Nvidia seems to be more and more desperate.
Announcement
Collapse
No announcement yet.
NVIDIA Cleans Up GSP Firmware Binary License
Collapse
X
-
Originally posted by blacknova View PostThe company just hit ~1 trillion on market capitalization recently. Desperate is the correct word here, right...
Interesting point amazon, alibaba, google... those developing their own AI accelerations and IBM/Redhat have a requirement in common they don't want blackboxes that they don't know what in heck is going on inside.
Getting to the top can be easy but staying there can get extremely hard.
- Likes 2
Comment
-
I don't know enough about how the internals of the hardware and software work, but I always wished Nvidia had a driver (or an option of the proprietary driver) that would initialize the card and handle things like temperature/clocks, but allow a hand-off from there to Mesa drivers. An unsupported/untrusted pass-through mode, for those of us who would rather use Nouveau or NVK.
It's just a bummer, my laptop with an MX250 was born, lived, and will likely die running the Intel iGPU because Nvidia locked driver devs out for so long.Last edited by mangeek; 31 May 2023, 11:55 PM.
Comment
-
Originally posted by oiaohm View Post
The problem here is that ~1 trillion on market capitalization means nothing to the parties most important to Nvidia future. Its not like that market cap is going to stop IBM/Redhat wanting all Linux kernel drivers to be mainline and open source. The ~1 trillion on market capitalization not going to stop amazon, alibaba, google.... from developing their own AI accelerators or change the core reason why they are doing this.
Interesting point amazon, alibaba, google... those developing their own AI accelerations and IBM/Redhat have a requirement in common they don't want blackboxes that they don't know what in heck is going on inside.
Getting to the top can be easy but staying there can get extremely hard.
As far as competitors building their own accelerators. That would require them to get access to advanced fab nodes and somehow not infringe on nvidia IP, good luck with that. At the moment no one else even comes close to what nvidia is doing in this field, they’re basically writing the standards themselves.
Comment
-
Originally posted by oiaohm View Post
The problem here is that ~1 trillion on market capitalization means nothing to the parties most important to Nvidia future. Its not like that market cap is going to stop IBM/Redhat wanting all Linux kernel drivers to be mainline and open source. The ~1 trillion on market capitalization not going to stop amazon, alibaba, google.... from developing their own AI accelerators or change the core reason why they are doing this.
Interesting point amazon, alibaba, google... those developing their own AI accelerations and IBM/Redhat have a requirement in common they don't want blackboxes that they don't know what in heck is going on inside.
Getting to the top can be easy but staying there can get extremely hard.
Sure some didn't work out as well as they had hoped, but that's what it takes to be on top: the first to innovate and be good at it (looking at you 🍏)
- Likes 1
Comment
-
Originally posted by oiaohm View Post
The problem here is that ~1 trillion on market capitalization means nothing to the parties most important to Nvidia future. Its not like that market cap is going to stop IBM/Redhat wanting all Linux kernel drivers to be mainline and open source. The ~1 trillion on market capitalization not going to stop amazon, alibaba, google.... from developing their own AI accelerators or change the core reason why they are doing this.
Interesting point amazon, alibaba, google... those developing their own AI accelerations and IBM/Redhat have a requirement in common they don't want blackboxes that they don't know what in heck is going on inside.
Getting to the top can be easy but staying there can get extremely hard.Last edited by MorrisS.; 01 June 2023, 07:49 AM.
- Likes 1
Comment
-
Originally posted by scottishduck View PostAs far as competitors building their own accelerators. That would require them to get access to advanced fab nodes and somehow not infringe on nvidia IP, good luck with that. At the moment no one else even comes close to what nvidia is doing in this field, they’re basically writing the standards themselves.
Nvidia is not alone in this space. Google and AWS are more focused on being able to deploy AI in more areas and be highly power effective. Nvidia really does not make the small highly power effective AI chips in any volume.
Hanguang from alibaba is focused on being Nvidia competitor why because they may not be able to import Nvidia latest and greatest AI chips. Alibaba china based IP infringement is not problem because China government has marked AI development as critical by any means including IP stealing. This is only tip of iceberg by the way.
Nvidia is not writing the standard by themselves. AI we have different competing standards between google, aws and alibaba and others that are not Nvidia.
Originally posted by mirmirmir View PostAs much as I hate novideo, you gotta admit, they made first GPU and succeed, they made GPU compute and succeed, they made ray tracing a thing and succeed, they've been maintaining their position on top for a while now.
Originally posted by MorrisS. View PostLinux end-users could prefer proprietary drivers to open source drivers, so this free choice should be allowed.
The first open-source release of GPU kernel modules for the Linux community helps improve NVIDIA GPU driver quality and security.
Basically Nvidia on newer cards prebuilt kernel module vs user/distribution built kernel module these days. Distribution built kernel module parties like Redhat can sign for secure boot. Closed source blob in kernel space CPU does not cut it any more. Nvidia is moving their trade secrets into firmware some of what Nvidia use to run in the CPU now runs in gpu card risv-v processor.
AMD when they open source the ATI driver did the same kind of thing.
Proprietary closed source OS driver lot of ways does not make sense due to the fact OS always need to update and be changed for security and performance reasons and the OS maker is required to sign these drivers by secureboot rules set down by Microsoft to keep Linux boot loaders signed by Microsoft. Proprietary firmware makes more sense and this can be protected and enforce by signing keys that Nvidia controls.
The big thing that has broken Nvidia closed source module under Linux is secureboot and the fact Redhat and other Linux distributions will not sign a binary blob OS drivers and will not develop validation systems for binary blobs for OS drivers instead want to see source code. Particularly thinking AMD and Intel does not have problems with this limitation it was why should Nvidia get special treatment.
Yes Nvidia customers what Linux systems to boot with secure boot on not having to turn it off just because they have Nvidia hardware to get performance.
The time of closed source OS kernel space drivers with Linux is coming to the end cause is Microsoft and secureboot and parties like Redhat being strict.
Remember any flaw in anything Redhat signs under secureboot rules Microsoft set down they will be 100 percent held accountable for. Nvidia giving Redhat a blob to run in kernel space and saying trust be bro sign it just does not fly. Nvidia got away with this basically for 2 decades it over now. This is why the change to firmware and added cpu to gpu for trade-secret stuff. Yes the reason why its is a point that its these Nvidia cards and newer is because older GPU don't have cpu/microcontroller on the GPU for running special stuff.
Yes it a problem with Nvidia firmware that there is not a dependable way to get firmware version loaded at this stage that will have to change as that will not be suitable going forwards either.
Originally posted by MorrisS. View PostIt is interesting to understand what the impact of this decision is about Nuveau and Nvidia cards management. What are the benefits? What about the adaptive power management mode?
Nuveau will need to pick particular versions of Nvidia firmware and ship them until Nvidia gets around to making stable API/ABI for the firmwares their cards use(not a simple process).
Last edited by oiaohm; 01 June 2023, 12:26 PM.
Comment
-
Originally posted by oiaohm View Post
No point talking that way. Nvidia proprietary kernel based driver has already ceased to exist for the newer cards.
The first open-source release of GPU kernel modules for the Linux community helps improve NVIDIA GPU driver quality and security.
Basically Nvidia on newer cards prebuilt kernel module vs user/distribution built kernel module these days. Distribution built kernel module parties like Redhat can sign for secure boot. Closed source blob in kernel space CPU does not cut it any more. Nvidia is moving their trade secrets into firmware some of what Nvidia use to run in the CPU now runs in gpu card risv-v processor.
AMD when they open source the ATI driver did the same kind of thing.
Proprietary closed source OS driver lot of ways does not make sense due to the fact OS always need to update and be changed for security and performance reasons and the OS maker is required to sign these drivers by secureboot rules set down by Microsoft to keep Linux boot loaders signed by Microsoft. Proprietary firmware makes more sense and this can be protected and enforce by signing keys that Nvidia controls.
The big thing that has broken Nvidia closed source module under Linux is secureboot and the fact Redhat and other Linux distributions will not sign a binary blob OS drivers and will not develop validation systems for binary blobs for OS drivers instead want to see source code. Particularly thinking AMD and Intel does not have problems with this limitation it was why should Nvidia get special treatment.
Yes Nvidia customers what Linux systems to boot with secure boot on not having to turn it off just because they have Nvidia hardware to get performance.
The time of closed source OS kernel space drivers with Linux is coming to the end cause is Microsoft and secureboot and parties like Redhat being strict.
Remember any flaw in anything Redhat signs under secureboot rules Microsoft set down they will be 100 percent held accountable for. Nvidia giving Redhat a blob to run in kernel space and saying trust be bro sign it just does not fly. Nvidia got away with this basically for 2 decades it over now. This is why the change to firmware and added cpu to gpu for trade-secret stuff. Yes the reason why its is a point that its these Nvidia cards and newer is because older GPU don't have cpu/microcontroller on the GPU for running special stuff.
Yes it a problem with Nvidia firmware that there is not a dependable way to get firmware version loaded at this stage that will have to change as that will not be suitable going forwards either.
GSP firmware binary license correction is only the first step. Next problem is no stable ABI/API between GSP and the driver. There is no API/ABI like AMD/Intel has to say I provide X list of functionality this is what allows older AMD/Intel driver to use newer firmware than what it was built with. Not all GSP versions even have a function to report version number of GSP loaded into card.
Nuveau will need to pick particular versions of Nvidia firmware and ship them until Nvidia gets around to making stable API/ABI for the firmwares their cards use(not a simple process).
Actual problem deals with the possibility mesa drivers allows the users to apply the benefits of hardware acceleration otherwise the GPU itself becomes completely useless. Moving from Nvidia proprietary drivers to MESA drivers causes the end-users to lose many benefits in some browsers, as example chromium or chrome in terms of video acceleration and video decode.Last edited by MorrisS.; 01 June 2023, 01:43 PM.
Comment
Comment