Tiny Corp Changes Course Yet Again With Plans To Offer AMD Radeon GPUs

[qarium]

LOL about the poeple complaining they should use instinct or w7900 and pay 3-5 times more for little real benefit, because of course you are not supposed to use hardware to it's full capacity, you should pay extra to get the rebranded, certified part that most certainly works (or does not) work better.

what a nonsense no one here said you should get a rebranded certified part... and this was also not the reason why it is usefull.

its not even about ECC ram... hell its only about 24GB vram vs 48GB vram.

the PRO W7900 is not a rebranded 7900XTX because the 7900XTX has only 24GB vram.

also did you ever check any instinct ? these instinct cards have much more vram than 24gb... mi100 has 32gb vram mi200 has 64 mi 300 has 96gb and so one...

[Quackdoc]

While the hardware sounds interesting and reasonably priced, the bipolar nature of TinyCorp makes me unlikely to recommend to my company that we buy this.

It's a shame that being open about issues and development process is being considered "bipolar" and not "open"

[ET3D]

It's a shame that being open about issues and development process is being considered "bipolar" and not "open"

Saying "going to work with AMD", "not going to work with AMD", "going to work with AMD", ... many times isn't "being open". It's a shame that you're calling it "being open", as it misses the entire reason this person is untrustworthy.

[NeoMorpheus]

As expected, the Ngreedia white knights are out in full force trashing AMD. Out of all places, its a shame such behavior.

This dude is trying to use the cards in a way that they weren't designed and tested for and is surprised and angry that his consumer cards aren't running his professional workload as well as he would like or that he isn't getting bespoke professional level support on consumer level equipment.
Almost like thats why professional products with better support exist, or heck, why AMD's entire semi-custom group exists.

I failed to see which Ngreedia gpus he is using, but if they are RTX, you can count on the lawyers sending some nice reminders of their eula.

[smitty3268]

It appears my info is outdated (it's been a long time since I've looked into it). I swear the last time I looked it up, ROCm on Linux required the proprietary user-space AMD driver.

It's long been open source, but it's the kind of open source with a crappy poorly documented build system, tons of non-upstream vendored dependencies (like customized llvm), etc. Meaning for all intents and purposes, you installed a prebuilt version of it via AMD's installers, and not by building it yourself or getting if as part of your distro. I believe recently some distros have started to try to provide it themselves and things have somewhat improved.

[chithanh]

their kernel driver is open source, and I believe the SDK for rocm is too, which I thought directly talks to that, unless there is a step missing im not sure what is closed source in the stack, aside from the bits mentioned.

The amdgpu kernel driver and ROCm user space stack are entirely free/libre and open source. Only the GPU firmware is proprietary, which is what tinycorp wanted AMD to open up.

That is not true. Many parts of the AMD GPU driver stack especially when it comes to compute are closed and even non-public.

Example are the GIM drivers for recent MxGPU cards. Only public Linux driver is 1.05 for old FirePro cards.

If you ever saw a Google Stadia Workstation Development Node in action (Lenovo produced a few before Google pulled the plug), those were loaded front to back with proprietary and non-public driver code from AMD.

My understanding is that ROCm should eventually replace all this but AMD is not fully there yet. Probably every BC-250 user can testify how horrible the driver situation is.

And that is even before talking about further AMD semicustom stuff like the Android drivers for Xclipse (Exynos 2200 etc.).

They are changing directions faster than a hunted rabbit.

Saying "going to work with AMD", "not going to work with AMD", "going to work with AMD", ... many times

I swear some people can't read nor understand probabilities and conditional sentences.

If the company says 70% confidence or there is a roadblock, that doesn't mean they changed direction when the 30% materializes or the roadblock is cleared. This happens all the time in less transparent companies too, just they don't talk but rather say they have nothing to announce.

[geerge]

Saying "going to work with AMD", "not going to work with AMD", "going to work with AMD", ... many times isn't "being open". It's a shame that you're calling it "being open", as it misses the entire reason this person is untrustworthy.

"Want to but need debug support"

"Debug support confirmed unavailable so can't"

"Actually debug support is available so can"

Tinycorp appears to be a developer led company, not a marketing led company. So you get openness and honesty about the process they're going through. Which you call untrustworthy. Lol. Don't get me wrong I understand that lack of marketing does make them appear less trustworthy to normal folk who want an appliance to do a thing, but it has the opposite effect on tech nerds, and it's the nerds who have been paying attention.

[ET3D]

I swear some people can't read nor understand probabilities and conditional sentences.

Does it matter if he keeps changing from "we probably won't work with AMD" to "we will work with AMD"?

It's like I'd narrate my software development, every attempt that doesn't work, and keep giving percentages to whether this will work in the end, while I keep being pessimistic about it. It's not something that inspires confidence. That and saying things like "They should immediately stop development of high end ML libraries and fix their basic shit."

I'm sure that some people are okay with how he comes across.

it has the opposite effect on tech nerds, and it's the nerds who have been paying attention.

It matters less who is paying attention and more who is paying money.

[varikonniemi]

what a nonsense no one here said you should get a rebranded certified part... and this was also not the reason why it is usefull.

its not even about ECC ram... hell its only about 24GB vram vs 48GB vram.

the PRO W7900 is not a rebranded 7900XTX because the 7900XTX has only 24GB vram.

also did you ever check any instinct ? these instinct cards have much more vram than 24gb... mi100 has 32gb vram mi200 has 64 mi 300 has 96gb and so one...

usually that would be called 7900XTX 48gb

not rebranded to something completely else

and mi series is something completely else and not comparable, meant for whole different workload and design.

anyhow you missed the point completely and stuck in the hyperbole. The point was that if you use the product according to the advertised specs, you must be special to say "it's not meant for that"

[mdedetrich]

I don't question about reliability, but my best guess is that you're talking about the hardware and I guess AMD hardware is as good as nvidia. Moreover, in the one of the last posts they were asking for AMD to opensource parts of the firmware to help them debug, not kernel or user space drivers.

Actually hes talking about the firmware (which is closed source), not the hardware. This is where the real sauce is, and NVidia's firmware is known to be rock solid compared to AMD's in this specific area (i.e. compute).

Having good hardware doesn't mean anything if the firmware also isn't good, and this is where most the complexity and value comes from.