The AMD Mantle Thread

[desprado]

The fun fact is that it isn't even a GTX 780 Ti, it's a GTX 780 Ghz.

another Fun fact is that it is not Sp it is MP.

 

[SiliconWars]

Same result. Besides the slower older drivers. HardOCP looks like a mess tho.

The difference apples to apples is 11% and that's vs a 290 non X. The polish guys are 40%+ faster than a 290X? Doesn't seem likely.

edit- and we know that [H] actually plays while reviewing.

The obvious message is that it's going to be very difficult getting a balanced MP result. Logic can still apply though and will be seen with gameplay graphs when Mantle keeps solidly high straight lines, while DX11 suffers huge drops when the battles get big.

 

[mikk]

Oh, and here is another benchmark from Hardwareluxx on a 64-man server - the same map Golem.de used:

http://www.hardwareluxx.de/index.ph...nisse-im-kampf-mantle-vs-directx.html?start=1

That's the problem with MP.

Tested on a i7-3960X 3,3 GHz @3,9 Ghz--> more GPU bound than CPU bound.

 

[desprado]

HardOCPs first enlarged result is almost always misleading. Its terrible.

Off the topic but i dont know about HardOCPs but Anandtech and Guru3d provides honest and good review.

 

[sontin]

I have no knowledge of this MT dx nvda driver ... but it should be pretty simple, before and after numbers for the MT driver. Where can we dig em up? Anyway, if they'd given the massive increase in min fps's as reported for mantle, this forum section would have lightened up as a beacon seen from space.. So I doubt it.... I didnt see it.

Nobody has cared about this. Now with the focus of the MP part and the Star Swarn test demo it is getting more and more obviously that AMD has a DX driver problem and not that the Windows world has a DX problem.

Fact is the Star Swarm test demo archives 85-90% of the performance of Mantle with a nVidia card and Deferred Contexts.

 

[Erenhardt]

Lies, they did post how many people were on the server.. Someone translated the statement on the previous page..

That was me. All it says is they use 64-player game mode, but do not say if the servers are empy/full/half-full/half-empty.

 

[Slomo4shO]

The fun fact is that it isn't even a GTX 780 Ti, it's a GTX 780 Ghz.

HardOCPs first enlarged result is almost always misleading. Its terrible.

Apples to apples at 2xMSAA from Jan 20:

Considering this is the R9 290 and not the R9 290X, I still don't see how PC.Labs were able to obtain a ~40% performance increase over the R9 290X...

 

[sontin]

Tested on a i7-3960X 3,3 GHz @3,9 Ghz--> more GPU bound than CPU bound.

Golem.de used a Core i7-3770K. So the difference shouldn't be so huge.

 

[desprado]

Nobody has cared about this. Now with the focus of the MP part and the Star Swarn test demo it is getting more and more obviously that AMD has a DX driver problem and not that the Windows world has a DX problem.

Fact is the Star Swarm test demo archives 85-90% of the performance of Mantle with a nVidia card and Deferred Contexts.

That is really a huge number if it is true.

 

[mikk]

Golem.de used a Core i7-3770K. So the difference shouldn't be so huge.

4 cores vs 6 cores. Big difference. And golem did use an OC 290x.

 

[cytg111]

Same result. Besides the slower older drivers. HardOCP looks like a mess tho.

So, ShintaiDK, those mantle numbers look good so far, dont you agree?

 

[Spjut]

Comparing AMD and Nvidia DirectX performance by using BF4 may be misleading, AMD doesn't have any incentive to make specific DX11 optimizations for BF4 since those cards will just use Mantle instead

 

[desprado]

4 cores vs 6 cores. Big difference.

In BF4 no there is not
with 3770k or 4770k u can minimize the difference by overclocking and even benchmark shows that
3960X@ stock has only has 10% of advantage over 4770k@ stock.

 

[krumme]

Johan on the MSAA perf:
http://forum.beyond3d.com/showpost.php?p=1824792&postcount=1055

 

[Abwx]

Tested on a i7-3960X 3,3 GHz @3,9 Ghz--> more GPU bound than CPU bound.

The polish used a 4.5ghz HW , with 25% higher frequency on average and higher IPC it compensate somewhat for the 50% more core of the 3960x , yet the fps they got on BF4 MP at 1080p with no AA and no AF are about the ones that this site got at 1600p 4x AA and 16x AF.....

 

[cytg111]

In bf4 no there is not
with 3770k or 4770k u can minimize the difference by overclocking and even benchmark shows that
3960X@ stock has only 10% of advantage over 4770k@ stock.

See? Right there, even in your own typing. Should read your own post, its an argument against you.

 

[SiliconWars]

I guarantee you this is about workload - the harder the battle is to render, the better Mantle does. That's the difference between the golem.de results and the pclabs results (or maybe it's just what abwx suggested above and the pclabs results are completely whack).

I also guarantee that if Nvidia had some 40% higher performance in BF4 MP we'd have heard about it long before now - from them. They'd be shouting about it from the rooftops.

 

[ShintaiDK]

Why ? Because you say so ? If so, that's not good enough.

Because the first assumption is its directly compared, and its not. Not exactly lacking examples of people going wrong with it either.

 

[Sisyphean]

Even though MP is the primary setting of BF4 (and the series) I have to question the validity of scores obtained. With such dynamic scenes it's not proper to make a real comparison of performance. Isn't it entirely possible that certain frames have elements in them that can cause big differences in performance? Given the state of BF4 I don't see how an MP test is valid on such a short notice test. And before you think I'm here defending AMD, I am not, I'd like to see this as objectively as possible (plus I haven't even used AMD hardware for almost a decade now).

I don't understand the spite in people over such a thing. I'm always exited to see new developments and increase of competition, which are all good things for the end user. Even if Mantle is a failure ultimately it's still a good thing that it came out because it shows that the involved parties still care about trying to innovate

 

[Grooveriding]

Because the first assumption is its directly compared, and its not. Not exactly lacking examples of people going wrong with it either.

So since you can't take the time to read what you are looking at and understand what is being represented, it's:

HardOCPs first enlarged result is almost always misleading. Its terrible.

Another classic case of attacking a website's results without a valid argument, just opinion, because one does not have anything valid to offer yet is not happy with the results.

 

[desprado]

See? Right there, even in your own typing. Should read your own post, its an argument against you.

yup my english is a bit bad and sorry for that.

 

[Slomo4shO]

Because the first assumption is its directly compared, and its not. Not exactly lacking examples of people going wrong with it either.

This is only an assumption if you just look at the graphs and don't actually read the review. HARDOCP reviews all have an apples to apples section and the initial benchmarks are based on highest playable settings. The assumption is just that, an assumption. Reading comprehension provides plenty of clarification :whiste:

 

[ASM-coder]

I don't understand the spite in people over such a thing. I'm always exited to see new developments and increase of competition, which are all good things for the end user. Even if Mantle is a failure ultimately it's still a good thing that it came out because it shows that the involved parties still care about trying to innovate

---^^^

 

[ShintaiDK]

So, ShintaiDK, those mantle numbers look good so far, dont you agree?

If Mantle is what it takes to get multithreaded DX performance out of AMD cards. Just wierd they needed a new API for that instead of fixing their DX driver. But I wonder how much the coffers can hold to sponsor companies to use it.

 

[DamnedLife]

I dont understand the complains about pclab. They tested the MP after the release of BF4:

http://pclab.pl/art55318-5.html

And a GTX780 was 28% to 43% faster than the 290X.

That's 780 NOT EVEN 780TI, no way those results are true NO WAY.