> Our mission:
> • Build a multi-architecture multi-vendor software ecosystem for all accelerators.
> • Unify the heterogeneous compute ecosystem around open standards.
> • Build on and expand open-source projects for accelerated computing.
If AMD got their act together they would go all in on open standards together with other actors in the segment, but maybe they really want to see if their rocm can repeat Cuda's success and lock in? I just want accelerators to become commodities and interchangeable like CPUs, let me write my code once and deploy on whatever hardware the user have, preferably without having to compile multiple versions.
There’s a lot of stuff Google could have spun off but they seem content to ignore things until they turn out the lights.
Like when he strolled into Twitter, acted like he knew better than everyone, delivered nothing and then left.
There is far more to achieving things in large companies than simply having great technical skills.
You weren't listening to the GP comment.
He's expected to succeed while being drowned in the swamp.
That's what most people mean and expect when they bring in an outsider.
"Fix our problems, but without changing anything to do with management."
Disclosure: I'm working on exactly that, so thank you so much for this!
I'm doing it on the high end of the spectrum, with MI300x, and a lot less drama.
Their entire stack on Linux is open-source. That part isn't a problem.
More concretely, what they open sourced calls closed source functions that run on the GPU which is where all the bugs are, and causes all the crashes.
In particular, what they did was open source the drivers but not the firmware, and then when the firmware has bugs nobody outside the company can fix it because it's not open source.
After all, your CPU almost certainly runs firmware with all the same issues. Not to mention all other GPUs from all other vendors.
Despite all of the noise-making, I don't think anybody has really convincingly made the case that the firmware quality is the main issue here, as opposed to the quality of the rest of the stack.
CPUs don't really have firmware in the same sense. CPUs expose basic primitives (machine instructions). Modern CPUs will translate them to micro-ops in microcode instead of executing them directly, but the mapping is relatively simple and bugs are correspondingly uncommon.
The GPU firmware is doing something complicated, and therefore buggy, but you're required to use their complicated buggy closed source code because it also doesn't directly expose simpler primitives you could use to build your own alternative to it.
> Despite all of the noise-making, I don't think anybody has really convincingly made the case that the firmware quality is the main issue here, as opposed to the quality of the rest of the stack.
The rest of the stack should be an abstraction layer that makes programmers not have to care what kind of hardware is under them, the same as they generally don't have to care if their CPU is from AMD or Samsung. But then you need the code to translate from that abstraction to the hardware, and that code needs to work. Which can either be accomplished by the vendor providing working code, or by the vendor providing sufficient documentation and source code for someone else to write working code. Providing neither of these is not effective.
What AMD should be doing at this point is both. Open source the existing firmware so that people who are trying to use their hardware right now have the ability to fix any problems they encounter themselves, while the company uses the money they now have to address their existing technical debt, which nobody reasonably expects to happen overnight. But will happen faster when it's not only the company but also the users working on solving the problem.
How would you classify e.g. the Intel Management Engine and the firmware that runs on it?
This is the most damning commentary on AMD's firmware code quality.
I mean, that's horrifically bad.
To think that is code for the flagship GPU for a company with a market cap of almost 300bn is just crazy.
and intel has a solution for this too, in fact…
I'm hoping Google spins off its TPU division and sells chips to everyone that wants them.
I'm hoping AMD gets their act together and that they open up their drivers and stack.
I'm hoping a lot of other hardware companies and compute stacks start making inroads on this.
Nvidia is not just bilking the whole world on margin, but they're also a limiting reagent in the grand scheme of progress.