The TSMC 20nm planar process seems quite bad, seeing how underperforming the Snapdragon 810 turned out.
So while this is great news because it will bring a powerhouse chip for the low price, I don't see Mediatek competing in high end with Samsung and Qualcomm until they sign some advanced fab deal.
Problem may not be the process but the design. 810 was a rushed SoC as Qualcomm found itself quite behind the competitors with no 64 bit architecture while professing that 64bit was useless at this point of mobile market evolution. We can see Qualcomm wrong strategy in the fact that they will have their own 64bit custom architecture quite late this year (or early next one) and had to use ARM standard IP for their top tier SoC. I would like to know how Qualcomm custom architecture is going to compare with ARM A72 under the point of performances, die size and power. For sure Qualcomm has to design a very good architecture or they risk to go out with an already obsolete design . Though good efficient architecture like the one created by Apple or nvidia require more die space = more expensive. And less cores. Will the cost justify the performance and most of all, how to fight against Cinese/Taiwanese producers that can just put on paper more and more cores with more and more GHz as marketing fud?
I think that the TSMC process is obviously the main issue here. Compare the performance of Snapdragon 810 with 14nm finFET Exynos that uses the same CPU setup, yet arguably less advanced interconnect and power management - Exynos pulls ahead by a mile, seemingly throttling much less.
Snapdragon designed by Qualcomm, Exynos designed by Samsung... you can't be sure that the process is the problem rather than the design. Apple's A8 is built on the same 20nm process and it seems to not have any problem.
That's a valid point, although very relative. The A8 throttles also, but doesn't have any competition within its ecosystem. Snapdragon 810 doesn't have absolutely any issues in regard of not being fast *enough*, it is merely not fast enough compared to its own competition.
The A4WP standard uses the wireless charging tech demonstrated by Witricity at TED in 2009 (http://www.youtube.com/watch?v=MgBYQh4zC2Y). I've been waiting years for this to come out. It's gonna be so cool. They've even demonstrated a solution for charging electric cars. Just place the wireless charger on your garage floor and you'll never have to think about plugging in your car at home ever again.
I can easily see Apple integrating a wireless charger into their iMacs so you can charge all your iDevices just by sitting at your desk.
I thought category 10 LTE (aka LTE Advanced) was supposed to get 1 Gbps with 66 MHz of spectrum. Instead they're showing off 450 Mbps with 60 MHz, which is only half of the original spec. Wth happened? Is there a category 11?
The MT8173 is very impressive, indeed! And a notable departure from the usual 8 x A7 Mediatek. Be it labeled midrange or not, I'd prefer the CPU config of the MT8173 over the current Snapdragon or Exynos high end SoCs.
At launch Core M was priced at some 280$ , A Tegra was 30$ and less at launch , something high end with integrated LTE is bellow 50$. So Core M is one the stupidest SoCs ever made.
Maybe on a single run, but what about sustained performance?
I highly doubt your claims. Remember the base clock of Core M. It is an extremely compromised design. And again, as mentioned in the other comment, in performance per price, there's absolutely no contest between the two. The difference will be ridiculous.
Will be very interesting to see benchmark results for devices with this SoC. 2+2 big.LITTLE is very appealing, at least on paper. Hopefully it translates into hardware as well as the theory says it should.
I tried SVP and or some scenes it worked very well (turning scrolling text into smoothly moving data at 24 fps instead of a jittery mess), but there were too many occasions where the movements looked really weird, too fast and unnatural. After trying different settings for some time I gave up. Kudos to Mediatek if tey got it right. It's a shame, though, that probably noone is interested in bringing this to traditional PCs without selling new hardware.
If MediaTek wants to sell to the other half of the US market now (as opposed to in a few years down the road), they need a CDMA2000 solution. Of the CMDA carriers, only Verizon has a more or less nationwide LTE network; and even they still have a handful of areas that they need to roam onto someone else's 2/3g network to maintain coverage.
Sprint, US Cellular, are nowhere near having LTE installed in all of their towers yet; and both are currently dependent on 2/3g roaming service from VZW to fill a lot of rural gaps. Until VZW is willing to offer LTE roaming, both of them will need to maintain CDMA 2/3g support even after upgrading all of their infrastructure. I suspect Verizon won't be willing to allow basic LTE roaming until either after they have nationwidish LTE-A that they won't share (to make sure their network is enough better to justify the premium prices they charge), or until they start shutting down their 2/3g networks to reuse the spectrum. The latter probably won't be happening anytime soon; back in 2012 they committed to maintaining 2g until 2021. OTOH they recently switched off 3g in Cleveland/Manhattan to reuse the spectrum for LTE. CDMA doesn't have a 3g voice capability, so it's the 2g network that they have more need to keep working.
Following up on my last post, VZW is only refarming half of it's 3g spectrum in the test areas; and has committed to keeping their 3g EV-DO network online through the end of 2019.
Both their 2 and 3g legacy networks only need 2.5mhz of spectrum to work; so we'll probably see additional intermediate refarming steps over the next 5 years as both networks are progressively wound down.
I guess MediaTek also needs CDMA2000 for China Telecom. That network is slowly rolling out LTE but they'll still be using CDMA for voice for years to come.
Qualcomm pretty much dominates the Chinese market at the midrange and high end, especially with China Telecom CDMA devices. It's funny that China has a similarly balkanized network environment as the US with OEMs making separate phones for each network. The Chinese market Galaxy S5 variants all run on Qualcomm, with the China Telecom version capable of dual SIM operation on TDD-LTE, FDD-LTE, CDMA2000 and WCDMA/GSM. Crazy stuff, even the Korean market variant doesn't have that global capability.
They actually have CDMA2000 (from a collaboration with VIA Telecom) in the new MT6735 and MT6753 SoCs that should be already shipping and will hit retail soon.
Noticed Miravision a few weeks ago in a shipping device (Elephone p6000 i think) but can't really find a list of everything it allows you to adjust and no clue if any popular colorimeters can take advantage of it. As for MT8173 some benchmarks showed up already but likely early software and not final clocks so retail perf should be plenty better (or more) https://gfxbench.com/device.jsp?os=Android&api... http://browser.primatelabs.com/geekbench3/1656634
What is the efficiency of these cordless charging systems? Or more specifically the losses as compared to wired charging? It's only a matter of time before the greenies poo-poo them if they aren't like 98% efficient;)
Even if it is prototype/early silicon, seeing the A72 this early is surprising. There is a chance end users could see A72 products this year while I was originally expecting them in 2016. ARM must have picked up their cadence behind the NDA curtain for this to happen.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
34 Comments
Back to Article
aegisofrime - Sunday, March 8, 2015 - link
Any details on what process the SoC is fabbed at?Novacius - Sunday, March 8, 2015 - link
28nm TSMC, but I don't know which process exactly.darkich - Monday, March 9, 2015 - link
Did MTK even announced anything beyond 20nm?The TSMC 20nm planar process seems quite bad, seeing how underperforming the Snapdragon 810 turned out.
So while this is great news because it will bring a powerhouse chip for the low price, I don't see Mediatek competing in high end with Samsung and Qualcomm until they sign some advanced fab deal.
CiccioB - Monday, March 9, 2015 - link
Problem may not be the process but the design.810 was a rushed SoC as Qualcomm found itself quite behind the competitors with no 64 bit architecture while professing that 64bit was useless at this point of mobile market evolution. We can see Qualcomm wrong strategy in the fact that they will have their own 64bit custom architecture quite late this year (or early next one) and had to use ARM standard IP for their top tier SoC.
I would like to know how Qualcomm custom architecture is going to compare with ARM A72 under the point of performances, die size and power. For sure Qualcomm has to design a very good architecture or they risk to go out with an already obsolete design . Though good efficient architecture like the one created by Apple or nvidia require more die space = more expensive. And less cores. Will the cost justify the performance and most of all, how to fight against Cinese/Taiwanese producers that can just put on paper more and more cores with more and more GHz as marketing fud?
darkich - Tuesday, March 10, 2015 - link
I think that the TSMC process is obviously the main issue here.Compare the performance of Snapdragon 810 with 14nm finFET Exynos that uses the same CPU setup, yet arguably less advanced interconnect and power management - Exynos pulls ahead by a mile, seemingly throttling much less.
CiccioB - Tuesday, March 10, 2015 - link
Snapdragon designed by Qualcomm, Exynos designed by Samsung... you can't be sure that the process is the problem rather than the design. Apple's A8 is built on the same 20nm process and it seems to not have any problem.darkich - Tuesday, March 10, 2015 - link
That's a valid point, although very relative.The A8 throttles also, but doesn't have any competition within its ecosystem.
Snapdragon 810 doesn't have absolutely any issues in regard of not being fast *enough*, it is merely not fast enough compared to its own competition.
nathanddrews - Sunday, March 8, 2015 - link
A4WP is what Intel demoed at IDF last fall, isn't it? When can we expect consumer products and adapters? Skylake?Pissedoffyouth - Sunday, March 8, 2015 - link
If they focused on 700mhz APT LTE they'd make a killing, as these networks are already live in NZ and Australia.Michael Bay - Sunday, March 8, 2015 - link
Australia is hardly a prime market for anything.Gigaplex - Sunday, March 8, 2015 - link
Certainly far more prime than NZ.Pissedoffyouth - Monday, March 9, 2015 - link
https://en.wikipedia.org/wiki/Asia-Pacific_Telecom...Bigger market than the US has for it's LTE networks, mate.
sonicmerlin - Sunday, March 8, 2015 - link
The A4WP standard uses the wireless charging tech demonstrated by Witricity at TED in 2009 (http://www.youtube.com/watch?v=MgBYQh4zC2Y). I've been waiting years for this to come out. It's gonna be so cool. They've even demonstrated a solution for charging electric cars. Just place the wireless charger on your garage floor and you'll never have to think about plugging in your car at home ever again.I can easily see Apple integrating a wireless charger into their iMacs so you can charge all your iDevices just by sitting at your desk.
sonicmerlin - Sunday, March 8, 2015 - link
Also the PMA and A4WP groups have agreed to merge in 2015: http://www.powermatters.org/new-in-the-media/338-c...nathanddrews - Monday, March 9, 2015 - link
No we just need to bring in the Wireless Power Consortium (Qi) and start selling some home upgrade kits!sonicmerlin - Sunday, March 8, 2015 - link
I thought category 10 LTE (aka LTE Advanced) was supposed to get 1 Gbps with 66 MHz of spectrum. Instead they're showing off 450 Mbps with 60 MHz, which is only half of the original spec. Wth happened? Is there a category 11?MrSpadge - Sunday, March 8, 2015 - link
The MT8173 is very impressive, indeed! And a notable departure from the usual 8 x A7 Mediatek. Be it labeled midrange or not, I'd prefer the CPU config of the MT8173 over the current Snapdragon or Exynos high end SoCs.darkich - Monday, March 9, 2015 - link
Indeed, this should be roughly as fast as the newest Exynos in multi, but much faster in single core.The single core performance should actually be higher than even in the A8X or Core M chips.
Novacius - Monday, March 9, 2015 - link
Core M should still be faster. The 5Y71 clocks with up to 2.9 GHz and has a much higher IPC, but even the 5Y10 should surpasss an A72.jjj - Monday, March 9, 2015 - link
At launch Core M was priced at some 280$ , A Tegra was 30$ and less at launch , something high end with integrated LTE is bellow 50$.So Core M is one the stupidest SoCs ever made.
darkich - Tuesday, March 10, 2015 - link
Maybe on a single run, but what about sustained performance?I highly doubt your claims. Remember the base clock of Core M. It is an extremely compromised design.
And again, as mentioned in the other comment, in performance per price, there's absolutely no contest between the two. The difference will be ridiculous.
darkich - Tuesday, March 10, 2015 - link
..oh, and also in performance per watt.The A72 is a smartphone core, far more efficient than A57
phoenix_rizzen - Monday, March 9, 2015 - link
Will be very interesting to see benchmark results for devices with this SoC. 2+2 big.LITTLE is very appealing, at least on paper. Hopefully it translates into hardware as well as the theory says it should.MrSpadge - Sunday, March 8, 2015 - link
I tried SVP and or some scenes it worked very well (turning scrolling text into smoothly moving data at 24 fps instead of a jittery mess), but there were too many occasions where the movements looked really weird, too fast and unnatural. After trying different settings for some time I gave up. Kudos to Mediatek if tey got it right. It's a shame, though, that probably noone is interested in bringing this to traditional PCs without selling new hardware.DanNeely - Sunday, March 8, 2015 - link
If MediaTek wants to sell to the other half of the US market now (as opposed to in a few years down the road), they need a CDMA2000 solution. Of the CMDA carriers, only Verizon has a more or less nationwide LTE network; and even they still have a handful of areas that they need to roam onto someone else's 2/3g network to maintain coverage.Sprint, US Cellular, are nowhere near having LTE installed in all of their towers yet; and both are currently dependent on 2/3g roaming service from VZW to fill a lot of rural gaps. Until VZW is willing to offer LTE roaming, both of them will need to maintain CDMA 2/3g support even after upgrading all of their infrastructure. I suspect Verizon won't be willing to allow basic LTE roaming until either after they have nationwidish LTE-A that they won't share (to make sure their network is enough better to justify the premium prices they charge), or until they start shutting down their 2/3g networks to reuse the spectrum. The latter probably won't be happening anytime soon; back in 2012 they committed to maintaining 2g until 2021. OTOH they recently switched off 3g in Cleveland/Manhattan to reuse the spectrum for LTE. CDMA doesn't have a 3g voice capability, so it's the 2g network that they have more need to keep working.
http://www.techtimes.com/articles/21527/20141204/v...
DanNeely - Sunday, March 8, 2015 - link
Following up on my last post, VZW is only refarming half of it's 3g spectrum in the test areas; and has committed to keeping their 3g EV-DO network online through the end of 2019.Both their 2 and 3g legacy networks only need 2.5mhz of spectrum to work; so we'll probably see additional intermediate refarming steps over the next 5 years as both networks are progressively wound down.
https://gigaom.com/2014/12/03/verizon-starts-killi...
https://gigaom.com/2014/12/05/verizon-is-transitio...
serendip - Monday, March 9, 2015 - link
I guess MediaTek also needs CDMA2000 for China Telecom. That network is slowly rolling out LTE but they'll still be using CDMA for voice for years to come.Qualcomm pretty much dominates the Chinese market at the midrange and high end, especially with China Telecom CDMA devices. It's funny that China has a similarly balkanized network environment as the US with OEMs making separate phones for each network. The Chinese market Galaxy S5 variants all run on Qualcomm, with the China Telecom version capable of dual SIM operation on TDD-LTE, FDD-LTE, CDMA2000 and WCDMA/GSM. Crazy stuff, even the Korean market variant doesn't have that global capability.
jjj - Tuesday, March 10, 2015 - link
They actually have CDMA2000 (from a collaboration with VIA Telecom) in the new MT6735 and MT6753 SoCs that should be already shipping and will hit retail soon.jjj - Monday, March 9, 2015 - link
Noticed Miravision a few weeks ago in a shipping device (Elephone p6000 i think) but can't really find a list of everything it allows you to adjust and no clue if any popular colorimeters can take advantage of it.As for MT8173 some benchmarks showed up already but likely early software and not final clocks so retail perf should be plenty better (or more)
https://gfxbench.com/device.jsp?os=Android&api...
http://browser.primatelabs.com/geekbench3/1656634
Hulk - Monday, March 9, 2015 - link
What is the efficiency of these cordless charging systems? Or more specifically the losses as compared to wired charging? It's only a matter of time before the greenies poo-poo them if they aren't like 98% efficient;)Hulk - Monday, March 9, 2015 - link
Doing some quick searches reveals lots of information but no real world efficiency numbers. That leads me to believe it's not good.DanNeely - Monday, March 9, 2015 - link
I have some vague recollections of 50-60% being typical.Kevin G - Monday, March 9, 2015 - link
Even if it is prototype/early silicon, seeing the A72 this early is surprising. There is a chance end users could see A72 products this year while I was originally expecting them in 2016. ARM must have picked up their cadence behind the NDA curtain for this to happen.bigstrudel - Wednesday, March 11, 2015 - link
2x High Power + 2x Low Power cores is a very smart configuration.4x High Power cores is just overkill and the battery life always suffers hugely because of heat.