Cat 14.2 B (Mantle update)

[davie jambo]

AA puts the bottleneck back on the GPU making mantle basically the same as dx.

Well that is not true

 

[Despoiler]

AA puts the bottleneck back on the GPU making mantle basically the same as dx.

With the exception that Mantle delivers much lower and consistent frametimes.

 

[Erenhardt]

Necro!
14.2 beta driver improves kaveri performance by up to 40% - whaaat?
http://benchmarkreviews.com/13238/amd-a10-7850k-performance-optimized-catalyst-14-2-driver/4/

That is quite a difference. I wonder how it stacks against a8-7600 now? Are both apu affected in the same degree? There is improvement in memory bandwidth benchmark, maybe that helped bandwidth starved 7850k.

Even Unigine Heaven benchmark sees 17% increase.

Shame the driver didn't make it in time for kaveri release, it would put it in different light.

How much more can they squeeze from that apu? I think there is a lot of room for improvements on the game devs side of things.

 

[Rvenger]

Actually if you read the entire review, that 14.2 driver improved on almost every metric and not just Mantle. Very good performance increases it seems but I don't think the increases was solely on Mantle.

 

[Erenhardt]

Actually if you read the entire review, that 14.2 driver improved on almost every metric and not just Mantle. Very good performance increases it seems but I don't think the increases was solely on Mantle.

Yes, I don't think that bf4 result was under mantle. Tomb Rider and Bioshock was definitely not under mantle. Its big boost in performance in dx, without mantle

If this thread is about mantle performance specifically, then please make my previous post a separate thread to avoid offtopic - which was not my intention.

 

[frozentundra123456]

Too bad they didnt show a discrete card like the 7750 or GTX750 to see if a similar improvement was made in performance with a discrete card.

Also, although it is a nice improvement, the absolute increase was at most 10fps. The percent increase sounds so good because it is starting from basically unplayable numbers. It does go from unplayable to "playable" depending on how picky one is, however.

 

[krumme]

Too bad they didnt show a discrete card like the 7750 or GTX750 to see if a similar improvement was made in performance with a discrete card.

Also, although it is a nice improvement, the absolute increase was at most 10fps. The percent increase sounds so good because it is starting from basically unplayable numbers. It does go from unplayable to "playable" depending on how picky one is, however.

I am pretty sure it would be the same with a discrete gcn card. Mantle just removes much of the cpu bottleneck. Using lower res the benefit would probably be yet higher.
40% is nice but if you want an average of say 80 fps on the huge maps with 64 players Mantle can easily give that for an ib i5 4c 4t non oc.
It all depens again how picky you are about your fps and how smooth they should be.

So mantle is basicly just giving you a 40% faster cpu in bf4. Better way to look at it. It might matter and give a huge even compettitive benefit - for some it just doesnt matter at all. Excactly as a faster cpu would.

 

[Erenhardt]

I am pretty sure it would be the same with a discrete gcn card. Mantle just removes much of the cpu bottleneck. Using lower res the benefit would probably be yet higher.
40% is nice but if you want an average of say 80 fps on the huge maps with 64 players Mantle can easily give that for an ib i5 4c 4t non oc.
It all depens again how picky you are about your fps and how smooth they should be.

So mantle is basicly just giving you a 40% faster cpu in bf4. Better way to look at it. It might matter and give a huge even compettitive benefit - for some it just doesnt matter at all. Excactly as a faster cpu would.

I just read the review again. 40% in BF4 is under mantle. But 33% in Tomb Raider is purely directX driver optimization.

Is it apu exclusive, or do gcn gpu benefit? I imagine 33% for all gcn GPUs would be groundbreaking and are more likely kaveri only.

 

[Rvenger]

If this thread is about mantle performance specifically, then please make my previous post a separate thread to avoid offtopic - which was not my intention.

Feel free to start a new thread, if you like.

 

[wand3r3r]

I've just upgraded to 14.2 and so far it's looking ok for mining.

 

[krumme]

I just read the review again. 40% in BF4 is under mantle. But 33% in Tomb Raider is purely directX driver optimization.

Is it apu exclusive, or do gcn gpu benefit? I imagine 33% for all gcn GPUs would be groundbreaking and are more likely kaveri only.

I think you are correct its kaveri related. But as said we need the discrete as comparison to tell the difference. But very interesting...

14.2 beta was a major, major, major release with new functionality - crazy. There must be a lot of bugs to iron out and optimizations to be made at least for the next half a year if not a year.

 

[Headfoot]

On the Cat 14.2's right now. But I still get the random hitches while on Mantle. Running 5239x1050 resolution, but I'm also on a 290 so I highly doubt its utilizing all 4GB of system memory, and it runs perfectly fine in DX without any hitching. I wonder if there is still a memory leak.

 

[3DVagabond]

Yes, I don't think that bf4 result was under mantle. Tomb Rider and Bioshock was definitely not under mantle. Its big boost in performance in dx, without mantle

If this thread is about mantle performance specifically, then please make my previous post a separate thread to avoid offtopic - which was not my intention.

No, not at all. It's about 14.2, but we were told that the only difference is in Mantle. So, if there are DX improvements, that's good info as well. Most games are DX, after all.

 

[frozentundra123456]

Too bad mobile kaveri is not out yet. The dx result is more interesting to me than mantle. For the desktop, I am still not a fan of apus. For the vast majority of cases, it still makes more sense to add a discrete card. But in mobile a 20 or 30 percent increase could make a big difference, where s discrete solution costs a lot more than one would expect kaveri to cost.