CES 2014: more details on Mullins/Beema

[Khato]

Turbo with better thermal/power management and perhaps better process (GloFo 28nm) suited for lower power/lower frequency APUs.

My bet's on the implementation of turbo being responsible for the increase in CPU performance metrics for Beema and Mullins. Such was an unfortunate feature for their previous low power entries to be missing.

Though the introduction of turbo also allows for some unfortunate games to be played with respect to both TDP and SDP if AMD so desires. Please note that these numbers are purely fabricated in order to illustrate the point - that 4.5W TDP could easily be achieved with a stock frequency of 800 MHz for the CPU and say 400 MHz for the GPU since when run at those speeds a full load would remain within that 4.5W TDP. But that's not the speed that either the CPU or GPU would be running at during independent benchmarks such as those provided in the PR materials. Oh no, for those the CPU could ramp up to a full turbo 1.5 GHz and use up almost all of the 4.5W TDP while for graphics loads the CPU would likely go to the lowest power state possible in order to give the entirety of that TDP to the GPU. A properly designed turbo is an excellent way to basically double performance/TDP without providing any actual improvements in efficiency.

Note that the same can also apply to their improved SDP numbers - they could quite easily apply to non-turbo frequency operation only.

With all that said, everything points to Beema and Mullins actually being a decent SoC for laptop and tablet use, especially compared to Kabini and Temash. I'm certainly not getting my hopes up as high as AMD's hype would like them to be for the reasons stated above... but they don't need to have some super duper improvement in efficiency to be great options for their intended purpose. Even at the same level of efficiency as Kabini and Temash the inclusion of turbo will increase how responsive the device is (really only applies to the low power Temash models) while connected standby/proper low power states will hopefully at least put its idle power consumption on par with Haswell ULT. Those were really the only major 'problems' that needed to be addressed - they were already doing quite well given their manufacturing process in terms of both CPU and GPU efficiency.

 

[sm625]

Oh no, for those the CPU could ramp up to a full turbo 1.5 GHz and use up almost all of the 4.5W TDP while for graphics loads the CPU would likely go to the lowest power state possible in order to give the entirety of that TDP to the GPU.

I would love it if AMD was that good at playing games with their turbo. But from what I've seen AMD turbo barely works.

 

[inf64]

I would love it if AMD was that good at playing games with their turbo. But from what I've seen AMD turbo barely works.

Piledriver's Turbo implementation (esp. in case of Richland) works really good. This goes for both mobile and desktop parts.

 

[IntelUser2000]

They can easily close the 20% CPU gap just by increasing clockspeeds. Also we don't know if Cherry Trail is just a shrink or if there are architectural improvements. Perhaps its just like BT, perhaps there will be a typical ''up to 5%'' Tick perf/clock increase (perhaps more?).

I don't buy the performance lead at all.

They use the same PCMark 8 benchmark to claim that the A10-6800K is faster than the 4670K, meaning huge emphasis on the graphics. That means its a partial graphics benchmark.

 

[inf64]

We'll have to wait it out I guess.

 

[Sweepr]

I don't buy the performance lead at all.

They use the same PCMark 8 benchmark to claim that the A10-6800K is faster than the 4670K, meaning huge emphasis on the graphics. That means its a partial graphics benchmark.

Thanks for pointing it out, already had my suspicions about this after Kaveri's marketing slides. I guess last year's Bay Trail-T will not be ''solidly behind in CPU performance'', let alone Cherry Trail-T.

 

[AtenRa]

I don't buy the performance lead at all.

They use the same PCMark 8 benchmark to claim that the A10-6800K is faster than the 4670K, meaning huge emphasis on the graphics. That means its a partial graphics benchmark.

http://www.futuremark.com/benchmarks/pcmark8

ΗΟΜΕ

The PCMark 8 Home benchmark includes workloads that reflect common tasks for a typical home user. These workloads have low computational requirements making PCMark 8 Home suitable for testing the performance of low-cost tablets, notebooks and desktops. Home includes workloads for web browsing, writing, gaming, photo editing, and video chat. The results are combined to give a PCMark 8 Home score for your system.

Edit : It can also use OpenCL

 

[inf64]

Also crying about graphics subtest is kinda pointless since its probably the most important part of the chip in this segment . It should be regarded as a big plus for a product to stomp the competition in graphics area. Look at Nvidia, they try so hard to update Tegra in order to get the best graphics performance out of it.

 

[AtenRa]

Oh no, for those the CPU could ramp up to a full turbo 1.5 GHz and use up almost all of the 4.5W TDP while for graphics loads the CPU would likely go to the lowest power state possible in order to give the entirety of that TDP to the GPU. A properly designed turbo is an excellent way to basically double performance/TDP without providing any actual improvements in efficiency.

Thats exactly how BayTrail turbo works

 

[AtenRa]

Also crying about graphics subtest is kinda pointless since its probably the most important part of the chip in this segment . It should be regarded as a big plus for a product to stomp the competition in graphics area. Look at Nvidia, they try so hard to update Tegra in order to get the best graphics performance out of it.

Not only NVIDIA, Intel puts more weight on iGPUs than CPUs for the last two microarcitectures and every x86 and ARM design will from now on.

 

[Abwx]

I don't buy the performance lead at all.

They use the same PCMark 8 benchmark to claim that the A10-6800K is faster than the 4670K, meaning huge emphasis on the graphics. That means its a partial graphics benchmark.

As pointed by Atenra PCM8 Home suite require at least
a DX9 GFX , so much for the graphics ponderation but sure
that some people prefer Sysmark...

 

[Khato]

Thats exactly how BayTrail turbo works

Did I ever say that it wasn't? No, I was just illustrating the fact that with the introduction of proper turbo frequency capability it's quite easy for AMD to cut the TDP of the SoC nearly in half and even increase peak performance capability without changing actual performance per watt at all.

Which, as said, is actually a very good thing. AMD is adding the key capabilities that were missing from the previous generation and hence at least entering the realm of being a viable option for these markets.

 

[AtenRa]

Did I ever say that it wasn't? No, I was just illustrating the fact that with the introduction of proper turbo frequency capability it's quite easy for AMD to cut the TDP of the SoC nearly in half and even increase peak performance capability without changing actual performance per watt at all.

Which, as said, is actually a very good thing. AMD is adding the key capabilities that were missing from the previous generation and hence at least entering the realm of being a viable option for these markets.

I dont disagree, I just pointed out that Intel BayTrail does exactly that

But, I still believe they will use a more Tablet oriented SoC design this time and perhaps a better process.

 

[mikk]

PCMark 8 is partially a GPU benchmark which is nothing new. That's why AMD likes to use it, so they can avoid real CPU benchmarks.

 

[AtenRa]

PCMark 8 is partially a GPU benchmark which is nothing new. That's why AMD likes to use it, so they can avoid real CPU benchmarks.

Well yea, its not like Cinebench that everyone using it every day in his PC/Laptop and now in their Tablets too

 

[Abwx]

PCMark 8 is partially a GPU benchmark which is nothing new. That's why AMD likes to use it, so they can avoid real CPU benchmarks.

CPU benchmarks from Bapco ?..Or the excel test that intel
cooked for MS.?..

Thing is that theses are not simply CPUs but CPUs + GPUs
but since Intel GPU part is lacking then we must discard
this part when doing benchs...

 

[Khato]

hing is that theses are not simply CPUs but CPUs + GPUs
but since Intel GPU part is lacking then we must discard
this part when doing benchs...

Was anyone stating that the PCMark 8 result should be discarded because it includes a graphics workload? A quick glance over the last few posts doesn't show such. No, all I see are comments on the revelation that it's not a pure CPU benchmark as previously believed which means that Silvermont likely still has the lead on that front. Quite rational really.

 

[mikk]

CPU benchmarks from Bapco ?..Or the excel test that intel
cooked for MS.?..

It doesn't change the fact that PCMark 8 is partially a GPU benchmark and AMD uses it frequently for this reason in lots of marketing slides and Roadmaps.

 

[Abwx]

Was anyone stating that the PCMark 8 result should be discarded because it includes a graphics workload? A quick glance over the last few posts doesn't show such. No, all I see are comments on the revelation that it's not a pure CPU benchmark as previously believed which means that Silvermont likely still has the lead on that front. Quite rational really.

That s a system bench with GPU requirement being DX9...
Anyway if the GPU was so important SB wouldnt stand
a chance against a lowish IB...

 

[Enigmoid]

That s a system bench with GPU requirement being DX9...
Anyway if the GPU was so important SB wouldnt stand
a chance against a lowish IB...

The numbers in PCmark are different in their ivybridge review.

Which shows why PCmark is pretty crappy.

On the slide you posted Ivybridge did not make that kind of gain over SB for the 3770k

(Which could have been marketed as "up to 35% faster". See where I'm going here.)

Edit: I believe that's because of the igp increase. dgpu does not help pcmark too much in certain tests while HD 4000 resulted in massive jump

 

[TrulyUncouth]

Shockingly this company chooses benchmarks that show their product in a favorable light. I see that some here are amazed by that. Next thing you know Intel will be showing slides with Quicksync enabled programs and CPU-heavy benchmarks... I can't believe AMD would start such a trend, irresponsible.

But really. In this segment all I want in a tablet is x86 and good enough CPU/GPU combo to play games from 2+ years ago on painfully low settings. Whichever benchmark can show me that best I support.

 

[Abwx]

Next thing you know Intel will be showing slides with Quicksync enabled programs and CPU-heavy benchmarks.

At least AMD took an independant bench , as for Intel ask Bapco and their SYSMARK or EXCEL TEST wich was "optimized" by Intel in a partnership with MS and if not enough there s Cinebench 11.5 wich run AMDs up to SSE2
while Intel is granted up to SSE4.2 thanks to their magic ICC that compiled
this soft , of course a bench that show that Intel CPU+GPU combined
offering is not as good as the competition , well , this bench is forcibly
crappy , never mind that for consumers it will be a mediocre choice
for their usages.

 

[TrulyUncouth]

I am sorry, I really didn't mean to start a debate about compilers and whatnot. I merely find it funny how everyone views their preferred companies as immaculate angels while the other guys are always pulling shady tricks and underhanded tactics.

It reminds of a great quote "We judge others by their actions, and ourselves by our intentions".

I think the ideal thing would be for the OEM's(Maybe ODM's?) to release the benchmarks that I assume they use internally to pick the SOC's for their tablets. Then we could all measure our e-peens objectively!

Edit: Just to add a bit more substance to my post- As much as I intellectually care about intel maybe cheating on benchmarks and would love for them to be punished if they did, I will be buying a high-end intel cpu for my new computer build this summer. Simply because they do perform well in the few real-life benchmarks that matter to me: GAMES

At least AMD took an independant bench , as for Intel ask Bapco and their SYSMARK or EXCEL TEST wich was "optimized" by Intel in a partnership with MS and if not enough there s Cinebench 11.5 wich run AMDs up to SSE2
while Intel is granted up to SSE4.2 thanks to their magic ICC that compiled
this soft , of course a bench that show that Intel CPU+GPU combined
offering is not as good as the competition , well , this bench is forcibly
crappy , never mind that for consumers it will be a mediocre choice
for their usages.

 

[Enigmoid]

At least AMD took an independant bench , as for Intel ask Bapco and their SYSMARK or EXCEL TEST wich was "optimized" by Intel in a partnership with MS and if not enough there s Cinebench 11.5 wich run AMDs up to SSE2
while Intel is granted up to SSE4.2 thanks to their magic ICC that compiled
this soft , of course a bench that show that Intel CPU+GPU combined
offering is not as good as the competition , well , this bench is forcibly
crappy , never mind that for consumers it will be a mediocre choice
for their usages.

Sysmark is trash, no one is disputing it.

But PC mark is just as bad.

AT ivy bridge mobile preview.

http://www.anandtech.com/show/5772/mobile-ivy-bridge-and-asus-n56vm-preview/4

Much higher score due to the igp + quicksync.

First things first, Intel’s IGP and Quick Sync can make a huge difference with PCMark 7—just take one look at the Computation score to see why. Interestingly enough, the HD 4000 actually outperforms the GT 630M in every single element of the PCMark 7 test suite, often by a fairly large margin. It’s not clear why the storage score in particular should differ by 8%, but that’s what we see.

Most notable differences.

Horrible benchmark.

 

[Sleepingforest]

I'm just hoping AMD can produce to match demand if these are as good some benchmarks show... (ahem. R9 280 and R9 280X, I'm looking at you. I wanted an upgrade this year!)