The State Of Open-Source Radeon Driver Features

[log0]

The GPU business is *very* competitive, and small differences in performance & features drive many of the buying decisions. The cost of driver development is the primary entry barrier for new competitors. Why would an established vendor give away their competitive advantage ?

There is a standard client lib already (mesa) -- are you sure you want to see it replaced with something 10x the size and optimized for one specific HW vendor ?

Interesting statement. You have got a working blob already. What is the point of supporting the open source driver then? It certainly can not be in your interest if "new competitors" use it for their advantage.

[archibald]

Interesting statement. You have got a working blob already. What is the point of supporting the open source driver then? It certainly can not be in your interest if "new competitors" use it for their advantage.

Because their customers/partners have asked for it.

[log0]

Because their customers/partners have asked for it.

And this customers/partners don't care about power management, performance and overall feature completeness? Who are this customers btw(I assume this info is not secret)? What are they doing with this open source driver?

[chrisr]

Am I the only who is HAPPY with the Mesa/Gallium/ATI stack?

Compared to fglrx, Mesa/Gallium are pure Heaven. With fglrx, I seem to remember that "fast 3D" was at the expense of stability, reliability, consistency and "gouge-your-eyes-out-slow" 2D. Ugh!

[archibald]

And this customers/partners don't care about power management, performance and overall feature completeness? Who are this customers btw(I assume this info is not secret)? What are they doing with this open source driver?

I should have clarified: I was reporting on what Mr. Bridgman has said in the past, I don't have inside knowledge of who these partners are/what they are doing.

[bridgman]

And this customers/partners don't care about power management, performance and overall feature completeness? Who are this customers btw(I assume this info is not secret)? What are they doing with this open source driver?

Other than power management, which was a whole lot simpler when we kicked this off back in 2007, I imagine they're pretty pleased with the features and performance. Launch-time support (buy new HW, install a recent distro, use the system) was a higher priority than features and performance.

The common thread among the customers was that (a) they were building big compute farms with our CPUs, (b) they were running Linux on those farms, (c) they did most of their related SW development on Linux, and (d) they wanted in-box support for the systems used for SW development and related activities.

[log0]

Other than power management, which was a whole lot simpler when we kicked this off back in 2007, I imagine they're pretty pleased with the features and performance. Launch-time support (buy new HW, install a recent distro, use the system) was a higher priority than features and performance.

The common thread among the customers was that (a) they were building big compute farms with our CPUs, (b) they were running Linux on those farms, (c) they did most of their related SW development on Linux, and (d) they wanted in-box support for the systems used for SW development and related activities.

I see. They are not really using your GPUs for computation related tasks, just happened to have them in their systems. But PM should still matter though.

[wpoely86]

I see. They are not really using your GPUs for computation related tasks, just happened to have them in their systems. But PM should still matter though.

They ARE being used for computational task. Cuda & OpenCL are becoming very populair in the scientific community. And we want big clusters with lots of flops. Cooling is an expensive issue with big clusters, so good PM is very welcome.

On the other hand, most cluster I know, all use Cuda (nvidia) for computations. The bad reputation of Catalyst isn't helping to change that.

[log0]

They ARE being used for computational task. Cuda & OpenCL are becoming very populair in the scientific community. And we want big clusters with lots of flops. Cooling is an expensive issue with big clusters, so good PM is very welcome.

On the other hand, most cluster I know, all use Cuda (nvidia) for computations. The bad reputation of Catalyst isn't helping to change that.

But not by the customers who wanted the open source driver. This is what I was wondering about.

And yes from my own experience nvidia is the preferred hardware for (serious) computing.

[jrch2k8]

They ARE being used for computational task. Cuda & OpenCL are becoming very populair in the scientific community. And we want big clusters with lots of flops. Cooling is an expensive issue with big clusters, so good PM is very welcome.

On the other hand, most cluster I know, all use Cuda (nvidia) for computations. The bad reputation of Catalyst isn't helping to change that.

well in hybrid clusters you are right many use nvidia/tesla but still pure CPU cluster are very common and will be for many generations cuz there are workloads than are too expensive or hard to paralelize enough to show any gain on a GPU and in this cases AMD still keep some muscle due to performance/$$$ ratio[bulldozer opteron with optimized codepath[AVX/FMA/XOR/MT/etc] are beasts]