AMD CPU speculation... and expert conjecture

[juanrga]

I think Mantle will be useful only when they start to support lower end GPUs and it benefits them. If you get 120FPS in DX11, another 5-10FPS means nothing.

Current games have been developed with the DX overhead in mind. Thus the MANTLE version only provides about 20-60% improvement. MANTLE true potential will be on next gen games, which will increase the number of draw calls by about 10x. This is what the Oxide demo on Kaveri was about

 

[gamerk316]

thief mantle benchmarks Thief Patch Enables TrueAudio And Mantle: First Benchmarks
http://www.tomshardware.com/reviews/thief-mantle-benchmarks,3773.html
http://www.anandtech.com/show/7868/evaluating-amds-trueaudio-and-mantle-thief

We see the expected (and predicted!) behavior of worse CPU's seeing better gains, and gains being muted on weaker GPU's. Mantles performance benefit seems best on low-CPU/high-GPU systems.

So again: I REALLY would like i3 benches, since i3's would figure to gain the most out of Mantle. You could make the argument an i3 is already a better buy then a 4170; how does Mantle affect the math? I'd REALLY like review sites to start doing due diligence on these types of benches...*cough*

 

[gamerk316]

I think Mantle will be useful only when they start to support lower end GPUs and it benefits them. If you get 120FPS in DX11, another 5-10FPS means nothing.

Current games have been developed with the DX overhead in mind. Thus the MANTLE version only provides about 20-60% improvement. MANTLE true potential will be on next gen games, which will increase the number of draw calls by about 10x. This is what the Oxide demo on Kaveri was about

Still haven't learned you lesson have you? As I've stated before: Draw calls are not a major bottleneck to GPU performance. And there are ways to get around that (doing a lot of work over a few calls, instead of doing a little work on many).

Sums up the problem better then I can: You can increase on part of performance under the sun by 1000000x; but if that part isn't a major driver of performance, you gain almost nothing. Draw calls are not a major performance factor.

 

[de5_Roy]

asrock's am1 mothebroards... with laptop power connector option!
http://www.anandtech.com/show/7869/asrocks-am1-kabini-motherboards-announced-including-19v-dcin-model
quite possibly the most interesting thing i've seen on a motherboard this year.
and i've the m2 slots and sas ports on haswell's 97 chipset motherboards!

AM1 parts available in parts of Euroland
http://www.fudzilla.com/home/item/34230-am1-parts-available-in-parts-of-euroland

AMD and Edios launch the Mantle version Thief
http://semiaccurate.com/2014/03/18/amd-edios-launch-mantle-version-thief/

the company announced that Mantle will remain proprietary for the time being, but that they are working to get a public specification for Mantle or a Mantle-like API out before the end of the year. As far as multiOS support is concerned AMD says that Mantle will continue to remain Windows only for now.

HSA’s next big moves
http://vr-zone.com/articles/hsa-mediatek-interview/74251.html

ASML pulls plug on new generation of computer chips
http://www.fudzilla.com/home/item/34236-asml-pulls-plug-on-new-generation-of-computer-chips

 

[jdwii]

thief mantle benchmarks Thief Patch Enables TrueAudio And Mantle: First Benchmarks
http://www.tomshardware.com/reviews/thief-mantle-benchmarks,3773.html
http://www.anandtech.com/show/7868/evaluating-amds-trueaudio-and-mantle-thief

htl's kaveri reivew
http://www.hitechlegion.com/reviews/processors/38897-amd-a10-7850k-kaveri-quad-core-apu-review

ht4u's (german)
http://ht4u.net/reviews/2014/amd_kaveri_apus_im_test/

AMD Catalyst™ 14.3 Beta V1.0 Driver
http://support.amd.com/en-us/kb-articles/Pages/latest-catalyst-windows-beta.aspx
oh enduro T_T

Ah those horrible scores for the dGPU makes me sick on that site, we really seeing horrible performance per dollar here this has to be a first for Amd except with their dGPU's which isn't really their fault.

 

[jdwii]

I think Mantle will be useful only when they start to support lower end GPUs and it benefits them. If you get 120FPS in DX11, another 5-10FPS means nothing.

Current games have been developed with the DX overhead in mind. Thus the MANTLE version only provides about 20-60% improvement. MANTLE true potential will be on next gen games, which will increase the number of draw calls by about 10x. This is what the Oxide demo on Kaveri was about

Amd's marketing slides are always cute

 

[jdwii]

thief mantle benchmarks Thief Patch Enables TrueAudio And Mantle: First Benchmarks
http://www.tomshardware.com/reviews/thief-mantle-benchmarks,3773.html
http://www.anandtech.com/show/7868/evaluating-amds-trueaudio-and-mantle-thief

We see the expected (and predicted!) behavior of worse CPU's seeing better gains, and gains being muted on weaker GPU's. Mantles performance benefit seems best on low-CPU/high-GPU systems.

So again: I REALLY would like i3 benches, since i3's would figure to gain the most out of Mantle. You could make the argument an i3 is already a better buy then a 4170; how does Mantle affect the math? I'd REALLY like review sites to start doing due diligence on these types of benches...*cough*

What really concerns me is that celeron getting basicly even performance compared to the A10, wonder what a 60$ pentium would look like. Actually i did build some pentium haswell dual cores and i put the 7770ghz in them i should actually buy a A10 and benchmark the two. However if i can't resell the machines what's the point.

 

[juanrga]

We see the expected (and predicted!) behavior of worse CPU's seeing better gains, and gains being muted on weaker GPU's. Mantles performance benefit seems best on low-CPU/high-GPU systems.

MANTLE main goal is to eliminate CPU bottleneck associated to the DX bloatware. Therefore you will see benefits on CPU bottleneck situations and almost no improvement in GPU bottleneck situations. Is not evident?

CPU bottleneck is not synonym for low-CPU. The typical situation will involve a low-CPU and a medium/high-GPU, but you can achieve a CPU bottleneck situation also with a top-end extreme i7 CPU. You only need to match the $1000 CPU with two or three 290X in crossfire and again MANTLE will brings evident performance benefits: ~50% more framerate.

 

[juanrga]

asrock's am1 mothebroards... with laptop power connector option!
http://www.anandtech.com/show/7869/asrocks-am1-kabini-motherboards-announced-including-19v-dcin-model
quite possibly the most interesting thing i've seen on a motherboard this year.
and i've the m2 slots and sas ports on haswell's 97 chipset motherboards!

Nice feature, but it is not anything new. DC-In power connectors have been available for years in thin mini-itx mobos.

HSA’s next big moves
http://vr-zone.com/articles/hsa-mediatek-interview/74251.html

Interesting reading, specially the part that mentions one part of Nvidia tech want to embrace HSA.

 

[jimmysmitty]

thief mantle benchmarks Thief Patch Enables TrueAudio And Mantle: First Benchmarks
http://www.tomshardware.com/reviews/thief-mantle-benchmarks,3773.html
http://www.anandtech.com/show/7868/evaluating-amds-trueaudio-and-mantle-thief

htl's kaveri reivew
http://www.hitechlegion.com/reviews/processors/38897-amd-a10-7850k-kaveri-quad-core-apu-review

ht4u's (german)
http://ht4u.net/reviews/2014/amd_kaveri_apus_im_test/

AMD Catalyst™ 14.3 Beta V1.0 Driver
http://support.amd.com/en-us/kb-articles/Pages/latest-catalyst-windows-beta.aspx
oh enduro T_T

Thief didn't show anything other than what we know and it also showed that the FX CPUs, even the highest end one seems to bottleneck GPUs.

I think Mantle will be useful only when they start to support lower end GPUs and it benefits them. If you get 120FPS in DX11, another 5-10FPS means nothing.

Current games have been developed with the DX overhead in mind. Thus the MANTLE version only provides about 20-60% improvement. MANTLE true potential will be on next gen games, which will increase the number of draw calls by about 10x. This is what the Oxide demo on Kaveri was about

Ahhhhh good old AMD marketing. See slide, it means truth.

I think Mantle has potential but it will only last if it benefits everything not just certain GPUs.

As I said, if DX and OGL get the same feature Mantle will be rendered pretty useless.

We see the expected (and predicted!) behavior of worse CPU's seeing better gains, and gains being muted on weaker GPU's. Mantles performance benefit seems best on low-CPU/high-GPU systems.

MANTLE main goal is to eliminate CPU bottleneck associated to the DX bloatware. Therefore you will see benefits on CPU bottleneck situations and almost no improvement in GPU bottleneck situations. Is not evident?

CPU bottleneck is not synonym for low-CPU. The typical situation will involve a low-CPU and a medium/high-GPU, but you can achieve a CPU bottleneck situation also with a top-end extreme i7 CPU. You only need to match the $1000 CPU with two or three 290X in crossfire and again MANTLE will brings evident performance benefits: ~50% more framerate.

I will wait till I see that in action. It is CPU bottlenecks that will call for that sort of ability.

It will probably also come in DX12 as I see that as being one of the only ways to really get benefits in performance with a API rather than just the GPU upgrades.

 

[eidolon171]

I have an a10 7850k and an R7 260x. Full Specs http://pcpartpicker.com/user/eidolon171/saved/3FgD
These are some of the things I have tested out on my rig.

FRAPS first...

Metro Last Night (medium settings) (igpu only) = 24fps
Metro Last Night (medium settings) (R7 260x) = 52fps

Hitman: Absolution (medium settings) (igpu only) = 35fps
Hitman: Absolution (medium settings) (R7 260x) = 47fps

Tomb Raider (ultra settings) (igpu only) = 30fps
Tomb Raider (ultra settings) (R7 260x) = 68fps

Batman: Arkham Origins (ultra settings) (igpu only) = 36fps
Batman: Arkham Origins (ultra settings) (R7 260x) = 70fps

Do you have minimum fps as well?

Passmark (64bit) CPU scores
http://postimg.org/image/5vj8tixj9/

Glad to see it in at the same level than i5-2500k, as predicted, and above i3 Haswell and FX-4350.

in synthetic benchmarks

To juanrga: minimum fps on the selected titles is 70% of average with the 260x and 60% of the average with only the igpu.
To jdwii: No those are not synthetics. They are real-world benchmarks taken from actual users who have uploaded their results to PCMark.

 

[jdwii]

PCmark is a synthetic benchmark.

 

[de5_Roy]

Thief didn't show anything other than what we know and it also showed that the FX CPUs, even the highest end one seems to bottleneck GPUs.

using a top card like 780ti or r9 290x at 1080p resolution will lessen gpu bottleneck in current games to the point where the bottleneck will shift to the cpu, regardless of brand. that's why, with directx, intel shows better performance while mantle lessens the gap.

a balanced build will make mantle's use moot for it's current purpose of reducing cpu overhead.

edit2: apparently microsoft has adopted mantle/as a part of directx12. LOL

 

[juanrga]

PCmark is a synthetic benchmark.

Which reproduces the average performance of real world applications, like several posters noted before...

 

[juanrga]

First hints Microsoft would replicate MANTLE to remain competitive

http://www.extremetech.com/gaming/177407-microsoft-hints-that-directx-12-will-imitate-and-destroy-amds-mantle

Then official announcements that AMD is partnership with Microsoft to develop "the infrastructure to render next-generation graphics workloads at full performance."

http://schedule.gdconf.com/session-id/828412

Finally confirmation Microsoft is adopting MANTLE and renaming it to DX12

http://semiaccurate.com/2014/03/18/microsoft-adopts-mantle-calls-dx12/

Also interesting:

http://www.brightsideofnews.com/news/2014/3/18/exclusive-amds-carrizo-apu-details-leaked---soc-for-the-mainstream.aspx

 

[GOM3RPLY3R]

I think Mantle will be useful only when they start to support lower end GPUs and it benefits them. If you get 120FPS in DX11, another 5-10FPS means nothing.

Current games have been developed with the DX overhead in mind. Thus the MANTLE version only provides about 20-60% improvement. MANTLE true potential will be on next gen games, which will increase the number of draw calls by about 10x. This is what the Oxide demo on Kaveri was about

Still haven't learned you lesson have you? As I've stated before: Draw calls are not a major bottleneck to GPU performance. And there are ways to get around that (doing a lot of work over a few calls, instead of doing a little work on many).

Sums up the problem better then I can: You can increase on part of performance under the sun by 1000000x; but if that part isn't a major driver of performance, you gain almost nothing. Draw calls are not a major performance factor.

I'm sorry for abruptly jumping in here, but I've been curious about Mantle. I've watched a couple of videos from real world testers (Linus, ToT, etc.), and I've found that from the videos and articles read, Mantle has been mostly pulling crap performance. Also the crap is better on Nvidia than the crap of AMD. I know it's still kinda new, and there have been improvement, but from the charts showing Mantles amazing performances seem to be from AMD or endorsed by AMD.

I'm not trying to pull sides, I'm just wondering why all I see is that insanely good performance and then the really crappy performances from real world testers.

 

[vmN]

One word: Marketing.

 

[gamerk316]

http://arstechnica.com/science/2014/03/new-cache-design-speeds-up-processing-time-by-15/

Think AMD should get on this, ASAP.

 

[Cazalan]

Sounds like AMD will be getting more money from Microsoft. Great for them!

 

[jdwii]

PCmark is a synthetic benchmark.

Which reproduces the average performance of real world applications, like several posters noted before...

Well it must not use any real world programs every other benchmark uses. Such as but not limiting to video encoding,editing software -handbrake it beats an I3 but loses pretty badly to a I5. Adobe After Effects 6 its 50% slower than a I5. In 7-Zip 9.2 its slower compared to its older A10 6800K. In MediaShow Espresso Video Transcoding Its 22% slower than a I5 and still slower than a 6300 even. But as long as its on level with a I5 in passmark i guess its ok.

 

[-Fran-]

I'm sorry for abruptly jumping in here, but I've been curious about Mantle. I've watched a couple of videos from real world testers (Linus, ToT, etc.), and I've found that from the videos and articles read, Mantle has been mostly pulling crap performance. Also the crap is better on Nvidia than the crap of AMD. I know it's still kinda new, and there have been improvement, but from the charts showing Mantles amazing performances seem to be from AMD or endorsed by AMD.

I'm not trying to pull sides, I'm just wondering why all I see is that insanely good performance and then the really crappy performances from real world testers.

In a nutshell, you have "lab" conditions and "real world" conditions. Both are different for each PC, and getting a beta product to work fine everywhere is hard to do. That's why they're beta, right? So, for those RL testers (mind giving a link or two?) could be using things not considered for the MANTLE driver first release or having serious bugs on their faces, etc. Anyway, RL != Lab; so differences are to be expected.

Also, I don't think that MANTLE is a marketing tool *only* (I do agree it helps a lot to give a good image for AMD). It's a slap in the face to Khronos and Microsoft for being lazy bastards (more MS than Khronos though). AMD has a problem with their CPUs and no graphics API is lessening the CPU burden making their CPUs look like crap. So the solution is to win the Console designs and then put MANTLE in play (geddit? heh). Makes perfect sense to me from a business perspective. If you have a disadvantage to your product that cannot be undone easily, look for a workaround.

Cheers!

 

[jdwii]

Lower level API was needed, didn't matter who made it, as long as it works on all hardware developers if anything Mantle proved to Microsoft that their API was bottlenecking performance in CPU limiting tasks which is important since die area on a piece of silicon is getting harder to increase without it being economically feasible. We need software to squeeze every bit of performance out of its hardware as efficient as possible.

 

[juanrga]

First hints Microsoft would replicate MANTLE to remain competitive

http://www.extremetech.com/gaming/177407-microsoft-hints-that-directx-12-will-imitate-and-destroy-amds-mantle

Then official announcements that AMD is partnership with Microsoft to develop "the infrastructure to render next-generation graphics workloads at full performance."

http://schedule.gdconf.com/session-id/828412

Finally confirmation Microsoft is adopting MANTLE and renaming it to DX12

http://semiaccurate.com/2014/03/18/microsoft-adopts-mantle-calls-dx12/

For those without S|A account

DirectX12 is nothing more than DirectX11 with Mantle API shamelessly integrated

 

[jdwii]

http://www.brightsideofnews.com/news/2014/3/18/exclusive-amds-carrizo-apu-details-leaked---soc-for-the-mainstream.aspx
Exclusive: AMD's Carrizo APU Details Leaked - SoC for The Mainstream

 

[gamerk316]

Lower level API was needed, didn't matter who made it, as long as it works on all hardware developers if anything Mantle proved to Microsoft that their API was bottlenecking performance in CPU limiting tasks which is important since die area on a piece of silicon is getting harder to increase without it being economically feasible. We need software to squeeze every bit of performance out of its hardware as efficient as possible.

Couldn't you then make the argument we should go back to hardcoded Assembly, rather then accepting the 5-10% performance hit of using a high level programming language?