[PCPER] beema and mullins

[blackened23]

I dont want a 3RD AMD thread locked becuase you guys love too much speculation and going off-topic...
this is a beema/mullins thread not a future intel[will own/ is better/ is etc] thread!

I discussed this with the mods. People are playing games to stifle and silence opinions that they don't like. Just because some people aren't "all in" on praising APUs nonstop. They agree that the concept of playing games to silence opinions through claims of "off topic" and "derailing" is nonsense.

The competitor of beema and mullins is relevant to the thread. Period. As much as you don't like it - it is relevant to the discussion. If someone brought up, say, the xbox 360 in this thread THAT would be off topic. The direct competitor to Beema and Mullins has COMPLETE relevance. So stop trying to say it's off topic, because it isn't.

 

[Sweepr]

I discussed this with the mods. People are playing games to stifle and silence opinions that they don't like. Just because some people aren't "all in" on praising your APUs nonstop. They agree that the concept of playing games to silence opinions through claims of "off topic" and "derailing" is nonsense.

The competitor of beema and mullins is relevant to the thread. Period. As much as you don't like it - it is relevant to the discussion.

Exactly. Just take a look at Bay Trail's thread. Most of these people did lots of Bay Trail vs Qualcomm, NVIDIA, Samsung, Apple, AMD (etc.) comparisons.

 

[monstercameron]

Hmmm.. Its looking like this is just the same kabini die.
Anyone find anything suggesting this is more than just a tweaked kabini

what about the arm trustzone ip?

 

[Ajay]

I discussed this with the mods. People are playing games to stifle and silence opinions that they don't like. Just because some people aren't "all in" on praising APUs nonstop. They agree that the concept of playing games to silence opinions through claims of "off topic" and "derailing" is nonsense.

The competitor of beema and mullins is relevant to the thread. Period. As much as you don't like it - it is relevant to the discussion. If someone brought up, say, the xbox 360 in this thread THAT would be off topic. The direct competitor to Beema and Mullins has COMPLETE relevance. So stop trying to say it's off topic, because it isn't.

Maybe we need AMD and Intel subsections in CPUs and Overclocking, while I'm fine with comparative discussions, it may be the only way we will keep threads open in an era where AMD is struggling and the h8ers on each side are becoming more vehement. I can see why AMD fans are feeling bullied.

 

[Ajay]

It's Merrifield (the smartphone SoC) that will be using a PowerVR GPU, supposedly one of the series 6 configurations.

Which is indeed because Intel's GFX still use too much power for that market due to the fact that the 'static' components simply don't scale down low enough. Which is to say that Intel would have to 'rip out' a number of features necessary for the other markets in order to get the feature set and hence power consumption to a level comparable to other smartphone graphics. Once you get to 14nm though the reduction in power thanks to the node shrink is likely adequate to fit in all those features and offer excellent performance.

It's actually rather amusing that it seems that Intel had an easier time scaling an x86 processor down to these power levels than it is with getting its graphics architecture to do the same. It'll be quite interesting to see how the current mobile graphics architectures - Adreno, Mali, and PowerVR - do with scaling up once Intel, NVIDIA, and hopefully even AMD are offering PC-compatible graphics to the market. (Though such only really matters if games take advantage of them.)

Do you have a link on this? I'm getting conflicting info on Merrifield , but I'm not having much success on Google.

 

[blackened23]

Maybe we need AMD and Intel subsections in CPUs and Overclocking, while I'm fine with comparative discussions, it may be the only way we will keep threads open in an era where AMD is struggling and the h8ers on each side are becoming more vehement. I can see why AMD fans are feeling bullied.

It can of course be discussed respectfully. But silencing an opinion just because someone doesn't like it, well, that's stupid. That's what these forums are all about, sharing opinions and ideas. You don't have to agree with 100% of them - debating and discussing the faults and merits of products is what forums like this are designed for. The bread and butter, if you will. If we were to go with the AMD fan idea of "on topic" here in this thread, we'd probably sound like robots talking about marketing slides. That is of course, nonsense - we can speculate and share opinions about beema/mullins AND competing products, there is NOTHING wrong with that. Heck, I hope both of these products are ALL THAT and more, a competitive AMD brings nothing but GOOD things to the tech community. Again, that's what this forum is for. Sharing ideas and opinions that you don't agree with. If someone is clearly out of line in attacking someone with an opinion, then let it be dealt with - make no mistake, the mods don't take lightly on personal attacks and over-the-top behavior and what not here from what i've seen. Yet, claiming something to be "derailing" when it is relevant to the product being discussed, that is of course nonsense. A direct competing product is relevant. We are not robots. WE are here to share ideas, again, whether they're agreed with or not. Airmont, specifically, is relevant because it is a direct competitor - now maybe AMD can pull an Apple. Can they? I don't know. Apple did miracles with their A7 design on 28nm. But stating that airmont discussion isn't allowed, well, to sound like a broken record - that is complete nonsense. Direct competing product.

IMHO

 

[Vesku]

Imo, compare it to Intel if you'd like but you should do it with published information from Intel vs the published information on beema and mullins. To just speculate IS off topic.

Otherwise, I could claim 14nm Atom core will be not only twice as efficient but also twice as fast as Ivy Bridge and Intel integrated Gen 8 graphics will destroy Nvidia and AMD discrete graphics all the way into the mid-range using 1/10th the power.

 

[Arachnotronic]

IF the 2x perf/watt is true, then maybe they can compete - at least for high resolution platforms (since AMD GFX will be better - Airmont is still IMG Tech, AFAIK). AMD desperately needs some design wins in the space.

Are these coming from TSMC or GFL?

Cherry Trail (Intel's tablet 14nm part based on Airmont) will be (from what I'm hearing) a 16 EU Gen 8 configuration.

 

[erunion]

Cherry Trail (Intel's tablet 14nm part based on Airmont) will be (from what I'm hearing) a 16 EU Gen 8 configuration.

16 EU? I find that incredibly hard to believe. Its going to be 5x the performance?

 

[Sweepr]

Cherry Trail (Intel's tablet 14nm part based on Airmont) will be (from what I'm hearing) a 16 EU Gen 8 configuration.

That would mean a huge performance jump compared to Bay Trail's 4 EUs (Gen 7).

14nm will probably allow this without a significant die size increase. I hope its true.

 

[Homeles]

16 EU? I find that incredibly hard to believe. Its going to be 5x the performance?

Well, I think it's not outside of the realm of feasibility. Consider the fact that Silvermont is underpowered with 4. I'd imagine 4 EUs is the "least common multiple," so more hardware comes in lots of 4. If 4 was too few, 8 on Silvermont would have been "just right;" we'd get more units at a lower clock speed, with an overall performance increase.

So if 8 is what we "should" have gotten, 16 isn't too unreasonable with scaling being somewhere in the realm of 2x.

 

[Centauri]

I'd really love to see AMD get some x86 tablet wins in 2014 because I'm chomping at the bit for a Windows 8.1 tablet that doesn't have Intel inside.

 

[Ajay]

It can of course be discussed respectfully. But silencing an opinion just because someone doesn't like it, well, that's stupid. That's what these forums are all about, sharing opinions and ideas. You don't have to agree with 100% of them - debating and discussing the faults and merits of products is what forums like this are designed for. <snip>...

I agree with you in principle. My concern is the posting vacations and thread closures that seem to happen more often when discussing AMD CPU products. I think that AMD's situation in the CPU market has made reasonable debate in CPUs and Overclocking on AMD products a tenuous proposition where a full vetting of the topic at hand to quickly turns into a debate of AMD vs Intel and then devolves into either chaos, or a thread more about Intel than AMD (where AMD proponents are effectively evicted). At times I have been guilty of making a snide comment or two because Intel's generous lead in x86 CPUs makes it easy to take cheap shots at AMD's products.

I've posted my suggestion for a fix here: http://forums.anandtech.com/showthread.php?t=2353394

If we could all remain rational and respectful, no change would be needed.

 

[Khato]

Well, I think it's not outside of the realm of feasibility. Consider the fact that Silvermont is underpowered with 4. I'd imagine 4 EUs is the "least common multiple," so more hardware comes in lots of 4. If 4 was too few, 8 on Silvermont would have been "just right;" we'd get more units at a lower clock speed, with an overall performance increase.

So if 8 is what we "should" have gotten, 16 isn't too unreasonable with scaling being somewhere in the realm of 2x.

There isn't really a least common multiple. If there was, then it would have been the 6 EUs that is Gen7 GT1... which actually would have been a very reasonable performance level for Baytrail. The real 'problem' with Baytrail graphics was simply that the performance target wasn't correct. Intel had to actually do a non-trivial amount of work to make a smaller GPU when they could have just used the 6 EU Gen7 GT1.

The same can somewhat be said of what's expected of Cherrytrail with 16 EUs - why exactly couldn't they have just kept as much in common with Gen8 GT2 and gone with 20 EUs? Is it really necessary to spare whatever fraction of a square millimeter that cutting the number of EUs gains? And likely come in below the competition?

That's one thing that definitely can be said for AMD - they actually devote a far more appropriate amount of die space to graphics. Though it is somewhat questionable if that's changing with Beema/Mullins given the speculation that it still might 'only' be 2 CUs. Which sounds quite likely since they don't have a process shrink to give them a larger budget.

 

[Ajay]

Well, I think it's not outside of the realm of feasibility. Consider the fact that Silvermont is underpowered with 4. I'd imagine 4 EUs is the "least common multiple," so more hardware comes in lots of 4. If 4 was too few, 8 on Silvermont would have been "just right;" we'd get more units at a lower clock speed, with an overall performance increase.

So if 8 is what we "should" have gotten, 16 isn't too unreasonable with scaling being somewhere in the realm of 2x.

Well Gen8 graphics must scale (power wise) much better than Gen7 since we are talking 4 times the EUs and roughly twice the die area. Intel's GPU team must really be cooking with gas to pull this off. On the plus side, I think it's Torvalds who has pointed out that Intel's iGPUs have much better IQ and support than all the top ARM SoCs. So this could be quite interesting if Intel really delivers 16 Gen8 EUs.

 

[Arachnotronic]

I'd really love to see AMD get some x86 tablet wins in 2014 because I'm chomping at the bit for a Windows 8.1 tablet that doesn't have Intel inside.

Bay Trail tablets are great and they're reasonably priced. I don't know why you would want to subject yourself to undue waiting/inferior products when what you want is available today.

Seriously, Windows 8.1 on a tablet is awesome and you are doing yourself a big disservice by not picking one up 🙂

 

[Arachnotronic]

16 EU? I find that incredibly hard to believe. Its going to be 5x the performance?

I'm just relaying what I've heard/seen...as unofficial as it is. Take it for what you will...just a post from some random guy on the internet 🙂

Also note that performance is dependent on clocks. But remember that Intel managed to significantly improve performance in the same power envelope on the order of 5x+ from the original, crappy Clover Trail in moving from 32nm -> 22n. Given that the Airmont CPU won't be a dramatic redesign (although I don't have any info on the core at this time), I would say the die budget is going mostly to the GPU.

 

[24601]

Bay Trail tablets are great and they're reasonably priced. I don't know why you would want to subject yourself to undue waiting/inferior products when what you want is available today.

Seriously, Windows 8.1 on a tablet is awesome and you are doing yourself a big disservice by not picking one up 🙂

Maybe AMD's solution will have a faster GPU in games?

Idk though, the x86 tablet players have been burned too many times before by TDPs that are too high (x86 tablet idea (and implementation) is older than a tablet from any other platform).

AMD wasn't able to get Temash into Tablets for this very reason.

We'll have to see how it plays out this time.

 

[Arachnotronic]

Maybe AMD's solution will have a faster GPU in games?

Idk though, the x86 tablet players have been burned too many times before by TDPs that are too high (x86 tablet idea (and implementation) is older than a tablet from any other platform).

AMD wasn't able to get Temash into Tablets for this very reason.

We'll have to see how it plays out this time.

My Dell Venue 8 Pro works in a tablet and it runs cool as a cucumber and performs extremely well. X86 has no problems getting into a smartphone/tablet - it's all about the microarchitecture/design point.

Silvermont is a <1W/core design while Temash is a ~2W core @ 1.5GHz and doesn't have any deep low power states. There's a reason AMD didn't win any designs this round, but let's hope Puma is a better core. Still can't overcome the fact that it's built on 28nm (laws of physics), but with some better design work on the total SoC, AMD should have something more competitive...

 

[relztes]

I think what we're going to see is aggressive turbo combined with lower base clocks. I'll bet you just about anything that PCMark 8 Home doesn't get the A6-5200 to anywhere near 25 W. Likewise, I suspect that 3DMark 11 is pretty easy on the CPU and AMD is not fully shifting the 25 W TDP to the GPU. So in reality, I'd expect some minor efficiency gains coupled with better turbo to fully utilize the available TDP.

Just to illustrate, take the A4-5000 and add a max single core turbo of 2.4 GHz and a max graphics turbo of 720 MHz. If AMD can do 4x 1.5 GHz in 15 W, they could probably do 1x 2.4 GHz as well. If AMD can do 6 CU at 775 MHz (8770m) at ~25 W TDP, they can probably do 2 CU at 720 MHz at 15 W. Now just find a single threaded CPU benchmark and a CPU-lite GPU benchmark, and you have 20% gains over the A6-5200 (assuming perfect scaling).

Of course, this isn't all bad. Kabini already has nice multithreaded performance. Turbo is just what is needed to improve the single threaded performance. But don't expect anywhere near double the performance per measured watt, just double the performance per TDP in cases where Kabini and Temash weren't actually using that TDP.

 

[monstercameron]

I think what we're going to see is aggressive turbo combined with lower base clocks. I'll bet you just about anything that PCMark 8 Home doesn't get the A6-5200 to anywhere near 25 W. Likewise, I suspect that 3DMark 11 is pretty easy on the CPU and AMD is not fully shifting the 25 W TDP to the GPU. So in reality, I'd expect some minor efficiency gains coupled with better turbo to fully utilize the available TDP.

Just to illustrate, take the A4-5000 and add a max single core turbo of 2.4 GHz and a max graphics turbo of 720 MHz. If AMD can do 4x 1.5 GHz in 15 W, they could probably do 1x 2.4 GHz as well. If AMD can do 6 CU at 775 MHz (8770m) at ~25 W TDP, they can probably do 2 CU at 720 MHz at 15 W. Now just find a single threaded CPU benchmark and a CPU-lite GPU benchmark, and you have 20% gains over the A6-5200 (assuming perfect scaling).

Of course, this isn't all bad. Kabini already has nice multithreaded performance. Turbo is just what is needed to improve the single threaded performance. But don't expect anywhere near double the performance per measured watt, just double the performance per TDP in cases where Kabini and Temash weren't actually using that TDP.

this makes alot of sense, just adding turbo can drastically alter the perf characteristics and provide a dynamic range of operatio much like intels methodology, I just wonder why they decided to launch temash/kabini without it?

 

[Arachnotronic]

this makes alot of sense, just adding turbo can drastically alter the perf characteristics and provide a dynamic range of operatio much like intels methodology, I just wonder why they decided to launch temash/kabini without it?

Turbo is difficult to implement/validate. It takes a lot of sophisticated logic to actually pull it off properly and to distribute/allocate power to each of the blocks correctly. It's a lot of work and a non-trivial achievement on Intel's and the ARM vendors' parts.

 

[24601]

Turbo is difficult to implement/validate. It takes a lot of sophisticated logic to actually pull it off properly and to distribute/allocate power to each of the blocks correctly. It's a lot of work and a non-trivial achievement on Intel's and the ARM vendors' parts.

Lets not give ARM or AMD too much credit in their past implementations.

Arm just jacks the voltages through the roof and that's what AMD did for Llano and ARM players do it to this day.

It'll be good if we see AMD implement a sandybridge like turbo (at least) in kaveri and its derivatives.

That's AMD's only hope in getting x86 into ARM tdp territory before 20 nm (and 20nm with finfet[16nm TSMC]).

 

[Abwx]

AMD stuck on 28nm for 2014 😱

Not a drama if they can release these early on 2014.

http://www.anandtech.com/show/7514/amd-2014-mobile-apu-update-beema-and-mullins

Update: It sounds like most of the performance gains come from frequency increases, while power improvements happen at the SoC level.

Not sure such frequency related high gains could
come from TSMC process , sound more like a GF
iteration.

 

[mrmt]

I think what we're going to see is aggressive turbo combined with lower base clocks. I'll bet you just about anything that PCMark 8 Home doesn't get the A6-5200 to anywhere near 25 W. Likewise, I suspect that 3DMark 11 is pretty easy on the CPU and AMD is not fully shifting the 25 W TDP to the GPU. So in reality, I'd expect some minor efficiency gains coupled with better turbo to fully utilize the available TDP.

This smells Richland all over again, when we got promised everything but in reality got only small improvements.