Slashdot Mirror
NVIDIA Begins Requiring Signed GPU Firmware Images
An anonymous reader writes: In a blow to those working on open-source drivers, soft-mods for enhancing graphics cards, and the Chinese knock-offs of graphics cards, NVIDIA has begun signing and validating GPU firmware images. With the latest-generation Maxwell GPUs, not all engine functionality is being exposed unless the hardware detects the firmware image was signed by NVIDIA. This is a setback to the open-source Nouveau Linux graphics driver but they're working towards a solution where NVIDIA can provide signed, closed-source firmware images to the driver project for redistribution. Initially the lack of a signed firmware image will prevent some thermal-related bits from being programmed but with future hardware the list of requirements is expected to rise.
Surely it is impossible to have an opensource software if it needs a key to build it into a runnable program?
I mean you have the binary but you cannot recreate it from the source without that key to sign it with. The key is part of the source and you don't have it.
I don't think you understand what these fake cards are.
They are legitimate $80 nvidia cards, flashed with modded firmware to enable cores and clock speeds that the silicon is incapable of handling (Hence why the chip became an $80 card in the first place, instead of a $300 card).
The modder then puts the flashed $80 card on e-bay for $200 and makes a sweet profit.
Quite often they're not even unlocking anything. Rather they're doing a dirty hack to change the bios information of the card to display something that it isn't. This isn't all that unfamiliar to those of us who were in the industry back in the mid to late 90's when scammers were resilking(cpu info used to be silk screened on, to counter this it's why all cpu's are now stamped) Cyrix cpu's as AMD and Intel. You only found out what the CPU actually was, when you plugged it into the board and it said "cyrix." And while there are cases of people doing this to binned parts, most of the time the links to enable those pathways are cut before they're made into a gpu to stop people from doing exactly that. And if you're wondering why, it's because Intel ran into a massive problem where fly-by-night companies would unlock the binned CPU, and then actually flashing the microcode to change what the CPU was.
The cheap and dirty way to unlock CPU's during that time period was to use a graphite pencil across a unfinished path. I think it was pin 14 or 23 on the board. Very nasty problems with Slot 1 cpus.
Om, nomnomnom...
You don't understand. All companies, AMD and Nvidia manufacture certain chips. Each chip has certain failure rates. When certain amount of cores fails, they are switched off in software and sold for less.
AMD does this. Nvidia does this. Pretty much everyone making complex chips does this. It's massively uneconomical to throw away an entire chip over partial failures.
Andy Ritger at Nvidia is already in talks with Ben Skeggs and Martin Peres with Nouveau. They're are going to hash out the details at XDC2014. The impact for Nouveau is in the packaging and distribution parts of the cycle, not development. Also, it was Nvidia who reached out to Nouveau, not the other way around. Nvidia has their reasons for doing this, but it's not an anti FOSS thing. It's more likely one of the more sane reasons posted above.
So everyone just relax their sphincters a bit....
You aren't understanding. Since it was explained fairly clearly, I'd guess you don't want to understand. But I'll try again anyway.
These chips are broken. So they are sold cheap. You don't want to pay full price for seconds. Before they sell them, they use software to set the broken parts as not working. Some of them aren't broken enough that you'll immediately notice, but that doesn't mean they aren't broken.
Usually the breaks are only in one area. Some die didn't burn properly, or traces weren't properly laid down. Whatever. So that area is sealed off. The manufacturer doesn't do a detailed investigation of exactly what's broken, just one that's good enough so they can figure out what needs to be sealed off to have a working chip. Then the sell the working chip (with reduced functionality) for a much cheaper price.
So if you don't need the full functions of the chip, you can buy the cheaper, reduced functionality, model at a cheaper price.
IC manufacturers have been doing this since the i8086, or maybe the i80186. (Intel was the first one I ever heard of doing it.)
This is a deal for those who don't need the functionality of the full model. It also cuts the prices for those that do, as selling the seconds defrays some of the cost of manufacturing.
Those who are removing the imposed limits and selling the seconds as if they were first quality are the ones who are cheating the customers. They are also impugning the name of the original manufacturer.
I think we've pushed this "anyone can grow up to be president" thing too far.
all of them, this is how binning works.
https://en.wikipedia.org/wiki/Product_binning
Chips are designed for the max freq of the specification. If they fail that spec, they are retested at a lower spec, and if they pass that spec, they are sold at that frequency. Why else do you find many diffrent chips in the same family run at diffrent speeds?
Many times the chip is %100 capable of running at faster speeds, but they had too much of the higher bin, and not enough of the lower bin.
But yes, taking a chip that didn't pass a higher speed, flashing it to the firmware of its faster/more capable cousin, and then selling it as such is ripping people off.