Yes, it conflicts with HaikuPorts Mesa package. You need to agree to uninstall it. Nebula fully replace HaikuPorts Mesa package.
I rebooted but it does not seem to work. Anything using GL crashes immediately. What logs could be useful for you?
/boot/system/var/log/syslog
Hm… says “nvidia: init_hardware: no supported device found”. Bummer
“nvidia” is very old Nvidia driver, it is a different thing. Look for nvidia_gsp and NVRM.
With “nvidia_gsp” I see nothing special, no errors. With “NVRM” I see this in particular: “NVRM: RmFetchGspRmImages: No firmware image found” and just before it “[!] can’t read firmware”
I did a quick checkup. I’ve got this device: “device 2420: GA103M [GeForce RTX 3080 Ti Mobile]”. According to the nvidia open gpu kernel modules this device is supported.
Check file attributes and search location of: gsp_ga10x.bin
It finds it in /boot/system/data/firmware/nvidia/570.86.16
What could possibly cause the driver to not find the firmware images in this directory?
It probably fails to detect required firmware file name, pass empty string and fails with file not found. That may mean unsupported hardware (Nvidia driver version 570.86.16). Adding more logging may help.
I did a quick check and that driver version (570.86.16) supports the “GeForce RTX 3080 Ti Mobile”. Definitely needs more logging since this driver/device combination should be supported.
Will this be integrate into haiku?
Maybe the problem is in “Mobile”. It may need some extra ACPI setup to be functional.
Hm… I took a look at the source code. nv_get_firmware(1575) would return nullptr if the firmware file is not found. But it does output “can’t read firmware” instead. So the problem is not that the file is not found but that it can’t be properly read for some reason. I checked the files. “/system/data/firmware/nvidia” is 555. “570.86.16” is 555. “gsp_ga10x.bin” is 444. To me this looks all correct. No idea what’s going wrong here.
Edit. The code line that has to fail is this one:
off_t readSize = read_pos(file.Get(), 0, &data[0], size);
if (readSize == -1 || readSize < size) {
I saw 3Deyes used the Nvidia GTX 3060 with Zink driver of Mesa 25.1.0:
The nebula driver uses the family GPU code names for firmware lookup so I don’t think the product names (i.e. * Mobile) are the issue.
I ran into an issue with a specific laptoip in which Haiku didn’t read/load certain firmware - but read/load the same firmware when using Linux/BSD.
Need to test the nebula driver on a working desktop with various “supported” cards and log the firmware results from syslog.
Ref: https://review.haiku-os.org/c/haiku/+/8011
This was planned for the upcoming Beta 6 release… (but not for Haiku x86 (32-bit))
It’s not even a problem with loading the firmware into the GPU. The problem is with reading the file. That’s what’s strange.
Hopefully this will expand support to Intel and AMD or glide, even though this one not used anymore, I really am trying to like unlock hardware acceleration instead of buying a very expense or god tier CPU that can run better than brute force software rendering cuz I don’t really wanna buy a AMD Ryzen 5 7500F or Intel Core i5-14600K or I was just gonna buy a thread ripper, but that cost around like a 9K so this would be very helpful if hardware acceleration really expanded it support
AMD’s EPYC with 192 cores … it’s about 10cm x 10cm and cost approximately 28000 Euro - you can get a motherboard that takes two. ![]()
Price was the reason I bought a 9600X - just for testing if Haiku would run before I buy a 9950X.
I’ll try and see if I can test Haiku on a 5600G (will work the same as 5700G). It pays very much to purchase an AM4 machine today with the high DDR5 prices (well DDR4 also rose a lot, but you get a very good motherboard with 3 PCIe slots and some NVMe slots for around 100 Euro).
I picked AMD over intel, because I checked out the performance of the CPU on compiling first. A low-cost AMD Ryzen 5 3600 outperformed my brother’s brand new high-cost Intel CPU (don’t remember which one it is). -Scroll down until you see the list.
The test results changes each day, because the Linux kernel changes (thus a new revision is used) and new users contribute to the tests. If you go on YouTube, you only see how many FPS a system can get in games, but the Timed Linux Kernel Compilation is useful for people who develop or build existing software. It may be slightly inaccurate, because some people buy slow RAM, others buy top-dollar RAM and tweak their settings to perform the best it possibly can, but often you get only a few percentage extra like 1% … 2% compared to a fairly good RAM (this applies to both AMD and Intel).