How much does Mantle benefit low end CPU setups, lets run a few tests.

[Kaapstad]

I have just tested Mantle and DX11 using the Thief game benchmark a single stock 290X and a 4930k in various different configurations using 2, 4 and 6 cores with Hyperthreading on and off. I have also used different CPU clockspeeds to see how much difference that makes.

Mantle fps in Red
DX11 fps in Green
C = Cores
HT = Hyperthreading

4930k @3.5ghz

2C = 44.4 43.9
2C2HT = 62.4 50.8
4C = 66.2 62.8
4C4HT = 66.0 64.1
6C = 66.2 65.9
6C6HT = 66.0 65.6


4930k @4.0ghz

2C = 46.7 48.6
2C2HT = 64.0 55.0
4C = 66.6 65.1
4C4HT = 66.0 65.6
6C = 66.3 66.1
6C6HT = 66.2 66.0


4930k @4.5ghz

2C = 50.6 47.1
2C2HT = 65.2 59.4
4C = 66.9 66.7
4C4HT = 66.3 66.6
6C = 66.5 66.6
6C6HT = 66.4 66.5


Some interesting things I picked up on

1. Why did Mantle fail to improve over DX11 with the CPU running just 2 Cores and no HT apart from where the CPU was @4.5 and Mantle showed a slight edge.

2. Mantle does look visually better when using 2 Cores and HT, less stuttering.

3. DX11 scored better than Mantle when the CPU was @4.5 in half the results.

4. Thief seems to run better with HT turned off when using 4 or 6 cores.

 

[Kaapstad]

Good benching kaaps, shame thief is such a terrible game though

Good Lord, I have not played the game !!!

 

[Kaapstad]

How does that CPU running 2 cores compare to say the Pentium g3230 iirc would that not give a better indication of how well mantle would work. Or have I completely missed the whole idea of mantle

Going by what I posted above, Mantle seems to make a significant difference if you are using an ivy bridge 2 core CPU with hyperthreading like in the link below.

http://www.overclockers.co.uk/showproduct.php?prodid=CP-430-IN&groupid=701&catid=6&subcat=1671

What is strange though is Mantle looks like it makes practically no difference if using the same CPU with the HT turned off.

What I was doing with the 4930k in the opening post was turning cores and HT off so that it was mimicking other IB CPUs, what still needs to be done is for someone to test it on the actual CPUs to see if they get similar results.

I am not familiar with your CPU so someone else really needs to answer that.

 

[Kaapstad]

You didn't actually state what the numbers indicate, averages I presume? Minimums are where Mantle is most interesting.

Quite a few situations where Mantle is making minimums 30-40% higher yet averages not change much at all.

I think what we are somewhat seeing, hugely higher minimums but sometimes very little increase in average if any, is indicative of frame pacing, which frankly is to be expected. Sacrificing really high frame rate for higher mins is fine by me.

Lack of frame increase with 2core no HT could simply be the game engine itself being limited by the cores. Maybe with only 2 cores it puts dx/mantle on one core and the main game engine on the other, when it detects more threads it will split the various bits across more threads freeing up the main thread to get a little more speed. However if minimums are up in that situation significantly then Mantle would still be helping.

The figures are averages, the minimums seemed reasonable whether using DX11 or Mantle. The really strange one with minimums was the 2 core bench, no HT with the CPU @3.5 - the reason it was odd was DX11 had a higher minimum, iirc it was about 17.5 compared to 14 which goes against what Mantle is supposed to do. Maybe it was just a fluke result but it was strange that Mantle did not perform when using just 2 Cores.

Also using 2 Cores and HT on with Mantle did give a better visual experience than DX11 with the same settings.

 

[Kaapstad]

I think a better test would be to use an actual weaker CPU instead of a £500 CPU and then try to gimp it.

For example. AMD Phenom II x6 1090T @ 4Ghz, R9 290 @ 1075 / 1400

DirectX: Min: 31.4
Mantle: Min: 57.1 (+88%)

DirectX: Max: 73.5
Mantle: Max: 89.6 (+20%)

DirectX: Avr: 45.4
Mantle: Avr: 67.8 (+50%)

Looks very different to yours doesn't it

All you are really showing is the difference between AMD CPUs and Intel ones, I may have been using a 4930k but don't forget that a 4770k is a better CPU for gaming.

 

[Kaapstad]

Sorry but am I missing something ? 4930k with 2 cores is equivalent to a low end cpu ?

They perform very similar

 

[Kaapstad]

Sorry but am I missing something ? 4930k with 2 cores is equivalent to a low end cpu ?

Your thread is about Mantle and weaker CPU's, run 2, 4, 6 and 8 Intel threads @ 3.5Ghz is not a weak CPU, just turning off cores does not constitute weak CPU's given that DX11 only uses one thread to render, PhysX uses another, object orientated calculations another.

Your doing it wrong, 4 of my cores @ 4Ghz are like your 4 cores at 2Ghz, one of my cores at 4Ghz is like one of yours at 2Ghz.

If anyone does not believe how bad a 4930k is on 2 cores and 4 threads @3.4ghz (memory @1600mhz).

Here is the proof

Go on Humbug you should have no trouble beating my score.

 

[Kaapstad]

Come on dude, put some effort in. Where are the 2 and 4 core results, what about different clock speeds?
As cap said, all you've really shown here is that a 4930K using 6 cores is better than a Phenom II using 6 cores under DirectX, which is hardly shocking is it? Is this part of the point of Mantle, to overcome the fact AMD haven't made a decent gaming CPU in years?

It feels to me a bit like people are trying to discredit Kaaps results because they didn't show Mantle being much better than DirectX. Given Kaaps rigs I suggest Kaap is a whole lot more likely to be unbiased than a number of people that only own recent AMD or Nvidia gaming GPUs. With 4 290Xs I'd imagine if Mantle could relieve any CPU bottleneck he might be getting (even with a clocked 4930k) I'm sure he'd be well happy (if he played either of the Mantle games). Just because Kaap isn't on a mission to stick it to anyone that owns an Nvidia card doesn't mean he wants to show Mantle is bad/pointless.
Maybe I'm reading too much into it...

+1

If people look closely at the results there is a big increase in performance using 2 cores and HT on with Mantle over DX11, this is exactly the place in the market that Mantle is supposed to benefit most.

 

[Kaapstad]

Like an i3.

It should also be noted that there is a huge increase for people with much older CPU's, i'm seeing a 50% increase in performance on a Phenom II.
Its gone from a big bottleneck to no bottleneck at all.

I would have thought people with an FX-63## would also see a huge increase.

In the first set of figures with the CPU @3.5, 2 Cores and 2 Threads (similar to an i3) Mantle had a 22.8% gain over DX11.

If your mobo allows it you should experiment with turning off cores to see the difference, in theory the less cores the bigger the improvement with Mantle.

 

[Kaapstad]

Read between the lines with this post...

After some research by myself i believe that the problem lies with the BF4 application itself and Dice/EA, rather than AMD. 7xxx series cards get nice gains on Thief and Starswarm but their gains on BF4 Mantle are more limited. You have to ask yourself why that is? Especially now after AMD have launched several more drivers. The problem is is DIce/EA and its an application issue. Hopefully it gets fixed in BF4, but the game has so many different issues on going.

However 7xxx still see gains on Mantle, nice gains in my HD7770 case.

Does CF work with Starswarm, if it does I may have to look into it.

 

[Kaapstad]

The problem here you have tested CF on DirectX what is in deed working and then Single GPU on Mantle.

CF works on DX11 but not on Mantle in Thief.

I would love to do some 4 card tests with Thief but at the moment there is no support.