AMD Beema/Mullins Launch Thread

[NTMBK]

I am assuming Puma+ SOC has two SATA ports just like Kabini (BGA and AM1) and just like Kyoto.

Two sata ports?

Two.

Why only two?

(Yes, I understand Laptops at most only require two SATA. One for 2.5" SSD or 2.5" HDD and one for optical drive.......but isn't there so much more potential beyond this considering the Kabini AM1 cat cores have AES-NI and are ECC capable)

Is adding two more SATA (to bring to total to four) something that drives up silicon die size too much?

How many mm2 for each SATA on 28nm?

I would just love to have a cat core MINI-ITX with ECC and four native SATA ports. Compared to what something from Intel costs with those features it would probably be a real bargain.

It's made for laptops and tablets, not microservers. If you really want more SATA ports then attach a SATA controller over PCIe, though then you still have a wasted GPU.

 

[ams23]

We really need to see real world app-specific power consumption measurements (in a real world thin chassis) to accurately compare and contrast perf. per watt. Lack of real world power consumption and battery life measurements is a huge letdown for these previews.

Most likely the peak (and idle) power consumption is way higher on the higher end Mullins SoC variants compared to Bay Trail SoC variants that you typically see in thin fanless 7-10" tablets.

Basically what AMD is doing is using more power hungry boost frequencies to increase benchmark performance while setting manufacturer-rated TDP no higher than it would be without boost frequencies. Other manufacturers do this too unfortunately, and it is a misleading practice. What reviewers need to do (which is easier said than done) is to move away from using manufacturer-rated TDP and instead use app specific power consumption data.

Keep in mind that if a tablet is big and thick enough, even some Haswell variants can easily fit inside a tablet (see Surface Pro 2 10.6" tablet which is actually even smaller than AMD's reference Mullins tablet). The Haswell SoC variant in the Surface Pro 2 tablet has MUCH higher CPU and GPU performance compared to the high end Mullins variant.

 

[Vesku]

No, i will not let got about the name, i was already piss off about calling R7 to the 7850K, they never did that until they adopted the Rx naming. Calling R6 to a 500mhz 128SP GCN 1.0... they think im a idiot? Tell me whats the difference to the R2 500mhz 128SP GCN 1.0 on Beema E2-6110? Why im the only one that see something wrong in here? and why i need to shut up about it?



Its a little early to say that, what about power and temps? When the A4-1200 was launched it also crushed Cedar Trail at tablets, but no one used it because of power and temps, OEM had to place them on cheap 14,15" notebook to get rid of them.

I don't think it's that outrageous, it's why they inserted the "Micro". The A10 and R6 is in relation to their ultra low power SoCs. A10 tablet is not going to be as beefy as A10 notebook which will not be as beefy as A10 desktop. Actually, I think going to that method of naming is pretty good it's just awkward having the "micro" in the middle of the name instead of the end. Credit to them for putting it where it can't be ignored.

Addressing the improvements from Temash, that skin temperature technology is interesting but I doubt they'll be in a rush to provide Linux drivers unless they do a u-turn on not embracing Android.

 

[Haserath]

I think AMD's idle power is too high to compete well with Arm designs. Anand showed the idle power being double a Qualcomm chip.

Without power numbers, it's hard to say where it stands... It looks really good under Anand's browser benchmarks though.

Unfortunately there's no chance in heck that CPU-Z is reading that right. You need a minimum voltage to keep a transistor on, which in turn is a function of the manufacturing node (it has a technical name at the moment that escapes me). While TSMC doesn't publish that data, no logic node I've ever seen can operate at that low of a voltage. Based on other TSMC products that minimum has to be at least 0.6v, if not higher.

Threshold voltage. Intel's trigate is supposed to be much lower than planar and theirs is somewhere in the 400mv range.

 

[Abwx]

Look like they re targeting more than BT :

In AMD’s own slides, they see the A10 Micro-6700T competing more with Intel’s Haswell Y chips, while the Micro-6400T and Micro-6200T compete with Bay Trail-T.

http://www.techspot.com/article/810-amd-beema-and-mullins-preview/page5.html


Edit : From Hardware.fr redactor :

AMD nous a confirmé deux points qui restaient en suspend, le fondeur (qui est GloFo et non TSMC)

"AMD cofirmed to us two point that were still not clear, the foundry (wich is GloFo and not TSMC)"

http://www.hardware.fr/news/13677/amd-annonce-beema-mullins-maj.html

Seems that i was right, perfs are not due only to the design but likely also to a better process in respect of frequencies.

 

[PPB]

I dont know if it is because the platform they are targetting isnt really into heterogeneous computing, yet (but ironically it loves fixed hardware functions), but am I the only one that thinks that Beema should have HSA enabled functions already? I find this segment a whole lot more interesting that even what mobile Kaveri may bring, and if AMD wants HSA to succeed, enabling it on what is probably their most interesting product in a long while would have been a good decision IMO.

Yes, I know that AMD already had to make concessions in CPU performance to make Steamroller in Kaveri HSA compilant, but I dont think the penalty hit would have been so bad, and I also think that the low power segment really could use HSA enabled hardware, as graphics performance is a key element and lately all chipmakers and putting more weight into their iGPs than on raw CPU perf (as visual media consuming is probably the most relevant use of tablets/phablets/ultraportables these days).

The rest of the chip looks great, AMD has been known of pulling these kind of tricks in the past (when they pulled a Thuban on the same process node, or a Richland out of Trinity in a rather short development cycle and in the same node, too).

 

[Insert_Nickname]

It's made for laptops and tablets, not microservers. If you really want more SATA ports then attach a SATA controller over PCIe, though then you still have a wasted GPU.

That's something the PCIe x4 slot could be used for. Though it'll add to cost.

I do think AMD made the right call here, these chips are targeted at mobile. The desktop versions are just a happy side-effect. Most laptops, especially in this price range, won't need more then two. Added SATA ports would just waste power.

The rest of the chip looks great, AMD has been known of pulling these kind of tricks in the past (when they pulled a Thuban on the same process node, or a Richland out of Trinity in a rather short development cycle and in the same node, too).

Trinity and Richland use the exact same chip. Richland just has a few more features enabled, and the benefit of a mature process.

 

[frozentundra123456]

I think AMD's idle power is too high to compete well with Arm designs. Anand showed the idle power being double a Qualcomm chip.

Without power numbers, it's hard to say where it stands... It looks really good under Anand's browser benchmarks though.


Threshold voltage. Intel's trigate is supposed to be much lower than planar and theirs is somewhere in the 400mv range.

I agree. The performance looks good, and turbo was badly needed. However, TDP /= power consumption, so I am reserving judgement until we see performance/temperatures/battery life in a more realistically sized form factor.

 

[Ancalagon44]

I agree. The performance looks good, and turbo was badly needed. However, TDP /= power consumption, so I am reserving judgement until we see performance/temperatures/battery life in a more realistically sized form factor.

Agreed. I think they will be slightly worse than the TDP numbers, but not much worse. I mean, Kabini was pretty accurate in its TDP numbers. If I remember correctly, the 15w A4-5000 usually used slightly less than 15w. So I think TDP will be fairly accurate here.

 

[NTMBK]

Edit : From Hardware.fr redactor :

"AMD cofirmed to us two point that were still not clear, the foundry (wich is GloFo and not TSMC)"

That's odd, because Anandtech clearly says that they were produced on TSMC: http://www.anandtech.com/show/7974/...hitecture-a10-micro-6700t-performance-preview Not really sure which is correct.

 

[NostaSeronx]

http://images.anandtech.com/doci/7974/Screen-Shot-2014-04-29-at-1.08.08-AM.jpg

Lots of stuff from Nolan.

 

[Shivansps]

I don't think it's that outrageous, it's why they inserted the "Micro". The A10 and R6 is in relation to their ultra low power SoCs. A10 tablet is not going to be as beefy as A10 notebook which will not be as beefy as A10 desktop. Actually, I think going to that method of naming is pretty good it's just awkward having the "micro" in the middle of the name instead of the end. Credit to them for putting it where it can't be ignored.

Addressing the improvements from Temash, that skin temperature technology is interesting but I doubt they'll be in a rush to provide Linux drivers unless they do a u-turn on not embracing Android.

So, lets tell Intel to call the Z3770 an Micro I7-4770T with iris Pro graphics under that logic.

Look at the E1-6010 and E2-6110... a whole 150mhz diff in freq, they are still called R2, even with the top A6-6310 the igp is called just an R4, thats OK!
They are pulling that kind of naming only with the tablet chips.

 

[SlowSpyder]

So, lets tell Intel to call the Z3770 an Micro I7-4770T with iris Pro graphics under that logic.

I don't see a problem with the naming, personally. The A4-A10 naming scheme doesn't necessarily mean performance on absolute terms, but per range of products. A10 of course being their high end part for this range.

Also, I don't see this confusing anyone. The only people who care what SoC is in their tablet know what they're looking for and will know the product's names. My mom isn't likely to buy one of these and call me up complaining that AMD duped her and she thought she was getting a full fledged A10 in her tablet. :whiste:


Assuming the power consumption is decent, AMD has a really nice part here.

 

[Enigmoid]

This is an SoC APU with R6 graphics, i dont see why you have such a problem with its naming.

Since it can finish Cinebench MT faster than Baytrail D it means that it can operate at adequate time at high Turbo frequency for most applications.

Thats not faster that's the same. You need to be above margin of error to be faster.


This is an awesome chip. This is how kabini should have been released.

Good, proper turbo implementation.

Though note that there is throttling to stay within tdp.

Its idle power is also looks noncompetitive.

Of course we do not know the actual power numbers. But if they turn out to be good then I'll probably buy one rather than Bay Trail.

 

[NostaSeronx]

That's odd, because Anandtech clearly says that they were produced on TSMC: http://www.anandtech.com/show/7974/...hitecture-a10-micro-6700t-performance-preview Not really sure which is correct.

If Beema and Mullins are on bleeding edge nodes, the only two are;
GlobalFoundries 28nm-SHP
GlobalFoundries 28nm FDSOI

28nm SHP has 22nm density and self aligned contacts and self aligned double patterning.
28nm FDSOI has true 28nm density and self aligned contacts.

TSMC and CPA nodes are; a mixed 32nm to 28nm density without self aligned contacts.

If TSMC 28nm to GlobalFoundries 28nm FDSOI then the die has shrunk to ~86 mm²
If TSMC 28nm to GlobalFoundries 28nm SHP then the die has shrunk to ~66 mm²
If TSMC 28nm to GlobalFoundries 28CPA then the die has shrunk to ~95 mm²
This is from ~105 mm².

 

[strata8]

I just noticed that Anand might've incorrectly read the idle power graph.



It clearly says Beema and Beema only in the graph, but Anand then says:

Beema and Mullins are definitely in a good place, however they still do consume more power at idle than the smartphone SoCs we typically find in iOS and Android tablets. AMD isolated APU power for the graph above and is using an “eReader” workload

And posts this...

...assuming that Beema and Mullins consume the same amount of power under idle. I find that unlikely, but maybe something was said in the presentation.

 

[Ancalagon44]

Good point strata8.

That makes AMD's graph look odd, as why would they compare power usage of Beema to Kabini? Mullins is the counterpart to Kabini, not Beema.

 

[AtenRa]

Good point strata8.

That makes AMD's graph look odd, as why would they compare power usage of Beema to Kabini? Mullins is the counterpart to Kabini, not Beema.

Beema replaces Kabini

Mullins replaces Temash

 

[coercitiv]

So, lets tell Intel to call the Z3770 an Micro I7-4770T with iris Pro graphics under that logic.

The A4-A10 naming scheme doesn't necessarily mean performance on absolute terms, but per range of products. A10 of course being their high end part for this range.

AMD's A-Series branding is similar Intel's Core branding. Core i7 brand is used for either hexacore, quadcore or dual core CPUs. Moreover, both AMD and Intel already used their second tier brands to designate chips with vastly different performance characteristics. (eg: Haswell Pentium and Silvermont Pentium)

Shivansps, if your point of view is right and this is indeed more than a storm in a tea cup, why don't you give it the attention it deserves and create a separate thread on the matter? You have the arguments, make them stand out from all the chatter in this thread.

 

[Shivansps]

I don't see a problem with the naming, personally. The A4-A10 naming scheme doesn't necessarily mean performance on absolute terms, but per range of products. A10 of course being their high end part for this range.

Also, I don't see this confusing anyone. The only people who care what SoC is in their tablet know what they're looking for and will know the product's names. My mom isn't likely to buy one of these and call me up complaining that AMD duped her and she thought she was getting a full fledged A10 in her tablet. :whiste:


Assuming the power consumption is decent, AMD has a really nice part here.

Ill agree its a nice chip if power is right, but you only need to take a look at the Beemas to figure out something is wrong with Mullins naming, its gets even worse when you check out Temash and Ontario APU/IGP names.

AMD its intencionally making a new scale with known model brands to create confusion. There is even a reason to skip A6 and R4 on Mullins? and R4 whould be already wrong Beema and Mullins should share the same IGP name scale, as Temash did.

But hey, its OK, nothing wrong here.

 

[Ancalagon44]

Beema replaces Kabini

Mullins replaces Temash

Oh.

In that case, Anandtech labelled their graph wrong, and this also affects their conclusion, since a Mullins SOC will thus draw less power at idle than a Beema SOC, and thus be closer to current leading mobile SOC designs.

 

[AtenRa]

Ill agree its a nice chip if power is right, but you only need to take a look at the Beemas to figure out something is wrong with Mullins naming, its gets even worse when you check out Temash and Ontario APU/IGP names.

AMD its intencionally making a new scale with known model brands to create confusion. There is even a reason to skip A6 and R4 on Mullins? and R4 whould be already wrong Beema and Mullins should share the same IGP name scale, as Temash did.

But hey, its OK, nothing wrong here.

Mullins has its own naming, they differentiate it from the rest of the AMD SKUs with "micro". That way it is clear that A10-Micro is the highest Mullins SKU and R6 is the fastest iGPU of the "Micro" family.

I would agree that it is way more difficult to understand if A6-6310 is Beema SoC of Richland APU. With Mullins(Micro) things are crystal clear.

 

[AtenRa]

Oh.

In that case, Anandtech labelled their graph wrong, and this also affects their conclusion, since a Mullins SOC will thus draw less power at idle than a Beema SOC, and thus be closer to current leading mobile SOC designs.

We have no idea if Mullins has lower Idle power than Beema.
But then again, comparing a x86 SoC Idle Power made for Tablets and Laptops to ARM SoCs made for SmartPhones is flawed.

 

[sm625]

All these reviews tell us absolutely nothing. Not one single useful piece of information about battery life. Absolutely nothing. For all we know these are going into 2 cell tablets that run for 2 hours before they die. (Or < one hour of gaming?) It may be able to play a game of DOTA, but we dont know if it will even last long enough to finish one.

 

[Shivansps]

Mullins has its own naming, they differentiate it from the rest of the AMD SKUs with "micro". That way it is clear that A10-Micro is the highest Mullins SKU and R6 is the fastest iGPU of the "Micro" family.

I would agree that it is way more difficult to understand if A6-6310 is Beema SoC of Richland APU. With Mullins(Micro) things are crystal clear.

I do not like the A10 on the name, they are skipping A6 for what reason? not even Beemas have an A10. But the Micros ones goes from an A4 to an A10? yeah, that seems ok.

But im mostly pissed about the IGP name scale, Kabini and Temash had the correct performance scale on them, on Mullins that just crap right there compared to Beemas.

And if add the A10 thing and 6700T on top of it, it gets way too much out of line. Even the numbers does not make sence compared to Beema, 6200T, 6400T, 6700T, Temash had the correct numbers too.

Temash and Kabini where OK, and this is all wrong, i wonder why.