Read more.Quote:
It will also look into a possible breakup and at shifting operations to low-cost countries.
Printable View
Read more.Quote:
It will also look into a possible breakup and at shifting operations to low-cost countries.
Investors...
A company still turns a massive profit but because it's less than perfect and they've had a hiccup, investors want blood and stuff the long-term potential as long as they get their quick fix.
It just seems like, if a company doesn't have a complete monopoly on a market then they're worthless and need to start taking peoples' jobs and cutting pay/investment in order to boost profits of people who do no actual work for said company.
I can't see how splitting the licensing and chip-making companies would benefit anyone except for short-term investor greed at the expense of long-term competitiveness of the company as a whole. Having separate segments complementing each other helps companies get through periods like this and allows more investment in the weaker segment to improve it. If the chip business was separate, would they be able to fund the chip-making part as well in order to compete better in the future? Plus you probably end up with higher overall spending because of duplicated costs with the separate companies, less co-operation, etc.
Snapdragon 810 was always going to be a relatively short-notice stopgap because of the MOAR BITS craze, and if the rumours are true it's a bit sad we might have missed out on their 32-bit Krait successor.
Look at Intel - until they folded the mobile business into their other segments, it was a consistent loss-maker (giving chips away for free tends to do that I guess). I wonder if investors will be wanting that split off to a separate company too as the margins aren't as rosy as their other markets in the short-term? :rolleyes:
Don't get me wrong, like any consumer I like to see good competition in markets and Qualcomm was getting worryingly close to a monopoly in Western markets but I don't think tearing up a company because of increased competition is good for, again, anyone but 'investors'.
/rant
it does seem that the current trend with mobile cpu's is 64bit versions of existing 32bit chips rather than improving the existing ones. I've got a first gen moto g at the moment and the 3rd gen is looking like the cpu will literally be the same cpu but with 64bit added... not exactly needed when the phone only has 2GB of ram max (according to rumours)
I mean the 64bit ARM cores A53 and A57 do improve on their 32bit predecessors the A7 and A15. However rumour has it Qualcomm was working on a 32bit successor to their Krait architecture but because of the market perception of higher number=better, they canned it in favour of off-the-shelf cores - the first time they've done it for a high end part. It's entirely possible this scrapped core could have been a better fit for the Snapdragon 810 than the A57 whose primary focus wasn't really mobile phone use anyway.
32bit processors can use >4GB RAM anyway, so that's not a big reason in and of itself to switch to 64. However the ARMv8 ISA does offer advantages over ARMv7 so there are good reasons to switch eventually regardless of RAM capacity, but there was no need to rush it out.
It's like the whole octa-core thing. It's different for big.LITTLE octa-core for obvious reasons, but I'm talking about e.g. the 8 identical-core A7 CPUs we keep seeing. It's just number-chasing.
@watercooled - agreed. I wish I could switch my pension investments to a plan that wasn't short sighted.
The mobile market is the most fickle market in the world and always has been
Krait is supposed to be high end, and ARM V8 gains you some performance which is what you want at the high end. Really, I think it was possible to see a move to 64 bit coming there quite easily, even if it isn't needed in the likes of the Moto E for a while.Quote:
Originally Posted by watercooled
Yeah that's what I mean - it was coming sooner or later, but aside from marketing it really wasn't a critical feature for the 810 time-frame. Like I say I'm going off rumours rather than hard facts but Qualcomm supposedly had another 32 bit custom core planned for the 810, and it does sound sensible since Qualcomm have typically used custom cores for their high end ever since Scorpion.
Aside from ARMv8 support, the 810 doesn't really offer much over the 805 on the CPU side - performance is about the same or even lower depending on thermals, despite being on a newer node.
I'm not saying 64 bit isn't useful, rather it's not really not a crucial feature for a 810-generation processor, especially not if it's at the cost of real-world performance just for the sake of a check-box in the marketing guff. http://arstechnica.co.uk/gadgets/201...heat-problems/
Having said that, Samsung have managed to do a good job of implementing the A57 in their Exynos line, especially the version in the S6 so it's not like the A57 is unsuitable for the role. Qualcomm's upcoming mid-range A72 part could be interesting.
The smaller cores have their place and are more than enough for a lot of uses, the issue is that real-world performance doesn't scale perfectly by adding more cores. For instance smaller cores can be significantly more efficient than their bigger counterparts which is part of the reasoning behind big.LITTLE; use the smaller cores when they're enough in order in increase efficiency, and have the scheduler switch to using the bigger cores when more performance is necessary.
Apple, who don't have to market their SoCs to people, use just two relatively large cores: http://browser.primatelabs.com/geekbench3/3041978
It's quite sad that a large portion of the SoC market is driven so strongly by marketing nonsense.
It is more complicated than that. The cores are not usually identical, so you get a little.notQuiteSoLittle effect that helps battery life, but on top of that when running benchmarks all 8 cores chime in and some people buy on benchmark results.Quote:
Originally Posted by Bambooz
The other thing is the cost of ARM licenses. The small cores cost less than big cores, and use *much* less silicon. It ends up cheaper to have those 8 cores, and cheap things are easier to sell.
So while not that technically clever, 8 cores does make a lot of business sense.
Edit to add: If you hate 8 core, have you seen the 10 core chip? http://www.fudzilla.com/news/mobile/...core-processor
The little.lesslittle concept makes sense if done right. In particular Huawei's Kirin 930 using what the describes as A53e cores in place of A57 for the 'big' cores. However I'd still like to see a proper analysis like the ones Anandtech have done on the new Exynos to see how beneficial it really is over having say just 4 'A53e' cores.
The big.LITTLE transition isn't perfect in energy terms and takes some effort to get it to work well, so as Anandtech found with some older SoCs, efficiency may not be all that much better than the big cores alone if it's not implemented properly. So with an even smaller performance delta between the cores, I wonder if there's really much of a benefit?
It's more the symmetrical 8-core SoCs I'm on about though; they're marketed on the premise of having a large number of cores and it's potentially harming the chances of realistically better-performing SoCs with lower core-counts. You then have to explain to people who are generally uninterested why fewer large cores are actually better.
Small delta? Using the upcoming X20 as an example, it might use A53 cores for the lower clusters, but they will be optimised differently. It isn't that one cluster is 1.4GHz and the other cluster is 2GHz, it is more that one cluster is optimised for performance and so is capable of 2GHz, whereas the other cluster is optimised for low power and hence can only hit 1.4GHz.Quote:
Originally Posted by watercooled
It sounds like the scheduler biases heavy workloads to faster cores and then powers off unused cores. That is a simpler system than big.LITTLE where work is migrated between clusters.
I wouldn't worry about the future of big cores though. The A72 looks on paper to be pretty epic, and that is a much bigger number than A53 ;)
I really can't find out much about 'A53e' though so I'm just assuming IPC is about the same as A53 with some implementation differences i.e. die size/clock speed trade-off?
I could be wrong, but like I say I've not been able to find much to substantiate that.
I think A53e is made up by marketing types :D
AIUI, ARM cores have options, that's probably just a performance variant.
AFAIK the options would be optional features like the cryptography extensions and uncore features like cache and interconnects; the uArch itself shouldn't change so performance differences should be attributable to clock speed and uncore.
The A53 is a mighty impressive core though - in many cases it's trading blows with Krait 200 despite still being a tiny in-order core! It's not really surprising they didn't increment it alongside the A72 as ARM have explained they're already pushing what is doable without going OoO, but I wonder they'll do going forward?
Qualcomm needs to partner with another to share the cost of design and lock-in a customer (guarantee sales). They have to make a choice now to either:
1) keep aiming for the top end to stay relevant and desirable with the risk of spending too much and not making profits; or
2) start cutting R&D so that they fall behind and start the sales+R&D+profit downward spiral into oblivion.
If you were competitive for only 6 months of the year you may survive. If you were constantly >6months behind then you start reducing prices to be picked for some cheaper/trailing devices but basically you are going to fail. Take AMD's situation as an example of the cost of cutting R&D (separating FAB) and not moving to process shrinks quick enough (28nm vs 14nm = 28nm losses).