The AMD Mantle Thread

[Carfax83]

another set of benchmarks not matching pclab

http://www.purepc.pl/karty_graficzn...nych_battlefield_4_mantle_vs_directx?page=0,2

again the mantle numbers from pclab.pl are low in comparison


Carfax83 i'll reply to your post tomorrow.... its late and im tired.

It doesn't look like they used Windows 8.1..

The PClab.pl numbers mirror those of some of the other reviews, particularly that Mantle gives greater benefits for lower performing CPUs, and that NVidia can catch up and even exceed Mantle's performance provided certain conditions are met:

1) The use of Windows 8/8.1

2) A powerful Intel Core series processor is used.

 

[mikk]

another set of benchmarks not matching pclab

http://www.purepc.pl/karty_graficzn...nych_battlefield_4_mantle_vs_directx?page=0,2

again the mantle numbers from pclab.pl are low in comparison


Carfax83 i'll reply to your post tomorrow.... its late and im tired.

Windows 7 once again.

 

[Hitman928]

One of the reasons why I, and many others come to Anandtech is because Anandtech uses thorough, detailed and consistent testing procedures.

That Overclockers review doesn't even list the drivers they used.. But more to the point, as someone else pointed out, that's not the in game benchmark.

That's obviously something else that is less reliable, as the GTX 780 is leading the Titan, something which shouldn't happen.. Anandtech uses the late game benchmark, which simulates a late game workload, and is way more consistent.

They have a link to their testing procedures, http://www.overclockers.com/overclockers-updated-video-card-testing-procedure/

So yes, they use the late game benchmark just like anandtech. Also, the Titan and the 780 have a difference of 1.7%, so if the benchmark is as cpu limited as you suggest, then there is nothing out of line with their scores. If you look at the Anandtech scores and then compare them to the overclocker's numbers, the scaling between cards is exactly what you would expect. You used old data to make a conclusion and now refuse to accept the new data presented to you that shows that your conclusion was off, even when it is Anandtech's own updated results.

 

[Carfax83]

All you're showing is a GPU limited benchmark, where a faster GPU performs better. That has nothing to do with CPU performance or CPU driver performance. Note how the 7970GE is faster than the 7970, if they were tied, then maybe you'd have somewhat of a point.... but you're not proving any point with that graph.

Really, it's not a very difficult concept to understand guys....seriously.

The benchmark is CPU or GPU limited, depending on the card. A GTX 780 is CPU LIMITED in the benchmark as it's so damn fast that it's maxing it out at 1080p.

The much slower GTX 670 on the other hand, is still GPU LIMITED.

You guys are missing the point by a mile. The GTX 780 has a near 50% lead over the 7970 GE (and the benchmark can't go any higher btw) despite being only 20% faster on average.

DX11 multithreading only reduces or removes CPU limitations. When there is no CPU limitation, there is no benefit.

The GTX 670 is nearly as fast as the 7970 GE, despite being a notably slower card generally speaking. Why? Because DX11 multithreading to remove any CPU limitation in it's performance.

If AMD got their act together and made a similarly performing DX driver as NVidia, then the 7970 GE's performance would be significantly more.

 

[Midwayman]

Problem is, AMD misled people on the marketing hype..........specifically to those with upper tier products and setups..........I don't remember hearing them state to people in that crowd that the benefits would be minimal at best..........did you? Or anyone else?

Nope, but OTOH they didn't promise gains to the high end setups either. I think you read what you wanted to hear into their statements.

 

[Carfax83]

They have a link to their testing procedures, http://www.overclockers.com/overclockers-updated-video-card-testing-procedure/

So yes, they use the late game benchmark just like anandtech. Also, the Titan and the 780 have a difference of 1.7%, so if the benchmark is as cpu limited as you suggest, then there is nothing out of line with their scores. If you look at the Anandtech scores and then compare them to the overclocker's numbers, the scaling between cards is exactly what you would expect. You used old data to make a conclusion and now refuse to accept the new data presented to you that shows that your conclusion was off, even when it is Anandtech's own updated results.

Ah, it's a good thing you showed me that, as now it makes sense.

Overclockers used 8x MSAA, which explains why the R290 did so well, and why the Titan and the GTX 780 scored so similarly.

So basically, the benchmark as Overclockers conducted it, is way more GPU bound than Anandtech's, which used 4x MSAA.

 

[Paul98]

Any word on what new GPU optimizations are being worked on?

 

[96Firebird]

Nope, but OTOH they didn't promise gains to the high end setups either. I think you read what you wanted to hear into their statements.

Or some of the hype in this very thread, like when SiliconWars claimed a single 290X would beat Titan SLI. D:

 

[Spjut]

Or some of the hype in this very thread, like when SiliconWars claimed a single 290X would beat Titan SLI. D:

It might if you use a Core 2 Duo

 

[blackened23]

It might if you use a Core 2 Duo

Using a low end CPU for 1 or 2 Mantle games isn't a reasonable proposition. It's an absurd proposition. Generally speaking, those who buy 500$-700$ CPUs don't pair their GPUs with extremely slow CPUs. If you can afford that level of GPU, you can afford a CPU that isn't garbage.

I understand that Mantle helps hide CPU bottlenecks, but again, buying a crap CPU for 1 game is a non reasonable and extreme proposition. Most people that have the budget for a 700$ GPU (which is what most 290X cards cost in the states) will have the budget for a good high end CPU. Heck, even if they don't, the 4670k can be had for only 200$. That is not much more than the best APUs, while the 4670k will just simply perform much better. On that note, even low end i3 CPUs outperform APUs and the C2D you mention across 98%+ of real world tasks so this isn't something that Mantle effectively addresses. It only addresses a few outlying scenarios in games with Mantle support.

I get that Mantle hides CPU deficiencies. That does not address the fact that Mantle does not have 100% coverage in all games. Not even close to 100% of games, it is far less than 1% of games especially in 2014 - there are 2 confirmed Mantle games with 2-3 "maybe" games. The reasonable buyer who can afford a 700$ GPU will not buy a CPU that performs okay when Mantle hides the problem in 1 or 2 games. They will buy a CPU that performs in all games. And a CPU that performs well in 100% of games is not expensive, either. (4670k).

 

[ASM-coder]

I understand that Mantle helps hide CPU bottlenecks, ...
I get that Mantle hides CPU deficiencies. (4670k).

Mantle does not "hide" anything. It reduces CPU dependence.

 

[blackened23]

I like how you misquoted me there and ignored the crux of my argument. When I say CPU deficiencies, i'm referring to APUs or the C2D mentioned by the poster above me. I'm not referring to the 4670k, again, I like how you misquoted me. Mantle does not have 100% coverage, period, so using a horrible CPU just for Mantle use is absurd. Period. The reasonable buyer, especially one who can afford a 600-700$ GPU for Mantle, will not spend 700$ on a GPU and then get the worst 50$ CPU they can buy. That's so silly it just blows my mind. They will buy something that performs in *everything*. They won't get a crap CPU that potentially performs in one or two games with Mantle while performing horribly in everything else.

If you can afford a 700$ GPU, you can afford a CPU that isn't crap. And even if you are budget limited, you can get a 200$ 4670k which is quite superior to a C2D or APU and performs well in 100%. Or you can buy an APU with Mantle which performs well in <1%.

 

[Hitman928]

Ah, it's a good thing you showed me that, as now it makes sense.

Overclockers used 8x MSAA, which explains why the R290 did so well, and why the Titan and the GTX 780 scored so similarly.

So basically, the benchmark as Overclockers conducted it, is way more GPU bound than Anandtech's, which used 4x MSAA.

Your arguments contradict themselves and have no evidence to support them outside of old data that's been fixed. You can believe whatever you want but I suggest you try and take a step back and try again.

 

[Hitman928]

I like how you misquoted me there and ignored the crux of my argument. When I say CPU deficiencies, i'm referring to APUs or the C2D mentioned by the poster above me. I'm not referring to the 4670k, again, I like how you misquoted me. Mantle does not have 100% coverage, period, so using a horrible CPU just for Mantle use is absurd. Period. The reasonable buyer, especially one who can afford a 600-700$ GPU for Mantle, will not spend 700$ on a GPU and then get the worst 50$ CPU they can buy. That's so silly it just blows my mind. They will buy something that performs in *everything*. They won't get a crap CPU that potentially performs in one or two games with Mantle while performing horribly in everything else.

If you can afford a 700$ GPU, you can afford a CPU that isn't crap. And even if you are budget limited, you can get a 4670k which is quite superior to a C2D or APU and performs well in 100%. Or you can buy an APU with Mantle which performs well in 1%.

I think the c2d quote was said mostly in jest. . .

 

[PPB]

I like how you misquoted me there and ignored the crux of my argument. When I say CPU deficiencies, i'm referring to APUs or the C2D mentioned by the poster above me. I'm not referring to the 4670k, again, I like how you misquoted me. Mantle does not have 100% coverage, period, so using a horrible CPU just for Mantle use is absurd. Period. The reasonable buyer, especially one who can afford a 600-700$ GPU for Mantle, will not spend 700$ on a GPU and then get the worst 50$ CPU they can buy. That's so silly it just blows my mind. They will buy something that performs in *everything*. They won't get a crap CPU that potentially performs in one or two games with Mantle while performing horribly in everything else.

If you can afford a 700$ GPU, you can afford a CPU that isn't crap. And even if you are budget limited, you can get a 200$ 4670k which is quite superior to a C2D or APU and performs well in 100%. Or you can buy an APU with Mantle which performs well in <1%.

And if you have an aged cpu and only afford to change the gpu and not the hassle of a platform (cpu+motherboard) change?

Mantle enables you to keep your cpu for longer, and only change the gpu when it no longer fits your perf target in a game. Save cash with unnecesary platform changes.

Fudsters blame the cpu for being weak when it is the software to blame most of the time. Mantle is analogous to really well coded production software, like renderers, video transcoding, etc. As those are the ones that are best as extracting ever single bit of performance out of your cpu.

 

[Carfax83]

Your arguments contradict themselves and have no evidence to support them outside of old data that's been fixed. You can believe whatever you want but I suggest you try and take a step back and try again.

Now it seems you're the one who can't accept the data. It clearly said in the link you provided that they used 8x MSAA, which would obviously generate different scores from the Anandtech benchmark which used 4x MSAA..

Unless you're suggesting that 8x MSAA and 4x MSAA have the same performance hit, there is no contradiction.

Also, the video I posted showing the performance between the 7850 and the GTX 660. How are you going to explain that one?

You guys just keep coming up with ways to discredit DX11 multithreading, when you should be discrediting AMD's lackluster DX drivers..

DX11 multithreading isn't as efficient as Mantle, but it does work if the drivers allow it to.

 

[blackened23]

And if you have an aged cpu and only afford to change the gpu and not the hassle of a platform (cpu+motherboard) change?

Mantle enables you to keep your cpu for longer, and only change the gpu when it no longer fits your perf target in a game. Save cash with unnecesary platform changes.

That would be a true statement if Mantle had 100% coverage in 100% of games. As things are, again, Mantle has 2 confirmed games in 2014 and a few more "maybe" games. So you're essentially holding onto an incredibly slow CPU for 1-2 games.

If you want to claim that's a realistic scenario, have at it, but PC gamers generally play a ton of games, not just 1-2. If you're stating otherwise, I would argue THAT is fud. Nobody plunks down 700$ for 1-2 games, and CPU is a significant issue in many games. And most of those games do not have Mantle support. Anyway, since PC gaming is extremely cheap with constant Steam sales, the typical PC gamer buys a ton of games. Heck, I bought tons of AAA titles that were a few months old for 10 bucks or less. There's no reason to limit myself to 1-2 games, because PC games are CHEAP. The reality of the situation is that anyone that can afford a 600-700$ GPU will equally be able to afford a CPU that performs well in 100%. Mantle is < 1%.

 

[GaiaHunter]

The benchmark is CPU or GPU limited, depending on the card. A GTX 780 is CPU LIMITED in the benchmark as it's so damn fast that it's maxing it out at 1080p.

The much slower GTX 670 on the other hand, is still GPU LIMITED.

It is clear that the GTX 780 is not CPU bottlenecked in that test, otherwise lowering resolution wouldn't increase FPS.

Lowering resolution sees an increase of 30% performance for the 780, a 34% performance increase for the 7970GHz, a 42% increase for the GTX 770 and a 45% increase for the GTX 670.

It is just a case of the hardware resources of NVIDIA offering a better balance in CiV5 than AMD resources.

The 290X/290 on the other hand have a different balance of resources compared to the 7970.

A CPU limited test looks like this.

And this is a case where the game engine favors AMD architecture instead of NVIDIA architecture.

 

[ASM-coder]

I like how you misquoted me there and ignored the crux of my argument. When I say CPU deficiencies, i'm referring to APUs or the C2D mentioned by the poster above me. I'm not referring to the 4670k, again, I like how you misquoted me.

I didn't misquote you, everything in the quotations was written by you.
Admittedly, the 4760k was out of context, but that was an unintentional over-site, my mistake.
My point being... Mantle doesn't hide anything. Your argument becomes tainted when prefaced with mischaracterizations.

 

[Hitman928]

Now it seems you're the one who can't accept the data. It clearly said in the link you provided that they used 8x MSAA, which would obviously generate different scores from the Anandtech benchmark which used 4x MSAA..

Unless you're suggesting that 8x MSAA and 4x MSAA have the same performance hit, there is no contradiction.

Also, the video I posted showing the performance between the 7850 and the GTX 660. How are you going to explain that one?

You guys just keep coming up with ways to discredit DX11 multithreading, when you should be discrediting AMD's lackluster DX drivers..

DX11 multithreading isn't as efficient as Mantle, but it does work if the drivers allow it to.

You keep trying to change the argument every time the data doesn't suit you. I never said anything about 8x and 4x MSAA having the same performance hit. The 7850 and GTX 660 example you posted is of a different game with old drivers on I believe an alpha build of the game. I've also never tried to discredit DX11 multithreading, only that the huge performance gains you originally claimed have no real evidence to back it up, and there isn't any that has been presented, only evidence to the contrary in regards to Civ5.

Like I said, I suggest you take a step back and look at it again and I think you'd come to a different conclusion. I have no skin in this game because I don't even really game any more and don't care one way or another about which brand performs better in any game. All I want is for mantle or DX or whatever to make more efficient use of hardware so that when I do decide to play a game twice a year, I don't have to upgrade my comp just to play that one game.

 

[ASM-coder]

That would be a true statement if Mantle had 100% coverage in 100% of games. As things are, again, Mantle has 2 confirmed games in 2014 and a few more "maybe" games. So you're essentially holding onto an incredibly slow CPU for 1-2 games.

They may play multiple games, but may only play 1 or 2 titles that really
stress their CPU, or that really need the bump in FPS.
Bioshock Infinite and many other games play fine on my aging setup.
BF4 is a stretch.

 

[smackababy]

So, is consensus in whether Mantle ridicules anything or not? I don't notice much of an improvement in BF4, but my setup apparently isn't what Mantle was aiming to improve much.

 

[Carfax83]

It is clear that the GTX 780 is not CPU bottlenecked in that test, otherwise lowering resolution wouldn't increase FPS.

I think you misunderstand the meaning of being CPU bottlenecked. Increasing resolution leads to being more GPU limited, and lowering it leads to being more CPU limited generally speaking.

The GTX 780 is CPU limited @ 1080p in the Civ 5 benchmark because it's MAXING IT OUT. Basically, the frame rate can't go any higher. Increasing the resolution to 1440p makes it GPU limited again.

I think I posted the wrong graph initially. I should have posted the 1080p graph from the get go..

 

[blackened23]

I didn't misquote you, everything in the quotations was written by you.
Admittedly, the 4760k was out of context, but that was an unintentional over-site, my mistake.
My point being... Mantle doesn't hide anything. Your argument becomes tainted when prefaced with mischaracterizations.

When you re-arrange my wording to make it sound like I said something that I didn't, you misquoted me. Period. Doing stuff like this makes it seem like you're trying to spread FUD, and i'd like to think that you aren't doing that. Don't try to make me sound like i'm saying something i'm not. You tried to make it sound like my statement about CPU deficiencies was related to the 4670k. The reality was, this is what I stated:

That does not address the fact that Mantle does not have 100% coverage in all games. Not even close to 100% of games, it is far less than 1% of games especially in 2014 - there are 2 confirmed Mantle games with 2-3 "maybe" games. The reasonable buyer who can afford a 700$ GPU will not buy a CPU that performs okay when Mantle hides the problem in 1 or 2 games. They will buy a CPU that performs in all games. And a CPU that performs well in 100% of games is not expensive, either. (4670k).

So what i'm saying here is the notion of pairing a high end GPU for Mantle with a low end CPU is absurd. And to be clear, when I say low end CPU, i'm referring directly to APUs and the C2D mentioned by the poster above me. The 4670k is a great CPU that performs well in 100%, and that is what I stated - a 4670k performs excellently in 100% of games. Whereas you can get a cheap APU or hold on to a C2D that performs well in less than 1% of games if they have Mantle support. Therefore, pairing a 290X with a low end CPU makes no sense. At all. If you can afford a 600-700$ GPU, that means you can afford a CPU that isn't junk.

Again, don't misquote me. Don't try to make it sound like I said something that I didn't. If you want to argue facts, have at it. Don't try to twist and re-arrange my words into something else. Doing so makes it seem like your intentions are nefarious or you're trying to spread FUD. Again, i'd like to think that you aren't doing that. Hopefully, your mis-quotation was a mere accident. Perhaps double check before hitting the post button. Thank you.

 

[hungtran]

Mantle's potential is tantalizing even for Intel CPU owners. It means someone running a i7-4770k today will not have to get a new processor to go with their next gen GPUs. Long term, AMD doesn't mind if NVIDIA adopts Mantle because higher Mantle adoption will close the gap between their CPUs and Intel's for consumer gaming. They want to move away from the node reduction race and bide time until eventually APUs will be sufficient to run most Mantle games on medium to high settings at above satisfactory frame rates. This is their ultimate goal.