Trinity & GCN Crossfire?!

[NTMBK]

Someone over on the CPU forums has apparently got his Trinity APU to crossfire with a HD7750: http://forums.anandtech.com/showpost.php?p=34812674&postcount=55 Of course, his proof consists of easily photoshopped screenshots, but still... if this is true, its very interesting. (I figured I'd post this in a new thread, as at the moment it's buried on page 3 of a CPU thread, and hardly going to get noticed by you GPU types. )

 

[Homeles]

Link got killed by truncation.

Here's a working link:
http://forums.anandtech.com/showpost.php?p=34812674&postcount=55

 

[MeldarthX]

I was in that sub also and find if this is new and works well......

Damn this changes a lot of recommendations for low end....and seriously well done to AMD to getting that to work.

 

[NTMBK]

Link got killed by truncation.

Cheers, should be fixed now.

 

[Homeles]

I don't see why Trinity wouldn't be able to do it. Richland can, with the same silicon. It's possible a recent driver update enabled it. I've not seen an asymmetric crossfire review for Trinity, but I'd imagine it'd be a nightmare for real world usage.

 

[SPBHM]

I don't see how this combination would work well... IGP = 384 VLIW4, 24 TMUs, 1866 DDR3, 775 = 512 GCN, 32 TMUs, 4500 DDR5

just look at how much faster the 7750 alone is compared to the IGP
http://www.hardware.fr/articles/878-10/gpu-f1-2011-civilization-v-battlefield-3.html

IF amd could make a CF between this, than they should enable CF for 7770+7870
or 6670 + 7750

it would probably have lot's of stutter, frames not being displayed correctly!?

Richland is only compatible (or at least recommended for) with some slower Mars (384 GCN) GPUs

they don't even mention the faster (higher clock and DDR5 like the 8790m) Mars GPUs, only the 650MHz with DDR3 as the highest model (8730m).

 

[BrightCandle]

I doubt anyone would want to use it in practice! But I am kind of interested to see the fraps output, I kind of expect it to be horrific but may as well see it.

 

[MeldarthX]

If we can get BGT to run some stuff for us would be very interesting to see how it fairs.....

 

[bgt]

If we can get BGT to run some stuff for us would be very interesting to see how it fairs.....

As far as I can see it here with f.i the Valley bench it runs ok. Also TR2013 runs ok in normal mode.
BTW I updated my input with more tests.

 

[MeldarthX]

BGT what benches you got to run - yea we're a pain - but if you can run with and without the crossfire a few benches......

I'm sure the op could put them at the top - what screen size do you play at?

 

[bgt]

what screen size do you play at?

1920x1080

 

[bgt]

I've not seen an asymmetric crossfire review for Trinity, but I'd imagine it'd be a nightmare for real world usage.

As far as I can see it it runs smooth.

 

[Maximilian]

Dont want to sound like im thread crapping but why is this a big deal? Did Xfire previously not work with APU's and the 7xxx series?

 

[BrightCandle]

As far as I can see it it runs smooth.

That is about as useful as a thumb up my arse, seen many of these things said about AMD's crossfire which is just a mess so I would rather trade in actual data.

Use fraps, capture the frame times and lets have the data from at least one popular game.

 

[Rvenger]

Dont want to sound like im thread crapping but why is this a big deal? Did Xfire previously not work with APU's and the 7xxx series?

Nope, side of box states the 6450, 6570 and 6670 are dual graphics capable only. This is interesting, I wonder if AMD has something going on behind the scenes with Richland.

 

[bgt]

Use fraps, capture the frame times and lets have the data from at least one popular game.

http://dl.dropbox.com/u/11003205/Benchmarks.rar

Never used Fraps before........can you do something with this?

 

[BrightCandle]

http://dl.dropbox.com/u/11003205/Benchmarks.rar

Never used Fraps before........can you do something with this?

Not sure why there are so many traces but I had a look at 2. I looked at the most recent frame times and the longest trace.

The latest trace shows periodic jumps of up to 16ms ever 30 frames or so and otherwise hovers around 20ms. That is pretty jittery, its certainly noticeable but its not awful. The big jumps down suggest multiple runt frames occurring so its quite likely FPS is borked in places and overstated. I wouldn't consider this playable.

The longer trace shows the classic signs of runt frames every other frame and severe microstutter for much of the rest of the trace. Microstutter is around 12ms when its not runting which is some of the worst I have ever seen. Not to mention the multiple jumps up into the 500ms range and plenty of other huge hitches.

So all in all in summary its showing worse behaviour than crossfire normally does and that is mighty impressive. Its a mess, a complete disaster and measurably unsmooth. Animation will be severely impacted as will the frame output. The practical FPS count is likely to be half of what is being displayed in the longer trace, without FCAT output can't be sure but it should feel like 24 fps or less.

Its that bad.

 

[bgt]

measurably unsmooth

But in reallife its not unsmooth....so there must be another factor correcting it for screen display(or Fraps does not measure it after the frames are corrected). The TR benchmark has no stutters I can see. Also playing TR shows no stutters(normal settings).