Vega 12 Support Now Queued In DRM-Next For Linux 4.17

[agd5f]

You do realize that Miners buying AMD GPUs for double the planned prize results in AMD getting a lot more money for their investments, resulting in more money to AMD,

AFAIK, AMD doesn't make any more money when retailers charge more. AMD still has the same MSRPs.

[dungeon]

People seems does not realise how markets vary, prices vary, performance vary, power consumption vary, benchmarks vary It vary, depends how you or someone else for you set it up, how you use WattMan let say, etc... You can only say - theoretical TDP is that, theoretical GFLOPS spike is this, etc... but it is all theoretical depends what you do, even marketing names are theoretical as one chip rev might appear in various form factors, etc...

Prices are not the same in US and in let say Bangladesh If some retailer charge more, that only usually describe current and that market condition. If you look at some parts of Europe prices for GPUs are nearly the same as before, but they were always higher than in US for example

When i read articles like this for example: Cryptocurrency Miners Bought $776M in GPUs in 2017, Mostly From AMD


I always asking myself, how in the hell they counted all that together Does miners always declare themselfs as: "Hey - i am a miner, not a a gamer" or so

[coder]

It would we weird to me only if we have released kernel with support for these chips, but these are not yet on the market. If some product containing these chips appears up to time of 4.17 kernel release, everything is considered fine

I think the idea is ideally to have support in a released kernel by the time the product launches. For a popular distro like Ubuntu, that would mean they should've gotten the support into 4.15, if possible, even though the product might not launch for months after that release is out the door. That hasn't been the case with their recent products, but only because AMD has been playing catch-up.

Just look at Intel. If you search the recent news-feed on this site, you'll literally see them committing support for products in the generation after next, just to make sure that when it finally launches, the software support is already out there and ready for it.

[coder]

Thanks AMD for not increasing production...

You act like they're 100% in control of all components and production capacity.

https://www.anandtech.com/show/12380/amd-to-ramp-up-gpu-production-ram-a-limiting-factor

[coder]

... results in AMD getting a lot more money for their investments, resulting in more money to AMD, ...

AMD and Nvidia aren't reaping most of the windfall, here. They have to deliver chips at contracted prices to their board partners. It's mostly the board makers and resellers that are getting the markups.

That said, it's good for AMD if they can sell every GPU they can make. However, given the above link about GDDR5 and HBM2 supply shortage, that's not even a given.

[dungeon]

Just look at Intel. If you search the recent news-feed on this site, you'll literally see them committing support for products in the generation after next, just to make sure that when it finally launches, the software support is already out there and ready for it.

I would just rather do not look at Intel at all, as Intel is not example to me at all. And why? Intel only do iGPUs and only in combo with their CPUs - that Lintendo sounds very easy-peasy, isn't it

AMD dGPUs works with CPUs even on ARM, Power, Mips, AMD, Intel CPUs (even works on these Russian CPUs )... AMD doing support for dGPUs, but also iGPUs in APUs, only with that info it is easy to realise who have more job to do. Do you even realise how dGPUs are much much more harder to support? Just look at the size of AMD driver, just count all asics they support, etc...

Intel would need to have 300 developers, just to support all that about AMD GPUs

[coder]

I would just rather do not look at Intel at all, as Intel is not example to me at all.

I'm only talking about the concept of software support preceding hardware availability. If you expect to have hardware on shelves before or coincident with the support being committed to the kernel, your position seems difficult for any company to defend for whom Linux factors significantly into their business strategy.

The software support needs to be out there, tested, and already deployed in mainstream products (i.e. major distro releases) by the time hardware reaches the shelves.

[dungeon]

AMD can do that if you wanna fund AMD to have at least 3 times more than Intel driver developers... because amount of job needed is much bigger, but amount of devs are less

They do the best they could, with a help of internal team. Should we note how this amdgpu driver, would likely never happened without that help

The software support needs to be out there, tested, and already deployed in mainstream products (i.e. major distro releases) by the time hardware reaches the shelves.

That sounds good, so just fund AMD to have 10 times more devs for linux driver and support would be just like on Windows, with all bells and whistles and one stop shop AMD would also squeeze down evereyhing to support only their own CPUs and that would be it, then they should stop producing dGPUs and only then we can compare that to Intel driver And of course do not complain if something is blobby, because bells and whistles are usually that on the market

BTW, which one AMD product to hit shelves - iGPU, mGPU, dGPU, PX, CFX, pro, embedded, etc... which one of these is your desired combo, with day one all working all profiled with released and opensource driver.

Joke aside, if AMD do only iGPUs for themselfs like Intel, you would already see it like this as that is easy peasy to support considering what all else needs to be supported

And nope we can't be like on Windows, to quote this for 18.3.4 driver:

A random system hang may be experienced after extended periods of use on system configurations using 12 GPU's for compute workloads.

https://support.amd.com/en-us/kb-articles/Pages/Radeon-Software-Adrenalin-Edition-18.3.4-Release-Notes.aspx

You see what these crypto gamers are doing there, that is their Vega 12

[bridgman]

What do you mean, more specifically? Which 'piece', and what 'overall plan'? And what is 'this' that should become the new normal? Drivers merged before public announcements, or are you talking about something else?

I was replying to the previous comment about how there was a lot of news (because of various parts of driver support appearing) before anything was known about the product...

I find it a little bit funny that we get so much news about a mostly unknown GPU. I'm not complaining, it's just weird.

... saying that publishing driver support before launch (in order to make support more readily available when the product is eventually launched) was hopefully going to become more common and hence less newsworthy over time. So yes, drivers merged before public announcements.

[lowflyer]

to merge big patch-sets so close to the next cycle otherwise he risks upsetting Linus Torvalds.

The real problem is to end up with broken code. Upsetting Linux Torvalds is not really the problem here.