Originally Posted by xoleras
Originally Posted by KyadCK
Ah that's right, you got the LE chips.
Ya, Mantle isn't a stuttery mess, you just got boned on having a Tahiti with 2GB.
Seriously, they're like the only cards with a problem, my Tri-Fire is butter.
20% is pretty good no? And more than that is the smoothness, which I will never let go since nVidia users certainly didn't. Still, I'm glad we can agree on something for a change. To Mantle, and the changes it forces in DX and OpenGL.
Hmm? I'm not saying the gains aren't good. I'm saying they're definitely not as good as I expected, most of them are around 10% or less for the high end 290X cards. Now 10% free performance, I welcome this. AMD seemed to indicate that the gain would be much higher, though. And i'm glad that DX is getting similar features in the future.
That said, in terms of smoothness if you look at the graphs - they're all generally sub 25ms so everything is smooth whether its DX or Mantle. It isn't a situation like the pre frame pacing fix days with 150ms or more. Anyway,....
One thing that interests me greatly is whether AMD can leverage Mantle in other ways. I remember several Glide games back in the day weren't so much faster than the D3D version, but the glide versions did have additional graphical features. Maybe AMD could do something like that for GPU limited games, and for the most part BF4 is very reliant on GPU unless you're using a cheese CPU. Just play to the game's strengths. Some glide games were about performance, but many more were about having those additional graphical features. Would be interesting, since a (mostly) GPU limited game doesn't play to Mantle's strengths that much IMO - the gains are good for the low end, not as good for high end.
Maybe. AMD has stated that Mantle can "do everything DX can and more". I would expect Thief to be the game using those extra features over performance though, not Battlefield.
Some of the people I talk to wanted screenshots of Mantle vs DX, so I took these a day or two ago, same spot as I did for the bench above; (View full size, they're 1080)
Ignoring the missing building in the DX one that had to do with loading time, the person noted that the AO and AA on the Mantle screen (wall on the right and trees/sand) looked better done on Mantle. I didn't really pay enough attention to if the trees move or not though, which would make a difference obviously.
If true, the better AO could help out a lot in Thief, as well as any other lighting and shadowing related effects. I guess we'll see in the future.
Originally Posted by Kinaesthetic
Originally Posted by KyadCK
You still have two 7870s unless you sold them. Go try those, they can use Mantle too. So can every other GCN card ever made.
Wow, assume much without reading? Would it make nVidia users feel better if we just erased all the nVidia scores from the benchmarks since they're irrelevant in this test anyway? Sheesh.
As he said in his post, Mantle is about removing CPU bottlenecks. You are not going to have a CPU bottleneck at 1600p Ultra with a single GPU
You want to know what WILL cause a CPU bottleneck? 7990/7970 Trifire on a 8320 at 1080p. Mantle
did much better than DX11
. 20% better in the exact same position.
Warning: Spoiler! (Click to show)
Ya that's done in the testing range, but it's also 1080p Ultra preset 4x MSAA. On an AMD CPU
I hate to be the insufferable person that points this out, but you do realize that both pictures show that there IS a CPU bottleneck in BOTH instances being shown. To say that there isn't is to completely discount all data. Because clearly there is one happening. Both in a single card configuration, and in a dual + card configuration.
Are the gains as high with a single card configuration than a dual + card configuration? No. Is there still a bottleneck clearly being shown? Yes.
I'm actually confused by your response. Both pictures are with Tri-fire. There is no single-card anything in there. Due to this, yes, there should be a CPU bottleneck in both pictures, because it's freaking Tri-fire on 1080p, what else is it going to be? Mantle's good, it ain't that good.
However, you assume that the gains Mantle brings to the table are CPU only
. That does not have to be the case. In the past, AMD has said that Mantle could, for instance, enable an easier way to do AA that wouldn't tax the GPU as hard. If that were/is the case, then the gains you see on the graph in the OP with 1600P single card would be as simple as that.
On the other hand however, if there is a CPU bottleneck on the single-card test, why does the DX11 Crossfire test (in the 2nd post) show massive gains (52 FPS vs 81FPS)? That is not an example of a CPU bottleneck. For that matter I'm still not convinced the Crossfire bench is a CPU bottleneck either since they raised the settings.