AMD CPU speculation... and expert conjecture

[sarinaide]

In other news: Mantle BF4 Frame Pacing, yay!

http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Battlefield-4-Mantle-CrossFire-Early-Performance-FCAT/High-End-C

What is especially worth noting is the consistency of frame times for Mantle, but is doing so at the cost of absolute FPS (observed and real). The same behavior occurs at 1440, even if the two results are now almost identical due to other bottlenecks.

Now for AMD:

Things flip here: Mantle CFX becomes the fastest config, but CFX DX11 becomes the SLOWEST, even slower then a single card running DX11. Things like better at 1440 as the CFX DX11 comes in second, but is closer to the single card solutions then CFX mantle. Likely the same issue as at 1080. Something is clearly mucked with the driver here.

Even more interesting are the AMD vs Intel result sets:

Intel is faster with DX, AMD is faster with Mantle. Which should NOT be the case; both CPU's should benefit to some degree.

I really don't know what to make of these results; they don't make sense on multiple fronts. But combine that with AMD's very poor DX11 results in Thief, we now have TWO games that use Mantle where it looks like AMD's D3D driver is performing worse then it should.

And before someone says "there's the anti-AMD bias again", answer me why AMD's DX11 performance SUCKS in Thief, or why BF4 DX11 CFX results look so poor.

I am not sure if you are complaining or not? My 2cents is I love the flatness of that graph, BF4 must be a dream to play at guaranteed FPS locked without fluctuations, also getting better the longer it goes into gameplay by the rise in FPS. mantle supported drivers will surely increase mantle performance over time but I would like to see other AMD cards and Nvidia cards. I can tell you that my GTX670 is massacred by BF4 with frame variances deemed extreme can go from 120 down to 45 up to 80 that with a Sandy bridge which quite honestly don't believe is the performance penalty. Just for the record my settings are:

24" LED 1080P - Medium textures, High Meshing, High Post Processing, No MSAA or HBAO/SSAO so in terms nothing that kills performance or shouldn't.

 

[logainofhades]

The R9 290/290x and the GTX 780/780ti are probably right on that boundary, if not pushing it. An HD 7970 and GTX 680 can lose 1-2% or so on PCI-E 1.1 vs 2.0 or 3.0. PCI-E 1.1 is still enough for 99% of users though.

 

[gamerk316]

I was simply commenting on how interesting the result set was.

Intel:

Single card: Mantle offers <5% performance gains. Similar FPS/Latency results.
Crossfire: Mantel offers SIGNIFICANTLY less performance then D3D11. Mantle crossfire latency is constant, indicating significant frame pacing is occurring.

Can't really comment on the CFX results due to the frame pacing kicking on, though one wonders why the DX11 mode isn't showing the same behavior in this regard...

AMD:

Single card: Mantle offers <5% performance gains. Similar FPS/Latency results.
Crossfire: Mantle is SIGNIFICANTLY faster; D3D11 result is worse then the single card results

Again, hard to reach a conclusion here, due to the poor showing for the crossfire test. This is almost certainly a driver bug that needs looking into.

What is interesting is the final result, performance wise.
1st: i7-3960x D3D11
2nd: A10-7850k Mantle
3rd: i7-3960x Mantle
4th: A10-7850k D3D11 [Almost certainly a driver problem based on single-card results]

And again, its worth noting that even on the "weak" A10-7850k, the difference between D3D and Mantle is minimal, maybe 10% on average. The Intel result using generic D3D11 is about ~35-40FPS faster on average then the A10-7850k using Mantle.

My main questions:
1: Why does the D3D11 Crossfire result look so bad on AMD CPU's?
2: Why does Intel lose SIGNIFICANT performance when using Mantle? [30 FPS is significant no matter how you spin it]

So I'm seeing a LOT of funnies here that don't really make any sense. I'm *hoping* its just a driver issue and the results will settle on a re-test, but something screwy is going on with the result set here.

 

[Rum]

In other news: Mantle BF4 Frame Pacing, yay!

http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Battlefield-4-Mantle-CrossFire-Early-Performance-FCAT/High-End-C

What is especially worth noting is the consistency of frame times for Mantle, but is doing so at the cost of absolute FPS (observed and real). The same behavior occurs at 1440, even if the two results are now almost identical due to other bottlenecks.

Now for AMD:

Things flip here: Mantle CFX becomes the fastest config, but CFX DX11 becomes the SLOWEST, even slower then a single card running DX11. Things like better at 1440 as the CFX DX11 comes in second, but is closer to the single card solutions then CFX mantle. Likely the same issue as at 1080. Something is clearly mucked with the driver here.

Even more interesting are the AMD vs Intel result sets:

Intel is faster with DX, AMD is faster with Mantle. Which should NOT be the case; both CPU's should benefit to some degree.

I really don't know what to make of these results; they don't make sense on multiple fronts. But combine that with AMD's very poor DX11 results in Thief, we now have TWO games that use Mantle where it looks like AMD's D3D driver is performing worse then it should.

And before someone says "there's the anti-AMD bias again", answer me why AMD's DX11 performance SUCKS in Thief, or why BF4 DX11 CFX results look so poor.

I am not sure if you are complaining or not? My 2cents is I love the flatness of that graph, BF4 must be a dream to play at guaranteed FPS locked without fluctuations, also getting better the longer it goes into gameplay by the rise in FPS. mantle supported drivers will surely increase mantle performance over time but I would like to see other AMD cards and Nvidia cards. I can tell you that my GTX670 is massacred by BF4 with frame variances deemed extreme can go from 120 down to 45 up to 80 that with a Sandy bridge which quite honestly don't believe is the performance penalty. Just for the record my settings are:

24" LED 1080P - Medium textures, High Meshing, High Post Processing, No MSAA or HBAO/SSAO so in terms nothing that kills performance or shouldn't.

That's because it's an AMD product and gamer is a troll/whiner about it. Just wait when DX12 eventually comes out gamer will be proselytizing us all about the wonders of M$ and Nvidia, oh wait... gamer does that already! I'm not sure who is worse, gamer or Juan?

 

[gamerk316]

So...you are arguing that a 20 FPS difference between a single card and a crossfire solution, in favor of the single card, is the correct outcome?

I just call it where I see it. I didn't hesitate to doubt DX11 adoption rates (one of the few times I overly pessimistic), I didn't hesitate to slam Larrabee, and at the end of the day, I was one of the few right from the very beginning about BD's shortcommings. I slam EVERYBODY when I see things go wrong. Its just AMD tends to make more poor decisions then most, in my eyes.

 

[-Fran-]

It's just one game, gamerk. Not taking validity out, but you're generalizing too much out of what a single DEV was able to do with MANTLE in its current state.

Not only AMD is to ask the question of what happened there, but the DEV as well. If you can find that Thief and upcoming titles behave the same, then you'll have a more valid point to make.

I do think that AMD is slacking seriously in their DX optimizing and agree with you there, but that's already something we've known for a good while. Remember that TechReport (and others) made AMD eat their words about pacing of their cards in XFire and DualVGA (APU+GPU).

Before getting something conclusive like "MANTLE sucks" or "MANTLE is the second coming of God", let other games show what their DEVs can do with it.

Cheers!

 

[jimmysmitty]

ASICs are already out for Litecoin. A friend has a few that are running 360KH/s and use way less power than his R9 290s do. As well those same ones will be updated to push out 16MH/s, qhich is about 16x faster than a R9 290X can even push out, and soon a 100MH/s one will hit.

I don't think they'll be that common nor cheap. Making a chip that can do the computation isn't hard, it's keeping it fed with a high speed memory bus that forces the costs up. The LTC algorithm was design's specifically to require large amounts of memory bandwidth as a way to prevent cheap custom chips from doing the computation entirely inside local registers.

http://www.gawminers.com/300-kh-s-single-gridseed-asic-miner/

My friend has 4 of them. That is $190 for 300KH/s, although he was able to push 360KH/s.

As well, look up the miners. There are ones at 16MH/s and even the creator behind Litecoin are not going to do anything to stop it.

I am not sure if these ASICs are pushing pure memory bandwidth or what as I am not that into it but it is still happening and they will take over.

Sure the 100MH/s is $10,000 but in order to get the same from a R9 290X you need 100 of them which ends up costing at best $55,000 dollars not including the PSUs needed, the fact that you would need multiple 15A/20A lines, motherboards, PCIe converters ect.

ASICs will keep taking over. I am just glad as it will let the GPUs settle in price for a bit.

It's just one game, gamerk. Not taking validity out, but you're generalizing too much out of what a single DEV was able to do with MANTLE in its current state.

Not only AMD is to ask the question of what happened there, but the DEV as well. If you can find that Thief and upcoming titles behave the same, then you'll have a more valid point to make.

I do think that AMD is slacking seriously in their DX optimizing and agree with you there, but that's already something we've known for a good while. Remember that TechReport (and others) made AMD eat their words about pacing of their cards in XFire and DualVGA (APU+GPU).

Before getting something conclusive like "MANTLE sucks" or "MANTLE is the second coming of God", let other games show what their DEVs can do with it.

Cheers!

My honest opinion is that Mantle wont be around very long. When devs have the access to the same ability in something they have used for 20+ years, they will jump back to it and Mantle will go on the side of the road. AMD does not have the R&D money to spend like MS does on something like DX.

And apart from what others say, I think DX 12 was in development before we found out about Mantle. Only because I don't think Microsoft will ever stop developing it for PC. Even with the changes to 8/8.1 those apps still can use DirectX so why stop at 11 if the apps could use 12 and get more performance? It would help sell tablets and such.

People just instantly jump on the "MS is copying blah blah" band wagon when in reality these things have been around forever. We are just now seeing them re-implemented in a way that wont BSoD our PCs every time the game takes a dump.

 

[juanrga]

Why is there a limit to PCIe if they wish to continue it? At the phyical level (I/O) they're exactly the same, thus why NVLink can plug right into PCIe slots.

PCIe 4.0 is 16GT/s
PCIe 5.0 would be 32GT/s
PCIe 6.0 would be 64GT/s

Which means NVLink 2.0 (50GT/s) sits between PCIe 5 and 6.

You say it's bringing competition but it's really just more exclusivity like PhysX. I doubt Intel or AMD will be making NVLink ports in any processors. And I doubt anyone besides NVidia will be making NVLink GPUs.

PCIe 5 and 6 doesn't exist. Only 4.0 is under development.

The links given before explain why PCIe cannot be scaled up at the NVLINK level: energy data movement, point-to-point, and signal integrity.

NVLINK is bringing competition to HPC market, because now AMD and Intel have to offer alternative interconnects or, better, accelerate their APU plans.

People, single cards don't even max out PCI-E 1.1 x16 yet; why are we talking about PCI-E bandwidth like its a massive limiting factor here?

PCIe bottlenecks is one of main factors behind current stagnant status of heterogeneous computing. It has a measurable effect in performance.

 

[juanrga]

On the GPU side I mentioned 20-40% which is more or less what AMD achieved from Richland to Kavari, maybe if its DDR4 bandwidth it is possible to see higher with some bandwidth opening up.

X86 side kavari at lower clocks was either better or more or less the same as Richland clocked higher so 10% doesn't seem like that much of an ask.

Gaming wise maybe 10%-15% combined performance over Kavari, depending on the game that may be lots or a little.

DDR4 official bandwidth will start at 2133 MHz. You can purchase faster 3000MHz DDR3 DIMMs today. I don't expect 4266 MHz DDR4 before 2016, which is too late for Carrizo.

 

[con635]

@gamerk: what you said seems plausible and the timeline kinda sorta fits. what doesn't fit is the boldness from amd. amd have been playing it safe since 2011. what would corner them so much to pull off a stunt like this now? imo everything amd tried to do with mantle hype got mostly undone by the mining craze. amd did make money, but it was for short term unlike what mantle-hype would have gained them. and the way developers and amd are handling mantle's software support is simply subpar and negligent.

It's just one game, gamerk. Not taking validity out, but you're generalizing too much out of what a single DEV was able to do with MANTLE in its current state.

Not only AMD is to ask the question of what happened there, but the DEV as well. If you can find that Thief and upcoming titles behave the same, then you'll have a more valid point to make.

I do think that AMD is slacking seriously in their DX optimizing and agree with you there, but that's already something we've known for a good while. Remember that TechReport (and others) made AMD eat their words about pacing of their cards in XFire and DualVGA (APU+GPU).

Before getting something conclusive like "MANTLE sucks" or "MANTLE is the second coming of God", let other games show what their DEVs can do with it.

Cheers!

Agree and we were told mantle can do away with afr altogether and let a multiple gpu config be seen as one big gpu so lets see what happens in the future, mantle has a good head start over dx12.

Can anyone point me to a review which compares in depth different cpus on mantle? All Ive seen in any depth is gpu comparisons, how about some athlon/apu vs fx vs intel etc in some sort of depth from toms?

 

[gamerk316]

PCIe bottlenecks is one of main factors behind current stagnant status of heterogeneous computing. It has a measurable effect in performance.

Provably false:

http://www.pugetsystems.com/labs/articles/Impact-of-PCI-E-Speed-on-Gaming-Performance-518/

And I had one a few days ago with 1.1 thrown in, but I can't find the bloody link.

Now, if you attempted to do lots of really small memory reads, yes, you'd lose a ton of performance. PCI-E works best when you do a very large data transfer and cache as much data as possible (hence, GPU's with 4GB VRAM installed). But like all external devices, latency is a problem. And if you have to do a lot of transfers, you performance will tank as your co-processor does work faster then you can pass over the bus. Increasing bandwidth hides this problem (larger transfers means you can do fewer of them, thus increasing performance due to less waiting), but doesn't solve the underlying problem of attempting to do a lot of reads across a bus that has high latency (relative to the rest of the system). That is ALWAYS a poor design when going over an external bus.

 

[jdwii]

On the GPU side I mentioned 20-40% which is more or less what AMD achieved from Richland to Kavari, maybe if its DDR4 bandwidth it is possible to see higher with some bandwidth opening up.

X86 side kavari at lower clocks was either better or more or less the same as Richland clocked higher so 10% doesn't seem like that much of an ask.

Gaming wise maybe 10%-15% combined performance over Kavari, depending on the game that may be lots or a little.

The thing is though Kavari was a big step, they went to GCN and they switched to steamroller. However with this next gen APU Amd is aiming at a 65watt TDP standpoint which seems almost impossible being that this APU is being built on the same 28nm design that's already pretty mature, now if they achieve their goals i don't expect any improvements on the performance side except minimal changes. Pretty sure they got the mindset where they want this module design to be over with so they can come out with the next big thing and that's probably what Jim Keller is working on with his team.
Maybe part might come in here. Maybe Jim Keller been working on excavator and pushing the CPU design as far as possible and also lets hope Amd includes GCN 1.1 for their next APU instead of 1.0.

 

[8350rocks]

The reason for MANTLE coming around is because Devs are tired of being tied to windows...myself included. MANTLE opens platforms to, yes, a specific architecture, but it allows them to port windows games to OSX/Linux via GCN cards. THAT is the point you are all missing. BTW, the work to port DX11 code to MANTLE literally takes 30 minutes...seriously. So, it is not much additional work at all to expose yourself to those other platforms that are literally almost untapped for even double A titles, much less triple A titles.

MANTLE is going nowhere...I could not care less what M$ does with DX12...I am not tying games to windows anymore, and MANTLE allows me to do that. Why do you think CryEngine got on board? EA? Square Enix/Eidos? The same reasons...they can back port those games when cross platform support comes.

The CPU overhead BS is the public marketing hype...the dev hype is "YES!!!! We can FINALLY be RID of M$ and Windows!!!!"

 

[con635]

For anyone interested I've had a play with star swarm demo, my hardware is pretty low end, turbo disabled 4ghz 4 core trinity and hd 7770. Dont think its really relevant but also 8gb ram, a55 asrock mobo and ssd.

Test settings are 'follow' at preset 'medium' settings, 360 second timed run.

D3D
Test Duration: 360 Seconds
Total Frames: 9508

Average FPS: 26.41
Average Unit Count: 4049
Maximum Unit Count: 5221
Average Batches/MS: 340.26
Maximum Batches/MS: 782.64
Average Batch Count: 14417
Maximum Batch Count: 80908

Mantle
Test Duration: 360 Seconds
Total Frames: 11765

Average FPS: 32.68
Average Unit Count: 4103
Maximum Unit Count: 5531
Average Batches/MS: 378.91
Maximum Batches/MS: 1554.99
Average Batch Count: 12038
Maximum Batch Count: 66758

Only 6fps more your thinking? Doesn't tell half the story which sort of makes me think review sites aren't either, on dx the fps dips to just 7 when there's lots of units on screen and if it was a game I'd say its unplayable imo. On mantle it dips to 24fps but remains smooth enough for an rts game.

Heres whats really interesting though, I fired up afterburner beforehand (unfortunately didn't actually save the log 🙁 maybe next time) and how mantle uses my hw looks phenomenal, in d3d the gpu usage is all over the place, in the places I'm sure were the 7 fps dips I'm actually cpu bottle necked, looking at the graphs of the cpu usage one above the other the core usage is all over the place, 2 hover 50% 2 hit 90 odds and the gpu drops.
On mantle my gpu stays at 98% for the entire test, the cores are all used evenly the graphs for each core look strangely similar stacked one above the other, the individual core usage is between 30 and 50% ie nowhere near a bottle neck.
vram usage is maxed on both d3d and mantle but mantle uses 1gb more system ram than d3d.
If anyone wants me to try any settings let me know, I'll try get a log of hw activity next time as well.

Edit, when I think about it, my d3d ss cpu usage is strangely similar to when I play arma/dayz, how could ms get away this rubbish for so long???

 

[logainofhades]

Open platform for games would be awesome. Games are the only reason I never took the time to learn Linux. I would gladly make the switch and learn if it meant better software support.

 

[con635]

The reason for MANTLE coming around is because Devs are tired of being tied to windows...myself included. MANTLE opens platforms to, yes, a specific architecture, but it allows them to port windows games to OSX/Linux via GCN cards. THAT is the point you are all missing. BTW, the work to port DX11 code to MANTLE literally takes 30 minutes...seriously. So, it is not much additional work at all to expose yourself to those other platforms that are literally almost untapped for even double A titles, much less triple A titles.

MANTLE is going nowhere...I could not care less what M$ does with DX12...I am not tying games to windows anymore, and MANTLE allows me to do that. Why do you think CryEngine got on board? EA? Square Enix/Eidos? The same reasons...they can back port those games when cross platform support comes.

The CPU overhead BS is the public marketing hype...the dev hype is "YES!!!! We can FINALLY be RID of M$ and Windows!!!!"

I dont think its just that, have you played with star swarm? I'd love to see the results of your rig and what afterburner says, see my above post

 

[de5_Roy]

Open platform for games would be awesome. Games are the only reason I never took the time to learn Linux. I would gladly make the switch and learn if it meant better software support.

yet amd keeps stating that mantle support will be tied to windows and gcn for now, with every catalyst release. if they really wanted to make an impact with linux, they woulda started with windows and linux, then android and (or only) steamos subsequently. so all that hype about "open" is, as of now, just hype.

edit: dx12 won't be any better. not due to the api, but due to microsoft's lock-out policies. might force o.s. upgrade onto users.

 

[anxiousinfusion]

Amd includes GCN 1.1 for their next APU instead of 1.0.

I was under the impression that GCN 1.1 had already been implemented with Kaveri parts.

 

[Reynod]

Some good links and comparative insight here guys ... keep it coming.

I'll try to get you all a "Mantle Badge"

 

[jimmysmitty]

The reason for MANTLE coming around is because Devs are tired of being tied to windows...myself included. MANTLE opens platforms to, yes, a specific architecture, but it allows them to port windows games to OSX/Linux via GCN cards. THAT is the point you are all missing. BTW, the work to port DX11 code to MANTLE literally takes 30 minutes...seriously. So, it is not much additional work at all to expose yourself to those other platforms that are literally almost untapped for even double A titles, much less triple A titles.

MANTLE is going nowhere...I could not care less what M$ does with DX12...I am not tying games to windows anymore, and MANTLE allows me to do that. Why do you think CryEngine got on board? EA? Square Enix/Eidos? The same reasons...they can back port those games when cross platform support comes.

The CPU overhead BS is the public marketing hype...the dev hype is "YES!!!! We can FINALLY be RID of M$ and Windows!!!!"

Why would devs want to move away from the main platform for consumers? Games are still not a major market for PCs in comparison to total sales but still game devs wanting to move off to another platform where sales would be only a small few and would also lose out on those other PC sales?

Unless other platforms get the support Windows has, I don't see anything taking over. Even Android is too segregated and still a weak OS in comparison.

For anyone interested I've had a play with star swarm demo, my hardware is pretty low end, turbo disabled 4ghz 4 core trinity and hd 7770. Dont think its really relevant but also 8gb ram, a55 asrock mobo and ssd.

Test settings are 'follow' at preset 'medium' settings, 360 second timed run.

D3D
Test Duration: 360 Seconds
Total Frames: 9508

Average FPS: 26.41
Average Unit Count: 4049
Maximum Unit Count: 5221
Average Batches/MS: 340.26
Maximum Batches/MS: 782.64
Average Batch Count: 14417
Maximum Batch Count: 80908

Mantle
Test Duration: 360 Seconds
Total Frames: 11765

Average FPS: 32.68
Average Unit Count: 4103
Maximum Unit Count: 5531
Average Batches/MS: 378.91
Maximum Batches/MS: 1554.99
Average Batch Count: 12038
Maximum Batch Count: 66758

Only 6fps more your thinking? Doesn't tell half the story which sort of makes me think review sites aren't either, on dx the fps dips to just 7 when there's lots of units on screen and if it was a game I'd say its unplayable imo. On mantle it dips to 24fps but remains smooth enough for an rts game.

Heres whats really interesting though, I fired up afterburner beforehand (unfortunately didn't actually save the log 🙁 maybe next time) and how mantle uses my hw looks phenomenal, in d3d the gpu usage is all over the place, in the places I'm sure were the 7 fps dips I'm actually cpu bottle necked, looking at the graphs of the cpu usage one above the other the core usage is all over the place, 2 hover 50% 2 hit 90 odds and the gpu drops.
On mantle my gpu stays at 98% for the entire test, the cores are all used evenly the graphs for each core look strangely similar stacked one above the other, the individual core usage is between 30 and 50% ie nowhere near a bottle neck.
vram usage is maxed on both d3d and mantle but mantle uses 1gb more system ram than d3d.
If anyone wants me to try any settings let me know, I'll try get a log of hw activity next time as well.

Edit, when I think about it, my d3d ss cpu usage is strangely similar to when I play arma/dayz, how could ms get away this rubbish for so long???

The whole purpose to Mantle is to help alleviate the CPU overhead. We had similar abilities in the old days but before D3D became the main, we also had applications and games causing the OS to crash.

This was the purpose to APIs, to allow for applications and drivers to crash and not take the OS with it. Now when you get a BSoD there is only one of two reasons, either a bad piece of hardware or possible OS corruption.

Without the APIs we would have all kinds of crashes and fun times.

And yes, it sucks but honestly it is also due to developers. The developers have been mainly focusing on consoles (most). That means PC hardware has advanced and not been used like it should have.

Maybe with DX12 and the new consoles we wont have that anymore.

Open platform for games would be awesome. Games are the only reason I never took the time to learn Linux. I would gladly make the switch and learn if it meant better software support.

Open platform has its downsides too.

Open platform for games would be awesome. Games are the only reason I never took the time to learn Linux. I would gladly make the switch and learn if it meant better software support.

yet amd keeps stating that mantle support will be tied to windows and gcn for now, with every catalyst release. if they really wanted to make an impact with linux, they woulda started with windows and linux, then android and (or only) steamos subsequently. so all that hype about "open" is, as of now, just hype.

edit: dx12 won't be any better. not due to the api, but due to microsoft's lock-out policies. might force o.s. upgrade onto users.

We shall see. If it comes out soon, I am willing to bet it will push to 7. Mostly because 8/8.1 adoption rates wont pick up fast enough.

We will see at some point but it really depends on adoption rates and if there is any platform updates needed.

Some good links and comparative insight here guys ... keep it coming.

I'll try to get you all a "Mantle Badge"

Can it be some sort of volcanic magma like badge that burns through the spam?

 

[juanrga]

PCIe bottlenecks is one of main factors behind current stagnant status of heterogeneous computing. It has a measurable effect in performance.

Provably false:

http://www.pugetsystems.com/labs/articles/Impact-of-PCI-E-Speed-on-Gaming-Performance-518/

I emphasized the part of my post that either you didn't read or didn't understand.

Here one work comparing compute abilities of APU vs discrete GPU over PCie:

The graphics processing unit (GPU) has made significant strides as an accelerator in parallel computing. However, because the GPU has resided out on PCIe as a discrete device, the performance of GPU applications can be bottlenecked by data transfers between the CPU and GPU over PCIe. Emerging heterogeneous computing architectures that "fuse" the functionality of the CPU and GPU, e.g., AMD Fusion and Intel Knights Ferry, hold the promise of addressing the PCIe bottleneck. In this paper, we empirically characterize and analyze the efficacy of AMD Fusion, an architecture that combines general-purpose x86 cores and programmable accelerator cores on the same silicon die. We characterize its performance via a set of micro-benchmarks (e.g., PCIe data transfer), kernel benchmarks(e.g., reduction), and actual applications (e.g., molecular dynamics). Depending on the benchmark, our results show that Fusion produces a 1.7 to 6.0-fold improvement in the data-transfer time, when compared to a discrete GPU. In turn, this improvement in data-transfer performance can significantly enhance application performance. For example, running a reduction benchmark on AMD Fusion with its mere 80 GPU cores improves performance by 3.5-fold over the discrete AMD Radeon HD 5870 GPU with its 1600 more powerful GPU cores.

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6031577&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D6031577

For anyone interested I've had a play with star swarm demo, my hardware is pretty low end, turbo disabled 4ghz 4 core trinity and hd 7770. Dont think its really relevant but also 8gb ram, a55 asrock mobo and ssd.

Test settings are 'follow' at preset 'medium' settings, 360 second timed run.

D3D
Test Duration: 360 Seconds
Total Frames: 9508

Average FPS: 26.41
Average Unit Count: 4049
Maximum Unit Count: 5221
Average Batches/MS: 340.26
Maximum Batches/MS: 782.64
Average Batch Count: 14417
Maximum Batch Count: 80908

Mantle
Test Duration: 360 Seconds
Total Frames: 11765

Average FPS: 32.68
Average Unit Count: 4103
Maximum Unit Count: 5531
Average Batches/MS: 378.91
Maximum Batches/MS: 1554.99
Average Batch Count: 12038
Maximum Batch Count: 66758

Only 6fps more your thinking? Doesn't tell half the story which sort of makes me think review sites aren't either, on dx the fps dips to just 7 when there's lots of units on screen and if it was a game I'd say its unplayable imo. On mantle it dips to 24fps but remains smooth enough for an rts game.

Yes average FPS only tells half the history, but even for average framerates, the increase is very important: those 6FPS more represent a 24% more performance.

 

[juanrga]

Some good links and comparative insight here guys ... keep it coming.

I'll try to get you all a "Mantle Badge"

I want the mine spelled as "mantel".

Spoiler

The reason for MANTLE coming around is because Devs are tired of being tied to windows...myself included. MANTLE opens platforms to, yes, a specific architecture, but it allows them to port windows games to OSX/Linux via GCN cards. THAT is the point you are all missing. BTW, the work to port DX11 code to MANTLE literally takes 30 minutes...seriously. So, it is not much additional work at all to expose yourself to those other platforms that are literally almost untapped for even double A titles, much less triple A titles.

MANTLE is going nowhere...I could not care less what M$ does with DX12...I am not tying games to windows anymore, and MANTLE allows me to do that. Why do you think CryEngine got on board? EA? Square Enix/Eidos? The same reasons...they can back port those games when cross platform support comes.

The CPU overhead BS is the public marketing hype...the dev hype is "YES!!!! We can FINALLY be RID of M$ and Windows!!!!"

Why would devs want to move away from the main platform for consumers? Games are still not a major market for PCs in comparison to total sales but still game devs wanting to move off to another platform where sales would be only a small few and would also lose out on those other PC sales?

Unless other platforms get the support Windows has, I don't see anything taking over. Even Android is too segregated and still a weak OS in comparison.

Windows is not the main customers platform anymore. I believe Toms had a article about that. Only about 25% of share is for windows. Linux and Apple are the rest.

In the second place, it was already explained that Microsoft recent strategy away from traditional PCs, toward tablets and Xbox1, motivated PC game developers to either (i) switch to linux or (ii) develop MANTLE in collaboration with AMD.

I already provided links to game developers confirming that Microsoft didn't have any plan for DX12 before MANTLE was presented.

Game developers that switched to linux also gave talks explaining why they were disappointed by Microsoft new strategy

http://www.zdnet.com/valve-windows-8-is-a-catastrophe-for-pcs-7000001634/

http://www.extremetech.com/gaming/167253-gabe-newell-made-windows-a-viable-gaming-platform-and-linux-is-next

 

[gamerk316]

Some good links and comparative insight here guys ... keep it coming.

I'll try to get you all a "Mantle Badge"

I want the mine spelled as "mantel".

Spoiler

The reason for MANTLE coming around is because Devs are tired of being tied to windows...myself included. MANTLE opens platforms to, yes, a specific architecture, but it allows them to port windows games to OSX/Linux via GCN cards. THAT is the point you are all missing. BTW, the work to port DX11 code to MANTLE literally takes 30 minutes...seriously. So, it is not much additional work at all to expose yourself to those other platforms that are literally almost untapped for even double A titles, much less triple A titles.

MANTLE is going nowhere...I could not care less what M$ does with DX12...I am not tying games to windows anymore, and MANTLE allows me to do that. Why do you think CryEngine got on board? EA? Square Enix/Eidos? The same reasons...they can back port those games when cross platform support comes.

The CPU overhead BS is the public marketing hype...the dev hype is "YES!!!! We can FINALLY be RID of M$ and Windows!!!!"

Why would devs want to move away from the main platform for consumers? Games are still not a major market for PCs in comparison to total sales but still game devs wanting to move off to another platform where sales would be only a small few and would also lose out on those other PC sales?

Unless other platforms get the support Windows has, I don't see anything taking over. Even Android is too segregated and still a weak OS in comparison.

Windows is not the main customers platform anymore. I believe Toms had a article about that. Only about 25% of share is for windows. Linux and Apple are the rest.

In the second place, it was already explained that Microsoft recent strategy away from traditional PCs, toward tablets and Xbox1, motivated PC game developers to either (i) switch to linux or (ii) develop MANTLE in collaboration with AMD.

I already provided links to game developers confirming that Microsoft didn't have any plan for DX12 before MANTLE was presented.

Game developers that switched to linux also gave talks explaining why they were disappointed by Microsoft new strategy

http://www.zdnet.com/valve-windows-8-is-a-catastrophe-for-pcs-7000001634/

http://www.extremetech.com/gaming/167253-gabe-newell-made-windows-a-viable-gaming-platform-and-linux-is-next

*sigh* Are you REALLY going to count mobile share when it comes to mobile platform.

Secondly: Mantle is currently only running on Windows.

Thirdly, Devs typically aren't told about major API changes until AFTER the API is finalized and released, which didn't happen until GDC last month. So of course they didn't know about it.

Fourthly, it looks like either 8.2 or Win 9 (however MSFT decides to package it) is bringing back the start menu, so Linux missed a chance to grow on one of MSFT's screw ups (again).

 

[juanrga]

Unreal ported to linux

https://www.unrealengine.com/blog/41-update-preview

 

[juanrga]

*sigh* Are you REALLY going to count mobile share when it comes to mobile platform.

The more popular laptop on Amazon doesn't run Windows. No sure if you want to count that as mobile, but in any case it is irrelevant to my former point.

Secondly: Mantle is currently only running on Windows.

On Windows 7, which is more than DX12 can do today. Moreover, there is plans to port MANTLE outside Windows.

Thirdly, Devs typically aren't told about major API changes until AFTER the API is finalized and released, which didn't happen until GDC last month. So of course they didn't know about it.

Developers showed surprise when asked about DX12, because they would be informed about its existence at least, even if not knowing all the details. Also don't forget Nvidia now pretending that "everyone" was informed about DX12 since first day.

Fourthly, it looks like either 8.2 or Win 9 (however MSFT decides to package it) is bringing back the start menu, so Linux missed a chance to grow on one of MSFT's screw ups (again).

Start menu is only one of the problems.