Guess what product AMD is revealing!

[Nanophys]

What would it suggest to you, if the 200%-250% perf/watt lead for Beema/Mullins is indeed accurate, yet design wins are low? Or let's lowball it for argument's sake and suggest a 150%-200% lead. intel's abysmal smartphone efforts have failed because their products there are failures, contrary to what intel was telling it's investors. That's not the case with Cat cores, they are very competitive and leaders in the majority of metrics, so logic says that something else must be influencing design wins.

I dabbled with 100 shares a while back just to feel like I supported fair play, but after witnessing the corruption and greed that is wall street, and the propaganda medium that it has turned tech forums around the internet into, I won't give them a dime to manipulate and then take. Greed is such an unadmirable trait.

You do realize those numbers are meaningless and no conclusions can be drawn from them? You can't draw any reasonable conclusions from them at all (similarly to how Intel's claims about Silvermont's wonders were pretty useless). There is a 0% chance that Beema/Mullins has even a 100% efficiency advantage over competition. Those types of things don't exist in the real world where anything resembling that low of hanging fruit would have easily been picked up with any firm putting out even a half-assed effort. Situations approaching that are a result of 1) fixed function hardware, 2) designed to niches, 3) large fab edge (which AMD certainly doesn't have), or 4) totally different design constraints.

Normally I don't get into these kinds of pissing matches, but saying the phrase "Or let's lowball it for argument's sake and suggest a 150%-200% lead." indicates a severe lack of common sense.

 

[Phynaz]

Funny, op derailed his own thread, lol.

Which can be fixed. Thread crapping on the other hand is a punishment.
-ViRGE

 

[blackened23]

^We'll just ignore that they have an inferior product as well. All of their prior SOCs had either poor performance or poor performance per watt. This wasn't the case with the Athlon / K7 which was an incredible product and better than Intel's offerings at the time. And despite what some would say, it sold incredibly well because it was a great product. I was one of those people buying them for all of my desktops, I certainly wasn't buying Intel at the time.

Unlike the Athlon products which were clearly better than the competition (and sold well as a result), AMD's current SOCs? Are not great products. Not even good products. They pale in every way imaginable to both intel's offerings in terms of performance per watt and overall performance for a given TDP. And i'm sorry, performance per watt is everything for mobile. Bay Trail has great performance per watt and great CPU performance, but it isn't perfect. That's a different situation than the SOCs that AMD has released thus far. So it's really tiring to see this victim card played out - just ask AMD to release better products. That's all you need to do. Not the victim card. Beema and Mullins may do that.

I don't know. Maybe beema or mullins will be good products. I will wait and see how they pan out, but i'm guessing they will be slight and small improvements over kabini or temash. That probably wouldn't be a good thing. But we shall see. Rest assured if they ARE good products, which prior AMD SOCs were not good products, they will get design wins. Period end of story. I don't buy this line of thinking I see here. If AMD releases a great product in terms of what the market wants, OEMs will pay attention to that; certainly they did so with the Athlon K7 line of CPUs.

 

[Abwx]

Everything seems to match up, the only potential differences I see are in this circled region. But it could just be the different method they used to highlight the transistor blocks.

There s a block above the left end of the ellipse that was added, seems that the GPU was improved and there are possibly GCN 2.0 CUs, the CPUs look the sames but a few mods at very low level would be indeed invisible on a die comparison.

 

[blackened23]

AMD had a good product during the Athlon days. AMD does not have an SOC that is remotely competitive by any real metric that OEMs care about - CPU performance, performance per watt, lack of android compatibility.. Again, let's not play the victim card of the issue being entirely paying OEMs. They just had inferior products in the past. This isn't the same situation as Bay Trail which has probably the best CPU performance of any android / x86 SOC on the market, and competitive PPW. It is a good product, but the issues with it were mostly (IMO) related to not having android compatibility which was recently rectified.

Again, let's not skew the main issue with AMD SOCs. Again, perhaps Beema and Mullins will change this. Perhaps they will be good products. And if they are, they will get design wins. Good products sell well, and the athlon sold well; intel was in disarray during those years. Because AMD had the better product. That is a large contrast to the current situation with AMD SOCs, but again, we'll see how Beema and Mullins actually do in the real world. I kinda suspect it will be a slight improvement over last years' products, which probably would not be a good thing.

 

[Maximilian]

Its gonna be some kinda quad core APU that has better graphics than intel chips but not good enough to displace discrete cards.

Same as the last umpteen releases. :\

 

[Enigmoid]

What would it suggest to you, if the 200%-250% perf/watt lead for Beema/Mullins is indeed accurate, yet design wins are low? Or let's lowball it for argument's sake and suggest a 150%-200% lead. intel's abysmal smartphone efforts have failed because their products there are failures, contrary to what intel was telling it's investors. That's not the case with Cat cores, they are very competitive and leaders in the majority of metrics, so logic says that something else must be influencing design wins.

I dabbled with 100 shares a while back just to feel like I supported fair play, but after witnessing the corruption and greed that is wall street, and the propaganda medium that it has turned tech forums around the internet into, I won't give them a dime to manipulate and then take. Greed is such an unadmirable trait.

AMD slides are saying perf/TDP not perf/W. Massive difference.

 

[Abwx]

AMD slides are saying perf/TDP not perf/W. Massive difference.

If the TDP is expressed in volts, ampere or whatever it would make a difference but since it s expressed in Watts perf/TDP = perf/Watt, all other theories are useless polemics.

 

[AtenRa]

Everything seems to match up, the only potential differences I see are in this circled region. But it could just be the different method they used to highlight the transistor blocks.

If those are real differences they are in the GPU portion of the die. Is Kabini GCN 1.0, 1.1 without true audio? If so that could be the difference, Beema/Mullins may be latest GCN versus Kabini not quite latest.

Have a closer look at the circles, there are differences in the design.
The big circle is in the iGPU, the two smaller ones are situated in the PCI-e and VGA area.

 

[ViRGE]

I'm only going to say this once: Intel's business practices are not part of this discussion. Stick to AMD's upcoming Cat products or take it to another thread.
-ViRGE

 

[Vesku]

Assuming the visible differences are real, this soon to be announced Cat family based APU has updated GPU and FCH (external I/O) any other tweaks are not visible, at least to me, as transistor blocks comparing the provided images:

 

[Homeles]

If no HDL, then can we say they finally applied the resonant mesh improvements in the cat line, AFAIK amd only talked about using on the big core line

RCM's fairly easy to spot on die. The clock drivers get significantly larger.

 

[Shivansps]

Have a closer look at the circles, there are differences in the design.
The big circle is in the iGPU, the two smaller ones are situated in the PCI-e and VGA area.

If the thing does not come with dual channel does not matter what they do to the igp.

Probably GCN 2.0, but irrelevant.

About the pci-e, If they upgrade PCI-E to 3.0 will be more than enoght.

 

[rainy]

If the thing does not come with dual channel does not matter what they do to the igp.

Most probably we will see dual channel MC in the next year (on 20nm) together with stronger GPU - 3-4 CU instead of 2.

 

[Vesku]

If the thing does not come with dual channel does not matter what they do to the igp.

Probably GCN 2.0, but irrelevant.

About the pci-e, If they upgrade PCI-E to 3.0 will be more than enoght.

Considering the full desktop APUs are just barely game worthy at low resolutions I don't think the lack of dual channel is that big of a deal in the types of devices these low power products they are aimed at. Does affect desirability for use in servers, but AMD seems to be going ARM for low power server APU.

Will be nice for home server use if and when the cat family gets it, though. The lack of dual channel does hurt them in the few reviews these types of products get.

 

[Shivansps]

Its really important for desktop too, if Kabini had DC it will be outperforming Haswell GT1, and it may be no so useless compared to a G1820.

 

[Vesku]

Its really important for desktop too, if Kabini had DC it will be outperforming Haswell GT1, and it may be no so useless compared to a G1820.

What exactly can Haswell GT1 allow to be done that Kabini can't due to its lack of dual channel? This mystery cat like die doesn't seem to have any more GCN units so no surprise there is no evidence of dual channel yet. Ignoring GT1, what exactly could 128 GCN units do with dual channel it can't with single channel other than get a slightly better unplayable experience in a game?

 

[Abwx]

Kabini already perform very well and dual channel is a tweak they still have up their sleeves to improve valuably its absolute perfs and perf/watt, so far Kabini start to show marginal perfs delta over 2133MHz RAM frequency, if Beema get 1866MHz RAM support it would be enough in the waiting of next gen uarchs that will indeed get a second memory channel wich is useless with only the current 128 SPs/2CUs.

 

[Homeles]

Nevermind.

 

[Shivansps]

Its destroying the lead on IGP that AMD had, its not only on desktop, on mobile the 2955U is a problem for Kabinis too.

DC on Haswell GT1 allows it to cach up with it, even surpass it, that is enoght to be a big problem for AMD.

 

[Vesku]

Its destroying the lead on IGP that AMD had, its not only on desktop, on mobile the 2955U is a problem for Kabinis too.

DC on Haswell GT1 allows it to cach up with it, even surpass it, that is enoght to be a big problem for AMD.

A desktop APU ~30-50W actual power under load CPU marginally, in graphics, surpassing a 15-30W SFF APU (AM1 socket products) is not what I'd call a big problem. Now if the next node version of AM1 APUs doesn't have at least double the GPU horsepower and dual channel, that would be a problem. It is a shame there are no visible CPU tweaks from Kabini on this mystery APU. Cat family compares well in CPU performance to Bay Trail but both are quite a bit behind Haswell Celeron/Pentium.

Considering, Cat family based AM1 looks to be replacing AMD's sub $100 big core APUs. Currently still just dual core Trinity and Richland priced against the Haswell Celerons and Pentiums. No sign of even the entry i3 competitor Kaveri A8-7600.

I think this is getting somewhat off topic as this Cat family based mystery APU is unlikely to be AM1. It's too soon to bump the launch AM1 products. This will most likely be bumping current mobile Kabini.

 

[mrmt]

AMD slides are saying perf/TDP not perf/W. Massive difference.

This. You need to go no further than what's in bold to know that the rest is pretty much useless. We are talking about AMD, which gave us Barcelona, Bulldozer, unsupported openCL and Kaveri, and in some of these cases there were AMD executives behind the flawed statements. They bent the truth some times and, outright lied on others. Any AMD marketing slide should be taken with gargantuan quantities of salt, they have no credibility.

 

[Vesku]

This. You need to go no further than what's in bold to know that the rest is pretty much useless. We are talking about AMD, which gave us Barcelona, Bulldozer, unsupported openCL and Kaveri, and in some of these cases there were AMD executives behind the flawed statements. They bent the truth some times and, outright lied on others. Any AMD marketing slide should be taken with gargantuan quantities of salt, they have no credibility.

Well, AMD's TDP ratings with Kabini seem quite reasonable based on third party reviews. It would be somewhat surprising if they tossed that with Beema/Mullins, didn't they start using Intel's SDP for that kind of "trickery"? IIRC, that's how they were trying to pitch Temash for tablets, an SDP rating.

It's not as if OEM product design teams are fooled easily by such sleight of hand anyway. Cooling and psu/battery will have to work on the actual product not just on paper.

Edit: I did remember correctly. AMD is using SDP for Mullins which they also how they rated Temash but sticking with TDP, how Kabini was rated, for Beema.

"First, we have Beema replacing Kabini, and with the change we get the AMD Security Processor (ARM Cortex-A5) and a reduction in TDP on some parts, with 10W being the minimum. Mullins does the same for Temash, only AMD uses SDP (Scenario Design Power) rather than TDP (Thermal Design Power), and the new APUs are ~2W compared to 3-4W for Temash. Apparently the TDP for Temash is 8W and the TDP for Mullins is 4.5W, and that’s what AMD uses for their performance per watt calculations."

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

AMD was unusually specific with their claims regarding Beema/Mullins:

Granted if they 'fudge' Beema TDP by a watt or two compared to how they rated Kabini that can magic up a chunk of their efficiency claims. Heck in the unlikely event that 15W TDP Beema draws the same actual power as 25W TDP A6-5200 it's still ~20% more efficient in perf/W at least in the two benchmarks AMD was willing to release. Impressive without a node shrink.

 

[piesquared]

http://www.computerbase.de/2014-04/...s1b-test/3/#diagramm-dbpoweramp-16-wav-zu-mp3

Since Kabini already competes with/ beats bay trail in perf/watt, and Beema/Mullins is 2-2.5x better than that, bay trail will not be competitive.

 

[mrmt]

Well, AMD's TDP ratings with Kabini seem quite reasonable based on third party reviews. It would be somewhat surprising if they tossed that with Beema/Mullins, didn't they start using Intel's SDP for that kind of "trickery"? IIRC, that's how they were trying to pitch Temash for tablets, an SDP rating.

If they were being forthcoming they would be talking about performance/watt, which is a metric that the industry and AMD itself recognizes as meaningful and is what they use in the press releases for their other products. Instead they are quoting performance/TDP, which is something rather useless, especially if the two TDP metrics aren't equal for AMD, something they didn't disclose. Plus one chip has turbo, the other does not. Depending on how aggressive turbo is on Beema and the testing conditions (starting temperature, cooling), that will really twist the results.

I'm sure that the even if Beema is more efficient than Kabini, the numbers will really fall short of reaching the numbers AMD is extracting with their performance/TDP mess. We'll know soon enough.