AMD 3D V-Cache Optimizer Driver To Be Merged For Linux 6.13

[sfjuocekr]

It would be more beneficial to have more fine grained control.

Have the whole system run on the frequency policy, but specific applications that the user selects use the cache policy instead?

Also, applications could probably be smarter about their behavior as well if this was exposed in some way.

[smitty3268]

To me it was a sign that they needed money so they implemented a strategy where people who used AMD were forced into buying more than one AMD GPU. The long term effect of that is that nowadays everything has CUDA support and ROCm is an afterthought if its even a thought at all.

CUDA was the only thing that mattered for compute before rdna existed, and it's still the only thing that matters today. I don't see anything AMD has done with their hardware which made an impact either way on that. CUDA simply got a big head start, and ROCm barely works, and that's all based on software not hardware.

[skeevy420]

Mainly in the US, here in France you can unionize pretty easily

I envy y'all. I wish my people had your peoples' courage and bravery.

[skeevy420]

If you care about professional uses, then obviously going with a card designed primarily for professional use-cases is the best option. Most consumers don't care about CUDA though.

The promise AMD made was that by designing rdna specifically for consumer use, it would be cheaper and better at that than by just getting a cut down pro card. Obviously that has failed, which I think is why they are killing off the line and falling back to their old strategy again.

Anyway, I'm not saying that's a terrible thing. It will certainly simplify creating the hardware for AMD, and maybe that will end up being better all around anyway, but it was at the very least a sign to me that they were wanting to cut down expenses on the gpu consumer side. These layoffs just kind of confirm that for me.

To me it was a sign that they needed money so they implemented a strategy where people who used AMD were forced into buying more than one AMD GPU. The long term effect of that is that nowadays everything has CUDA support and ROCm is an afterthought if its even a thought at all.

While most consumers don't care about CUDA, enough of them do which is why it is so prevalent and easy-to-use on Linux. ROCm has been a nightmare.

Most consumers don't care about Linux. There are probably more CUDA users on Windows than there are people running Linux desktops.

Most consumers don't know what a tariff is. It didn't stop them for voting for them. What's a tariff? What does it mean to be a member of the E.U.? Did Biden drop out? We might be individually smart here and there, but we're collective morons.

Most consumers

We are the Borg. Prepare to be assimi....

*watches 2 girls 1 cup*

Locate the nearest transwarp conduit.

[avis]

Community Service Bulletin:

Upvote this post if you'd like to downvote Avis.

Check the mindful reply that got 11 upvotes: https://www.phoronix.com/forums/foru...02#post1505802

Maybe you could stop posting BS and actually ignore the replies that are explicitly marked as offtopic.

"Community service", FFS. Please spare me.

[mitch971]

combined with a lot of the world not having very strong labor laws helps to prevent unionization efforts

Mainly in the US, here in France you can unionize pretty easily

[smitty3268]

smitty3268 Moving home and professional users to the same platform will help make AMD more competitive with NVIDIA. You buy NVIDIA and you have CUDA and all the current features available on whatever GPU you buy.

If you care about professional uses, then obviously going with a card designed primarily for professional use-cases is the best option. Most consumers don't care about CUDA though.

The promise AMD made was that by designing rdna specifically for consumer use, it would be cheaper and better at that than by just getting a cut down pro card. Obviously that has failed, which I think is why they are killing off the line and falling back to their old strategy again.

Anyway, I'm not saying that's a terrible thing. It will certainly simplify creating the hardware for AMD, and maybe that will end up being better all around anyway, but it was at the very least a sign to me that they were wanting to cut down expenses on the gpu consumer side. These layoffs just kind of confirm that for me.

[Type44Q]

</offtopic>

Community Service Bulletin:

Upvote this post if you'd like to downvote Avis.

[skeevy420]

Is the consumer lineup really going back to cdna? So fp64 will be good again? That would be nice.

UDNA

[skeevy420]

I don't know, man. A part of the 4% layoffs are resources in AMD Link. Considering AMD is a fraction of Nvidia's market share, and a fraction of AMD graphics customers use AMD Link, they've finally realized that it's not worth burning money to keep this feature afloat. This thing has to support iOS phones/tablets, Android phones/tablets, Apple TVs, and Android TVs. There's security, marketing, and there's an aspect of how these integrate with Adrenaline and the pile of bug tickets it brings. So, for a fraction of a fraction of their shrinking market share in discrete GPUs, the cost just doesn't add up and are better served instead by third party vendors.

Also, the stock market in regards to AMD is very irregular and does not make sense... And while the hiring season align with the industries due to grads availability, Lisa ain't gonna wait for her projects to get started for the sake of stock manipulation lol.

Well, it actually IS good business to do any necessary layoffs during the yearly purge. As a Linux user I forgot that Link even existed. I forget a lot of the features my GPU should actually be able to do since Linux doesn't support them. Outside of the actual AMDGPU driver in the kernel, AMD's software support is atrocious.

And regarding stock prices, practically every year that AMD doesn't have a flagship product release there's a dip in the price that starts around August to November and ends between November and February.

smitty3268 Moving home and professional users to the same platform will help make AMD more competitive with NVIDIA. You buy NVIDIA and you have CUDA and all the current features available on whatever GPU you buy. You buy AMD and you have to check with a matrix consisting of GPU models and AMD's Features to see what GPU actually supports what. Then you finally find your GPU only to see that it's one of the models with an asterisk where it says some shit like "We support GFX_1234 but not in your model of GPU. It should work but your results may vary and we won't help you if you have problems." Moving everything to the same platform should help alleviate that problem of needing a feature matrix.

AMD did that as a way to entice people to buy more than one GPU. The problem is that means a professional home user would need one Pro GPU at work, one Pro GPU at home, and one Consumer GPU at home. Ain't nobody wanna buy three GPUs.

Instead, people would rather just buy one Consumer NVIDIA GPU at home and a Pro NVIDIA GPU at work since the Consumer NVIDIA GPU supports all the features of the Pro GPU, it's just not as fast, but that's good enough for prototyping and trying things out as well as they don't have to deal with multiple GPUs in the same system or have to have one system per GPU. I really fucking hope that's what AMD's endgame is.

I hope they give up trying to compete with NVIDIA at that level. At some point it's like watching Lamborghini and Ferrari. Cool. Many fast. So much win. Anyways, I can't afford that, where's my Honda Accord GPU?