Mantle API presentation by AMD, DICE and Oxide - AMD Summit 2013

[the54thvoid]

Pixie dust and black magic.

Put me down for a black pixie and some magic dust instead.

 

[TheMailMan78]

The old saying goes "If its to good to be true, then chances are its not". These performance slides are reminiscent of a Chuck Norris meme.

 

[xenocide]

That's because i7-4770 is in 350$ price range while A8-7600 is below 120$. yo yo yo

http://i.imgur.com/8axFUzi.jpg

I'll bite--it's got nothing to do with price, it's more of AMD's shit marketing, assuring you their products are better than they really are. AMD shows these pretty slides using a bunch of lopsided synthetic benchmarks in outlandish configurations that give them performance benefits. All of those are in reference to the performance of the IGP--which almost anyone will tell you Intel has continually lagged behind AMD in, but that's to be expected considering AMD nearly bankrupted themselves buying up the second largest GPU manufacturer a few years back, they kind of had a head start.

All of that aside the issue at hand is that Oxide Games is claiming using DirectX that an A8-7600 performs about the same as an i7-4770 in "CPU Intensive" situations--that is false. There are hundreds of real world benchmarks that suggest that is nearly impossible. If we look at current AMD APU's we see the CPU portion performs at best around the same level as an i3-3220--which has half as many cores and threads as the cited i7. It is inconceivable that a 4 Core AMD CPU would offer even remotely similar performance to a 4 Core Intel CPU with Hyperthreading in CPU-heavy scenarios. Basically the slide doesn't make sense--if they are making use of the iGPU the APU should win hands down, if they are using a discrete card the Intel CPU should decimate the APU. They are relatively close, and that is confusing as hell.

 

[RoutedScripter]

big thread on overclock.net got deleted before I could save a ton of slides ... but im not in a hurry basically, i can wait 7 days for the talk along with the slides.

 

[HD64G]

Here are the 88 slides: http://imgur.com/a/4WiVM/all

 

[bpgt64]

Meh, having access to a Micro center means intel CPU's at ~200 bucks (4770k). Kevari seems like a 15% overall performance boost. Interested in seeing how Mantle stacks with an APU setup.

 

[TheoneandonlyMrK]

I'll bite--it's got nothing to do with price, it's more of AMD's shit marketing, assuring you their products are better than they really are. AMD shows these pretty slides using a bunch of lopsided synthetic benchmarks in outlandish configurations that give them performance benefits. All of those are in reference to the performance of the IGP--which almost anyone will tell you Intel has continually lagged behind AMD in, but that's to be expected considering AMD nearly bankrupted themselves buying up the second largest GPU manufacturer a few years back, they kind of had a head start.

All of that aside the issue at hand is that Oxide Games is claiming using DirectX that an A8-7600 performs about the same as an i7-4770 in "CPU Intensive" situations--that is false. There are hundreds of real world benchmarks that suggest that is nearly impossible. If we look at current AMD APU's we see the CPU portion performs at best around the same level as an i3-3220--which has half as many cores and threads as the cited i7. It is inconceivable that a 4 Core AMD CPU would offer even remotely similar performance to a 4 Core Intel CPU with Hyperthreading in CPU-heavy scenarios. Basically the slide doesn't make sense--if they are making use of the iGPU the APU should win hands down, if they are using a discrete card the Intel CPU should decimate the APU. They are relatively close, and that is confusing as hell.

Whilst I get what your saying about the slides, implying Amd pick and choose benches is a bit silly when its a direct game comparison(admittedly vague on specs) and its also slightly biased, do you think intel plays fair with benchmarks, or that their intel compiler (often used) is not in effect a cheat on benches against Amd's arch.
.

 

[erocker]

So... any demo using an actual game yet? No? Yawn...

 

[Frag_Maniac]

So... any demo using an actual game yet? No? Yawn...

Good point, graphics demos can too easily be slanted toward the strengths of the software they're testing. It's a lot like the way synthetic benchmarks yield big results when games don't.

 

[xenocide]

Whilst I get what your saying about the slides, implying Amd pick and choose benches is a bit silly when its a direct game comparison(admittedly vague on specs) and its also slightly biased, do you think intel plays fair with benchmarks, or that their intel compiler (often used) is not in effect a cheat on benches against Amd's arch.
.

I don't deny any of that, but we also have no real world examples to compare them against. The point I'm making is this;

They claimed DirectX caused significant CPU bottlenecks, yet the DirectX portion has a near identical difference between a very powerful Intel CPU and a not so powerful AMD APU (using the CPU portion). They claimed Mantle would remove the need for a very powerful CPU entirely, yet the difference is between the AMD APU and Intel CPU is almost identical when Mantle is implemented. They claim that example is a CPU heavy workload, but somehow a CPU with better threading and substantially better per thread performance is barely beating an APU. All of these things cannot be simultaneously occuring is the point I'm making. If the benchmark is GPU bound there should be no real difference between the two CPU's no matter what API is used, if it's CPU bound the Intel CPU should be blowing the doors off the APU in the bottom graph.

The ~20fps gain with Mantle is a bit more than I expected, but not outlandish.

 

[HD64G]

I don't deny any of that, but we also have no real world examples to compare them against. The point I'm making is this;

They claimed DirectX caused significant CPU bottlenecks, yet the DirectX portion has a near identical difference between a very powerful Intel CPU and a not so powerful AMD APU (using the CPU portion). They claimed Mantle would remove the need for a very powerful CPU entirely, yet the difference is between the AMD APU and Intel CPU is almost identical when Mantle is implemented. They claim that example is a CPU heavy workload, but somehow a CPU with better threading and substantially better per thread performance is barely beating an APU. All of these things cannot be simultaneously occuring is the point I'm making. If the benchmark is GPU bound there should be no real difference between the two CPU's no matter what API is used, if it's CPU bound the Intel CPU should be blowing the doors off the APU in the bottom graph.

The ~20fps gain with Mantle is a bit more than I expected, but not outlandish.

You must be very confused. The diffirence w/o mantle is 40-50% between Kaveri and i7. When Mantle is on it is <10%. Where do you see it doesn't work as intended? It just stops games giving the main role to CPUs. And having a $150 CPU in games using mantle will be the same as using a $300 one. Simple as this.

 

[xenocide]

You must be very confused. The diffirence w/o mantle is 40-50% between Kaveri and i7. When Mantle is on it is <10%. Where do you see it doesn't work as intended? It just stops games giving the main role to CPUs. And having a $150 CPU in games using mantle will be the same as using a $300 one. Simple as this.

Show me a CPU Intensive program where a 4C i7 performs only 50% faster than a 4C APU. Every benchmark I see pegs it at at least 100% faster.

 

[RoutedScripter]

So... any demo using an actual game yet? No? Yawn...

Silence before the storm.

You must be very confused. The diffirence w/o mantle is 40-50% between Kaveri and i7. When Mantle is on it is <10%. Where do you see it doesn't work as intended? It just stops games giving the main role to CPUs. And having a $150 CPU in games using mantle will be the same as using a $300 one. Simple as this.

I don't know if those guys are just that stupid or trolling the hell out of this thread.

 

[erocker]

Silence before the storm.

Are they going to be showing any BF4 w/Mantle at CES this week? I though I heard something about it. I'm also curious to see how it's coming along with Star Citizen. I left a couple questions for the devs about it... It might be too early for an answer though.

 

[Mathragh]

Are they going to be showing any BF4 w/Mantle at CES this week? I though I heard something about it. I'm also curious to see how it's coming along with Star Citizen. I left a couple questions for the devs about it... It might be too early for an answer though.

Apparently there has already been a demonstration, according to this dutch site:
Edit: also got a note at anandtech.

They tested with an R9 290X and a kaveri APU, difference between directx and mantle in BF4 was 45%. They also state that this percentage is a bit variable, with some set-ups and scenes having a lower boost, but with stuff like crossfire the boost could be well in excess of 45%.

For a game that is supposed to be quite CPU-agnostic(as long as there are atleast 4 threads) thats quite a big boost imho.

 

[crazyeyesreaper]

single player not multiplayer thus kinda worthless as no one really gives a flying f about BF4 single player lol

 

[Mathragh]

single player not multiplayer thus kinda worthless as no one really gives a flying f about BF4 single player lol

Wasnt multiplayer even more CPU-bottlenecked than singleplayer?

 

[crazyeyesreaper]

yup pretty much depending on the map / player count etc but then with server sync and other background items such anti cheat systems etc who knows maybe single player benefits more and thus got shown off. That or its just an internal benchmark that they used as a quick demo to prove it does run lol

 

[Mathragh]

yup pretty much depending on the map / player count etc

Then logically you'd expect an increase in excess of 45% with lower end CPU's that usually struggle in MP, since mantles main advantage is supposed to be alleviating CPU bottlenecks. I Suppose atleast.
We'll have to wait for more info as always, but the 45% is way over what I was expecting, especially in singleplayer, where the CPU performance hardly seemed to matter.

 

[crazyeyesreaper]

One thing to keep in mind Single player is heavily scripted meaning easy to optimize and probably easy to work with mantle and get a large performance boost. Multiplayer on the other hand is unpredictable so performance difference could be less or more depending on situation / map / etc

 

[Mathragh]

One thing to keep in mind Single player is heavily scripted meaning easy to optimize and probably easy to work with mantle and get a large performance boost. Multiplayer on the other hand is unpredictable so performance difference could be less or more depending on situation / map / etc

Yeah as usual we'll have to wait a bit longer, but as Erocker stated he "heard something" and I came across some info regarding the demo so I decided to share it

 

[TheMailMan78]

Apparently there has already been a demonstration, according to this dutch site:
Edit: also got a note at anandtech.

They tested with an R9 290X and a kaveri APU, difference between directx and mantle in BF4 was 45%. They also state that this percentage is a bit variable, with some set-ups and scenes having a lower boost, but with stuff like crossfire the boost could be well in excess of 45%.

For a game that is supposed to be quite CPU-agnostic(as long as there are atleast 4 threads) thats quite a big boost imho.

 

[Recus]

So Mantle is 290X? Not sure if serious.

Spoiler

6790K with GTX 580.

 

[the54thvoid]

So Mantle is 290X? Not sure if serious.

Spoiler

6790K with GTX 580.

I'm not being funny, what do you mean?

 

[brandonwh64]

So will mantle install like directx does or with the GFX drivers? I still love these slides comparing a I7 IGP to a APU since the I7 was DEF made for GFX performance