No, AMD Will Not Be Opening Up Its Firmware/Microcode

[starshipeleven]

As far as I'm aware a clean room wall is still legal.
https://en.wikipedia.org/wiki/Clean_room_design

Clean room design has some chances of working if the stuff you are reverse-engineering isn't a ridicolously complex affair like a modern GPU.

There is a reason why most examples of it are from the nineties or older.

[starshipeleven]

The microcode is not only active at initialization time but also at runtime from what I read:
https://www.semiaccurate.com/forums/...585#post269585

That's what I said with "set how they should work at low level". The microcode tells the hardware components how to operate at all, as the complexity of modern systems does not allow to design hardware that is fully ASIC.

Well, what are the tasks of an operating system?
IPC
Memory management
Scheduling
Networking
Storage

Wrong, any microcontroller running a bare-metal blob does the same.
What makes an OS is the ability to be a host of processes, offering APIs that abstract all the operations above so that a process does not need to give a fuck about how to allocate memory or operate a NIC.

An OS is still an OS even without networking and storage (it is a RAM-only OS, loaded in RAM by the bootloader).

Anyway, as bridgman has noted, storage isn't managed by the GPU but by the driver, though I have a hard time imagining how this is going to happen without any supporting code or logic for that on the GPU itself. The considerable speed gains that were demonstrated by AMD supposedly come from not having to move the data via the host, but directly between GPU and SSD.

newsflash, it's called DMA, allowing devices to have direct access to RAM and other system resources on pcie bus on their own without asking permission to CPU (and OS) each time they need it.

https://en.wikipedia.org/wiki/Direct_memory_access

High-performance networking features also RDMA (remote DMA), which is allowing remote hardware access to the same stuff.

And yes, this isn't fun and games for a security standpoint, which is why a long time ago they invented IOMMU or vt-d features, that sandbox this so that a device cannot just go and hack BIOS for lulz.

And besides, I think your comment shows that you failed to fully grasp the FSF position on the matter.

FSF position on the matter is not realistic. Only way you can drop these proprietary blobs is if you have fully open hardware, and THAT is the thing you should lobby for, not harassing companies that make closed hardware to release as open stuff they cannot possibly release.

[artivision]

First of all microcode and firmware are two totally different things. I don't understand what we actually need the second. I also don't understand why we must load the first and why is so big and needs to exist outside cache.

[starshipeleven]

he believes that we the FLOSS community can not do this but in fact he will be wrong.

I'm with him. FLOSS community produces a lot of hot air, but actually making something that is practical? Nope.

For example, there is RISC-V which should allow true FLOSS hardware. Is FSF or even Stallman doing anything about it spreading the good news and campaigning for its usage/adoption? Nope. Only hot air and whines at the usual suspects for the usual stuff.

[starshipeleven]

Bridgman does not believe me wen i tell him that in the end we will have full opensource hardware down to the firmware and if it is not AMD hardware then we will have a solution WITHOUT AMD...

FYI: AMD is also a platinum or gold founder member of RISC-V foundation.

[MaxToTheMax]

I'm looking forward to the day semiconductors STOP getting faster. I think this will affect the economics of computing in some positive ways:

• No more upgrade treadmill
• Open hardware designs are given time to catch up to the proprietary ones, especially as patents begin to expire
• Drastically reduced prices as companies have to compete on price instead of performance
• Second-hand electronics become viable options just like second-hand cars
• Devices like lapops and phones are built to be more durable, reducing e-waste
• The gradual end of the digital divide as everyone has basically the same computing power as everyone else
• As hardware stops evolving, the gradual end of proprietary device drivers and binary blobs, since it's now possible to fully reverse-engineer a design without it becoming obsolete

I'd still like computers to get a *little* faster before they stop, though.

[starshipeleven]

No more upgrade treadmill

They will do that with features or power consumption, or even simple software support (which is 100% artificial). In many areas (cough-Android-iOS-embedded-cough) it's already like that.

Open hardware designs are given time to catch up to the proprietary ones, especially as patents begin to expire

More like proprietary ones stagnate or disappear outright as no company finds profitable keeping them in production.

Drastically reduced prices as companies have to compete on price instead of performance

If you are thinking companies are selling stuff at a loss, think again.

Second-hand electronics become viable options just like second-hand cars

Already true for most stuff.

Devices like lapops and phones are built to be more durable, reducing e-waste

Total bullcrap, the devices are built to break and locked to a specific firmware version just to introduce a concept called "planned obsolescence", that is obsolescence of hardware that is still technically fine.

The gradual end of the digital divide as everyone has basically the same computing power as everyone else

Communist utopia, you should be punished for thinking this is even possible giving human behaviour.

As hardware stops evolving, the gradual end of proprietary device drivers and binary blobs, since it's now possible to fully reverse-engineer a design without it becoming obsolete

Sure, you can reverse-engineer stuff that isn't produced anymore as the company has long since died or retooled for something else entirely.

[pal666]

First of all microcode and firmware are two totally different things. I don't understand what we actually need the second. I also don't understand why we must load the first and why is so big and needs to exist outside cache.

feel free to develop your own gpu without those downsides

[karolherbst]

[*]Drastically reduced prices as companies have to compete on price instead of performance[/list]

This is your main mistake why your assumption is just wrong: If prices fall a lot, then performance/price also falls, just like today. If you can't scale vertically (higher freqs) you simply scale horizontally (more cores), so how does this help with anything you just said?

To sum it up: if CPUs drop half in price, you can get double performance for the same price. It will only stop if the price stay _constant_, and this will only happen if your research really doesn't find anything usefull, which is highly unlikely.

[pal666]

I'm looking forward to the day semiconductors STOP getting faster

you don't need to wait. gift your property to church and go live in cave