The Dirty List Of GPUs With Open-Source Drivers Gone Wildly Wrong
This morning I shared the list of the 60+ graphics cards being tested under Linux for a set of very interesting articles coming up in the days ahead in this massive Linux graphics comparison in celebration of Phoronix.com's 10th birthday next week. While all of the graphics cards were tried, with the open-source drivers there were notable failures with both the AMD Radeon and Nouveau drivers.
The first round of testing with all of these graphics cards were when using the Linux 3.15 kernel as of a few days ago along with Mesa 10.3-devel and the updated DDXes as well, with the user-space components being obtained from the Oibaf PPA on Ubuntu 14.04 LTS x86_64. While a majority of the graphics cards tested successfully completed the OpenGL and 2D benchmarks fine on the open-source Intel/AMD/NVIDIA Linux graphics drivers, it wasn't without quite a few failures too.
Within the article next week, I'll cover more closely all the struggles encountered with each of the graphics cards tested. Besides being the biggest GPU comparison ever conducted at Phoronix, it will also likely be the longest set of articles too... However, for those wanting the short-story, embedded below are just my unedited notes I recorded when trying out the large assortment of graphics cards. Here were the problematic NVIDIA GeForce GPUs with the Nouveau tests of this latest driver stack:
8500GT: Lost sync/display ater starting tests.
8600GT: Hung on Reaction Quake 3 (fail ttm validate), Warsow, and Unvanquished
8800GT: GPU lockup when loading Tesseract, dmesg issues when running Reaction... display ultimately disappears quite some time into the testing process. Also some corrupted text on booth
9600GSO: After a while, font rendering was really poor
GT 240: mode-set, then corrupted screen -- as always been the case for that card
GT 520: Hung while loading Unvanquished
GTX 550 Ti: hangs on Reaction. Unvanquished failed with DRM errors about failed to idle channel along with Warsow. GPU Lockup with Xonotic
GT 610: extremely slow, monitor would stop displaying after a while
GTX 750 / 750 Ti: Poor Maxwell Nouveau support for now
GTX 780 Ti: display failed to light up, after few tries, got it running on LLVMpipe only
GTX TITAN: Bad rendering with Unigine tests and Unvanquished
All NVIDIA Cards: obviously no good re-clocking...
There were several cards running into mode-setting display issues, while for other cards there were stability issues when running some OpenGL workloads. The tests most easy for thrashing the driver were Unvanquished and Xonotic. For performance testing, all of the cards suffered due to the already well-covered re-clocking issues.
On the AMD side, the list of Radeon/FirePro issues on the open-source driver stack included:
X1800XT: bad rendering on some tests
X1950PRO: failed to schedule IB when it came to Tesseract test...
HD 2400PRO: stability problems soon as testing began....
HD 2900XT: system would hang after booting, lose display
HD2600 pro: ring 0 stall
HD 3650: Stalls
HD 4870X2: lights up, but no CrossFire open-source support...
HD 5750: lose display after a while of testing...
HD 7850: Randomly the system would hang when running Unigine Tropics. Been the recent LLVM issues to note.
HD 7950: lost display during 2d benchmarking...
R7 260X: Also broken reclocking
R9 270X: Many GPU faults detected..... si_restrict_performance_levels_before_switch failed, ring 3 stalling, etc.... Triggered by Unigine Tropics. Unvanquished also causes the problems.
R9 290: Open-source OpenGL Hawaii support broken....
RadeonSI broadly: register allocation issues with Xonotic on higher quality settings.... / broken register spilling
FireGL V8600: mouse cursor worked, but system unresponsive
FirePro V8750: black screen
The most common open-source Radeon issues were stalls with the older R600/RV600 GPUs, some issues with the more demanding OpenGL tests on RadeonSI, the Hawaii GPU support being broken, and a couple other issues.
Stay tuned for more information and benchmarks in the days ahead.
The first round of testing with all of these graphics cards were when using the Linux 3.15 kernel as of a few days ago along with Mesa 10.3-devel and the updated DDXes as well, with the user-space components being obtained from the Oibaf PPA on Ubuntu 14.04 LTS x86_64. While a majority of the graphics cards tested successfully completed the OpenGL and 2D benchmarks fine on the open-source Intel/AMD/NVIDIA Linux graphics drivers, it wasn't without quite a few failures too.
Within the article next week, I'll cover more closely all the struggles encountered with each of the graphics cards tested. Besides being the biggest GPU comparison ever conducted at Phoronix, it will also likely be the longest set of articles too... However, for those wanting the short-story, embedded below are just my unedited notes I recorded when trying out the large assortment of graphics cards. Here were the problematic NVIDIA GeForce GPUs with the Nouveau tests of this latest driver stack:
8500GT: Lost sync/display ater starting tests.
8600GT: Hung on Reaction Quake 3 (fail ttm validate), Warsow, and Unvanquished
8800GT: GPU lockup when loading Tesseract, dmesg issues when running Reaction... display ultimately disappears quite some time into the testing process. Also some corrupted text on booth
9600GSO: After a while, font rendering was really poor
GT 240: mode-set, then corrupted screen -- as always been the case for that card
GT 520: Hung while loading Unvanquished
GTX 550 Ti: hangs on Reaction. Unvanquished failed with DRM errors about failed to idle channel along with Warsow. GPU Lockup with Xonotic
GT 610: extremely slow, monitor would stop displaying after a while
GTX 750 / 750 Ti: Poor Maxwell Nouveau support for now
GTX 780 Ti: display failed to light up, after few tries, got it running on LLVMpipe only
GTX TITAN: Bad rendering with Unigine tests and Unvanquished
All NVIDIA Cards: obviously no good re-clocking...
There were several cards running into mode-setting display issues, while for other cards there were stability issues when running some OpenGL workloads. The tests most easy for thrashing the driver were Unvanquished and Xonotic. For performance testing, all of the cards suffered due to the already well-covered re-clocking issues.
On the AMD side, the list of Radeon/FirePro issues on the open-source driver stack included:
X1800XT: bad rendering on some tests
X1950PRO: failed to schedule IB when it came to Tesseract test...
HD 2400PRO: stability problems soon as testing began....
HD 2900XT: system would hang after booting, lose display
HD2600 pro: ring 0 stall
HD 3650: Stalls
HD 4870X2: lights up, but no CrossFire open-source support...
HD 5750: lose display after a while of testing...
HD 7850: Randomly the system would hang when running Unigine Tropics. Been the recent LLVM issues to note.
HD 7950: lost display during 2d benchmarking...
R7 260X: Also broken reclocking
R9 270X: Many GPU faults detected..... si_restrict_performance_levels_before_switch failed, ring 3 stalling, etc.... Triggered by Unigine Tropics. Unvanquished also causes the problems.
R9 290: Open-source OpenGL Hawaii support broken....
RadeonSI broadly: register allocation issues with Xonotic on higher quality settings.... / broken register spilling
FireGL V8600: mouse cursor worked, but system unresponsive
FirePro V8750: black screen
The most common open-source Radeon issues were stalls with the older R600/RV600 GPUs, some issues with the more demanding OpenGL tests on RadeonSI, the Hawaii GPU support being broken, and a couple other issues.
Stay tuned for more information and benchmarks in the days ahead.
54 Comments