Don't Look For Open-Source AMD CrossFire Anytime Soon

[GreatEmerald]

I'd imagine some kind of primitive benchmark or hardware performance database that Mesa could reference to gauge the relative strength of accelerators to know how to partition workloads.

Hmm, I think a profiling tool would be best. It would create profiles for each game with the ratio being different (because some cards do some things better than others, and games make use of different features).

[Calinou]

For most uses (gaming, not scientific/rendering use), there's a fix: use a single card. This way you use way less power (less noise/heat) and you don't lose much, to not say "anything", since games need SLI/CrossFire profiles to work best. This also helps against stuttering which is common...

[Kivada]

For most uses (gaming, not scientific/rendering use), there's a fix: use a single card. This way you use way less power (less noise/heat) and you don't lose much, to not say "anything", since games need SLI/CrossFire profiles to work best. This also helps against stuttering which is common...

No, we need to get the oss drivers to the point we can do this:

[curaga]

How is crossfire necessary for that? Use a high-end 6-way eyefinity card.

[Calinou]

How is crossfire necessary for that? Use a high-end 6-way eyefinity card.

It would be too slow, but that use is stupid anyway.

[bridgman]

Do you mean "5 monitors like that is stupid", "doing it on a 6-output card is stupid", or "the 4 way crossfire demo was stupid" ?

Multi-display gaming is one of the major use cases for the high end cards.

[curaga]

Crossfire scales badly + causes micro-stutter = one is better off with a high-end single card.

With four cards you get maybe 2x the perf of a single card, for 4x the power & price, with the stutter making it less enjoyable.

[zanny]

Crossfire scales badly + causes micro-stutter = one is better off with a high-end single card.

With four cards you get maybe 2x the perf of a single card, for 4x the power & price, with the stutter making it less enjoyable.

This doesn't mean that load balancing gl workloads across providers in Mesa would have the same issues. It would almost certainly have workload balance issues, and idle sync time overhead, but Crossfire has its bugs because its a black box. Can't say the same about a publicly improvable solution.

[curaga]

You sir are an optimist

Both SLI and CF work badly, and it's in direct business interest of both companies for them not to suck. A couple volunteers outdoing two focused companies that have thrown a lot of resources at the problem would be pretty unlikely.

[zanny]

You sir are an optimist

Both SLI and CF work badly, and it's in direct business interest of both companies for them not to suck. A couple volunteers outdoing two focused companies that have thrown a lot of resources at the problem would be pretty unlikely.

Just like how it is impossible for DRI PRIME to be better than Optimus and AMD's Switchable Graphics? (and imo, it is).