the newer Nvidia Maxwell cards now require signed BIOS firmware or else not all low-level functionality gets exposed so the future trend, at least for Nvidia, is to prevent you from changing the firmware.

Nvidia is claiming that there were dishonest sellers in Asia or Russia (forget which) taking low-end cards and flashing them with the BIOS of high-end cards and selling them as such. Apparently there were some cards which would basically function but of course with no more performance than they orginally had and probably a lot more bugs. Supposeduly Nvidia intends to make that scam impossible with the signed firmware. Similar scams are known to exist in the sale of USB flash drives, along with the "badUSB" firmware attack code that can be loaded into some USB drives. They might be next for signed firmware, but the trend becomes an ugly one.

One option would be for makers to exempt from the signed firmware requirement their top shelf items, plus those that cannot pass for anything other than what they are. I cannot see an Ebay seller convincing anyone that a tiny, half-height card with a tiny passive heatsink or even smaller fansink is a 250 watt. $1,000 gaming card.

There is also this question: is the "signed firmware" requirement burned into ROM somewhere in the GPU, or is it on a flash rom somewhere that a specialized reflashing board (like a bus pirate for GPU's) could change? A motherboard requiring a signed UEFI would be a mess, but if the requirement was in any flash location on the entire board it could be changed with a bus pirate type tool. A signed UEFI requirement burned into Intel CPU's would be a real mess, especially if AMD followed suit. After that, all signed firmware booting only signed operating systems, and/or ISP's refusing connections from "unauthorized" computers.

[duby229]

I don't see why not - it would be difficult to change the vbios around, but certainly possible. In fact for a bunch of these cards, you can get them into highest mode by telling nouveau to switch to that mode at boot (nouveau.config=NvClkMode=0f). But not always.

Well, if the driver supports boot time ability to change clock, that would be ok, just need a card that you know will work like that.

[david_lynch]

As always, FPS doesn't tell the whole story. I had a Dell GX 270 desktop with an older nvidia card, and had always used the nvidia binary blob, which worked, and enabled decent 3D performance. However, after a distro upgrade upon which the Nouveau driver became the default, I was pleasantly surprised at the improvement in 2D performance for everyday use. For instance, grabbing an xterm and dragging it around the screen was smooth and snappy under Nouveau, which made me realize how sluggish the 2D performance had been with the binary blob. The nouveau driver was much more pleasant to use, for normal destop activities. The gaming performance wasn't there, but nowadays it's getting closer.

Just my 2 cents