Samsung launches its first 5G-integrated CPU
- Thread starter JournalBot
- Start date
Not sure if facetious...
Bixby is software, not hardware. But Bixby Lens, which is exactly as much like Google Lens as it sounds, will take direct advantage of the NPU for much more rapid response times when invoked. The NPU is also used directly by the camera app itself to recognize items visually, cuing it what to "focus" on automatically and how to choose scene processing types according to what classes of object are in frame.
Upvote
6
(23
/
-17)
You know, if my phone could reliably do 10-20 mbps, consistently, there are very few times that I would not be satisfied with that.
That's more than fast enough for video at resolutions that will be clear on a 6" screen, and large downloads like OS and app updates generally happen in the background anyway (or on wi-fi), so I'm not really waiting for it.
We'll see how 5G matures, but for me, for now, it's not a big selling point.
That's more than fast enough for video at resolutions that will be clear on a 6" screen, and large downloads like OS and app updates generally happen in the background anyway (or on wi-fi), so I'm not really waiting for it.
We'll see how 5G matures, but for me, for now, it's not a big selling point.
Upvote
57
(57
/
0)
Upvote
103
(104
/
-1)
We wouldn't have congestion if the government owned the airwaves and tower infrastructure and leased them to providers. The problem with the current setup is that each provider owns a different set of airwaves in multiple markets. For example, a provider may have 600 MHz in 1 market and 700 MHz in another.
If competitors competed over the top, phones could automatically shift to a less congested portion of the spectrum. This would also solve another issue: competition.
EDIT: For the anti government types, the federal government provides most of the funding for the maintenance and expansion of the interstate highway system. While I'm not at all a socialist, services such as healthcare, internet connections, and cellular talk/text/data, all of which are critical to today's society, should definitely be handled by one entity.
If competitors competed over the top, phones could automatically shift to a less congested portion of the spectrum. This would also solve another issue: competition.
EDIT: For the anti government types, the federal government provides most of the funding for the maintenance and expansion of the interstate highway system. While I'm not at all a socialist, services such as healthcare, internet connections, and cellular talk/text/data, all of which are critical to today's society, should definitely be handled by one entity.
Upvote
54
(63
/
-9)
Chemical Tribe
Ars Legatus Legionis
Socialist.
Seriously don't let the shit heads on the right tar and feather good policies.
Upvote
34
(41
/
-7)
Bixby is Also AI and leverages blockchain crytpto currency foolsgold (now selling at $.000000001324323 per coin, and it is going to moon shot soon)
Calling Bixby anything but annoying is giving it way to much credit.
Also what I find best about this isn't the intergrated 5G, it is 5G modem at 8nm. building all of these chips on the latest dies will definitely lower costs, and power usage.
Upvote
-8
(2
/
-10)
I mean, the 'non-critical' parts of the economy probably amount to less that 30% of GDP... so don't worry about disavowing the socialism label at that point.
More on topic, Anandtech has done a lot of analysis on the custom core architecture Samsung has been building up and they have sounded very promising the last two generations, but each with a fumbled element that keep them from being competitive with Qualcomm and Apple's best, hopefully these rectify those problems and we can get another good SoC in the race.
Upvote
13
(13
/
0)
How did Intel drop the ball? They used to be process leaders and now they're at least 1 generation behind the Taiwanese.
As the article mentions, that difference has put AMD demonstrably ahead of Intel..
Right now, it doesn't matter that much as the Taiwanese are friendly with the U.S. However when the Chinese inevitably invade Taiwan, we're going to lose access to cutting edge tech.
As the article mentions, that difference has put AMD demonstrably ahead of Intel..
Right now, it doesn't matter that much as the Taiwanese are friendly with the U.S. However when the Chinese inevitably invade Taiwan, we're going to lose access to cutting edge tech.
Upvote
-11
(4
/
-15)
The Chinese are never going to invade Taiwan. They prefer having cities and not nuclear craters.
Upvote
1
(7
/
-6)
It's not even on my radar, and give how it works, I don't think having it in a phone is worth the extra cost of having it in the phone.
The site-to-site line of vision thing to a ground station/repeater/etc. makes sense, because then one can switch to wifi and be fine. But a CELL PHONE? I'm truly having a hard time wrapping my head around the reasoning behind putting this in a cell phone to begin with.
Perhaps I missed the memo (that happens a lot), but how would 5G be superior to 4G under normal cell phone use when it's almost certain that your cell phone won't be able to pick up a 5G signal in the first place. Or is that a mistaken impression of its capabilities as applied to a cell phone under normal use circumstances?
I'm apparently not connecting dots there...
With respect to the processor scale and SOC layout, that's kind of neat. Faster is better in anything electronic. And just out of curiosity, is that putting the modem in the SOC a trend in the rest of the industry a result of the anti-trust decree against Qualcomm?
Upvote
4
(4
/
0)
Shirley, you're joking. The one I know of isn't even willing to work with his own appointed staff and keeps firing them instead.
Upvote
53
(55
/
-2)
Why does this article make repeated comparison references to the last-gen Snapdragon 845 (as used in the Pixel 3), but not the current-gen Snapdragon 855? The Snapdragon 855 uses TMSC's 7nm process, and has a much better GPU than the Snapdragon 845. I'm interested in how the Exynos 980 compares to the Snapdragon 855 with a separate 5G modem.
We'll also probably see the Snapdragon 865 with an integrated 5G modem announced in December at Qualcomm's annual summit in Hawaii.
We'll also probably see the Snapdragon 865 with an integrated 5G modem announced in December at Qualcomm's annual summit in Hawaii.
Upvote
22
(22
/
0)
I'm not seeing the use case for millimeter transceivers in handheld devices myself, but in the sub-6GHz band where these chips operate the 5G protocols do allow for a more efficient use of the bandwidth than 4G does. As an in-place replacement of 4G it would be worth doing, if it weren't for all the already deployed incompatible hardware.
Upvote
9
(9
/
0)
"5G" refers to two entirely separate frequency ranges, FR1 and FR2. The really big speed increases happen at FR2, which is millimeter-wavelength, and you're exactly correct—it's pants-on-head ridiculous to put in a phone, which does not seem to stop vendors from putting it in phones anyway.
5G FR1 is sub-6GHz, and has roughly the same penetration and propagation characteristics as existing 4G. It's also not necessarily a whole lot faster than 4G; 4G is theoretically capable of 1Gbps (to a stationary target) already. The proof will be in the pudding as to how much of an improvement 5G FR1 really is once the carriers are supporting massive numbers of 5G customers, the way they are supporting massive numbers of 4G customers now.
Upvote
13
(15
/
-2)
Do Samsung SoCs ever show up in phones for North America? Even Samsung's own phones tend to use Qualcomm for the NA version.
Upvote
7
(7
/
0)
Ah, okay, thanks for that explanation. I didn't realize 5G was spread out over that much spectrum. It makes much more sense now.
Still not a feature I'd be looking for, but I can see how it would be more useful than I thought it'd be.
Upvote
1
(1
/
0)
Well,, the article would have been better served if author just sited the source press release from Sammy. However, the article did cover the issue with the FR2 (mm waves) that makes it not so critical in mobile handset use.
Only area where it may be handy in a mobile phone might be in stadiums. ball parks and concert venues - so it might facilitate decent phone calls and allow Uber/Lyft some data to arrange smooth pickups near major sporting events or music festivals. IF that's your thing, don't hold your breath, as networks still may take few years to deploy them widely (and who knows for mobile cellsites) - past meaningful lifecycle for the chip that peaks in first couple years.
Imagine if everyone had meaningful data connectivity at Coachella - the havok it would wreck on FB/IG servers....
Upvote
2
(2
/
0)
So whose foundry is doing 8nm? Samsung's own? Was hoping for more detail on this, as I've mainly heard of 7nm and 10nm with the other big players.
Upvote
1
(1
/
0)
It's not like we had four previous generations to figure out that for one or two years there is nothing to be gained besides less money, battery and sometimes convenience.
But for some reason every single new generation that comes along it all starts again.
Upvote
2
(2
/
0)
Nobody is going to get nuked, but China IS going to continue to integrate with Taiwan.
They had a damn good run but no civil war lasts forever, especially when we're talking about the countries that are #56 and #1 on the global population lists, and the main two who speak Chinese.
I'm just hoping Taiwan brings more respect for human rights to China's mainland, now that they are modernized enough to be able to afford that (when most of your population lacks modern diet and medicine, heirarchy of needs kicks in).
https://en.wikipedia.org/wiki/Chinese_Civil_War
https://en.wikipedia.org/wiki/List_of_c ... ed_Nations)
https://en.wikipedia.org/wiki/List_of_o ... _territory
Note that footnote.
Upvote
-9
(1
/
-10)
Nah they went for a fixed bridge design instead.
(hint a reference to the misspelling)
Upvote
5
(5
/
0)
I'm pretty sure Samsung's own Galaxy S series are more popular than the Pixel phones and fairly certain "big-deal kid" would be the S855 since its in basically all the major phones this year (and almost certainly will be in the Pixel 4 that is probably due out soon). What makes your comment even more bizarre is that you referenced the S8 and S9+ but ignore the S10 series (which has been out for I think 6 months now) or say the just released Note 10.
Samsung even has recently announced a new Exynos chip on their own 7nm process that uses EUV. Its basically a die shrunk version of the one they had on 8nm (they had 8nm Exynos out in the non-US S10 series).
And the bits at the end about AMD and Intel are just...weird since they don't really add much. I'm a little surprised you didn't reference how integrating the modem into the SoC was a big reason why Qualcomm's Snapdragon SoC's were so popular (as that was the overriding reason, it saved companies costs - with Apple being one of the few holding out because they rely on their own processor design which necessitates using other companies' modem chips with rumors suggesting Apple is planning on doing their own 5G modem, which will leave them skipping 5G until then).
Sorry to be so critical (I think you're a new writer/contributor for Ars maybe?), and I'm guessing you're not as adept on hardware (I'm guessing based on the little blurb about you, that you're more of a software person, hence why you feel the Pixel 3 is the current popular reference)?
Upvote
9
(9
/
0)
Yeah because there is a fundamentally limited amount of wireless spectrum there can be only so much competition. You could never have 100 different carriers with 100 different networks in every location as we don't have the wireless spectrum for it and never could. Splitting the majority of it across 3 or 4 major cell providers just restricts most (google Fi is a little odd and there is roaming in some cases) people to using just the portion their carrier happens to have at their location. It also requires building a lot of redundant infrastructure in rural areas for coverage when it might not really be needed for providing enough bandwidth for the area. In cities that's handy for aggregate bandwidth to provide enough for everyone to use but in many rural area's that probably isn't needed other than everyone wants to claim they have coverage there.
A single carrier with everyone else basically being an MVNO would be the most efficient option from a wireless spectrum utilization perspective and from an infrastructure perspective. It doesn't have to be government run, private with heavy regulation to avoid monopoly abuses could work, but I don't see how the US gets to either one from where we are and with the power's that be swapping every 4-8 years.
Upvote
3
(3
/
0)
I'm not quite sure it's totally crazy on a phone. Very situational definitely. Premature to be throwing into phones now. I'd agree. It does have it's potential use cases in very high user density environments like Airports/ sports stadiums/convention centers/ casinos etc. You'd need a lot of transmitters due to the LOS issues and users would definitely be roaming on and off the mm band stuff quite a bit but it's hard to ignore the benefit of the large amount of spectrum available up in those bands. The LOS limitations becomes an advantage when you are trying to reuse the spectrum. Each conference room in a convention center could have a mm band base station in the ceiling with each room reusing the same spectrum.
Add on good FR1 5g coverage and you'll end up with a better experience for all. There will be lots of bandwidth in the FR2 stuff for anyone who is on it and that will mean less congestion on the FR1 bands so better results for everyone using that.
Upvote
6
(6
/
0)
Yes, its Samsung's own. Their 8nm is just a revision of their 10nm (and thus much less advanced than TSMC's 7nm, let alone Samsung's own 7nm (which uses EUV) or TSMC's 7+ (which uses EUV as well and so is substantially different from their 7nm; it is what the high end phone SoC's like Apple's newest A series in the iPhone that should be out shortly, will be utilizing).
Think of 8nm in relation to 10nm like 12nm was to 14 and 16nm. In the past it'd be called a half node, but these days all of that stuff is mostly marketing terms (and the nm numbers themselves have lost most of their meaning as often times most of the transistors aren't to that level; I believe TSMC's 7nm for instance is closer to Intel's 14nm than it is to Intel's upcoming 7nm - which will use EUV, and Intel's 7nm is probably be a bit better than the other 7nm based EUV processes from Samsung and TSMC; but those two will be onto 6 and 5 nm processes by the time Intel gets their 7nm products out, and so they'll be closer to equal - assuming Intel doesn't have the issues with 7nm that they've been having with 10nm).
Upvote
5
(5
/
0)
Samsung is using its own (Korean) tech. TSMC is the one that brought 7nm to AMD (not sure why the article commends AMD) and to Apple (in Sept).
If somehow, the trade war escalates enough that TSMC is in trouble, we as Americans have a lot more to worry about than just the best semiconductor process.
Upvote
0
(0
/
0)
Just commenting to say I appreciate Jim using the 'FR1' and 'FR2' terminology so we don't have the usual conflation of '5G' solely with its deployment on mmWave. Please continue this in future articles on 5G (where appropriate, of course).
Upvote
20
(20
/
0)
And it's not just the cellular 5G modem, also the connectivity (Wi-Fi and Bluetooth) on this mobile platform is integrated into the SoC itself.
This was also the case on last years S5E9610/S5E9609 (10nm), that was used for the Galaxy A50 and the Motorola One Vision among others.
Upvote
0
(0
/
0)
Because they got complacent and suffered through a generation or two of bad management.
All the fabs had issues in the past 10 years transitioning so its not even just Intel. Intel did well going 32->22->14 (mostly, 14 had some issues early on), but moving to 10nm has been a mess for them. During that time the other fabs had big issues going below 28nm (I think they even had some issues going to 28nm, and I think GF had some issues with 32nm which is partly what made AMD's early Bulldozer chips especially poor although most of that was the chip design). They initially even skipped 22/20nm because of all the issues they were having. Their 16 and 14nm processes went better but were actually closer to 20nm in most regards (so the xxnm numbers became mostly marketing terms and not terribly representative of the actual process size for most of the transistors made). Intel's 14nm is superior to TSMC's 16nm and Samsung (and GF, who licensed Samsung's 14nm after they couldn't get their own up and running quickly enough). I don't know if even 10nm at TSMC or Samsung was even that close to 14nm either. But with Intel having major issues with 10nm, they have advanced beyond Intel for the first time (I'd guess TSMC's 7nm is better than Intel's initial 10nm which was used in a chip last year).
There's been big changes in the process tech. FinFETs, and now the move to EUV, with both increasing the costs and complexity (newer processes now cost more per transistor, which it used to drop the cost per transistor with normal planar processes; which I think Intel moved to FinFETs pretty smoothly at 22nm). That's one of the reasons adding to the issues, there's extra complexity and cost required to design newer chips on newer processes. There was also a bottleneck in that the fabs had to wait for enough EUV machines to be made so they could even start to integrate them into their fabs.
I actually think that has less to do with it than AMD's design. Much like how poor design made Bulldozer worse than GF's 32nm (which was worse than Intel's 32nm process used for Sandy Bridge; the Phenom 2 chips which used GF's 32nm for instance had higher IPC than Bulldozer), which exacerbated the issues. AMD made some smart design decisions that made Zen 2 a big improvement (for instance the I/O dice, which are actually made on GF's 14 and 12nm), more than they'd have gotten from basically just shrinking Zen 1 to 7nm. We really see the benefits on the server side where AMD absolutely destroys Intel in performance (let alone perf/$) with their 64 core EPYC chips, because most of those decisions especially pay off there.
On consumer side, Intel's topology and clock speed advantage enables them to match AMD (the 8core 16 thread 9900K at 5GHz can match or even outdo AMD's 3900X 12 core/24 thread CPU in many tasks including multi-threaded; we'll see how the higher clocked and even higher core/thread count 3950X does; incidentally, its possible that Zen 2 Threadripper might be the gaming king if it gets the highest binned chiplets, coupled with the extra cache and memory channels over Ryzen; Intel's higher core count chips use a different topology and don't clock as high which is why the 7700/8700/9900K chips have been the top performing gaming chips these past few years). AMD made changes to Zen 2 to help mitigate issues of the chiplet design (namely extra large caches to reduce latency issues), which is why Zen 2 saw an especially big boost in gaming performance compared to Zen 1.
Samsung isn't Taiwanese. Even GF (which is in US/Europe) was going to leap frog Intel until they decided to call it quits (they were close to production with their 7nm process, they would've probably beat Samsung since the latter chose to wait for EUV for their 7nm process, whereas TSMC and GF were waiting til after their first 7nm processes to do that).
Like I said, Samsung would be an option, and it seems that GF's owner is probably going to spin them off to someone who could finish their 7nm. And they could license tech from elsewhere. The improvements to process technology come about via worldwide development, so if China tries to lock others out from TSMC, then TSMC will probably fall behind as China has tended to lag behind in developing stuff like that on their own. Plus if China did that, I'm pretty sure the US government would fund fabs. Heck, GF's owner is trying to leverage the current US administration's anti-Chinese sentiment to squeeze money out of a bunch of companies through patent litigation (claiming TSMC is using those patents without a license).
Upvote
5
(5
/
0)
Statistical
Ars Legatus Legionis
5G means more aggregate throughput for the tower. It is funny how the people saying I don't care about gigabit cellular I just want a reliable 20 mpbs can't see that the way you get a reliable 20 mpbs per user for hundreds of users using a tower means a tower with hundreds of megabits if not gigabits of throughput. As you are all the other users on the tower use more and more data the tower gets more congested.
You can't have one without the other. If you want a reliable 20 mpbs you can either reduce the number of users or increase the tower throughput. Which do you think is more likely going forward.
I am not saying rush out and buy a new 5G smartphone this is going to be a long slow process but the only way forward is more throughput because the number of users is going to increase, the number of devices is going to increase, and the data used per user and per device will continue to increase.
Upvote
10
(10
/
0)
Here's a non-disruptive fix: spectrum auctions would only be made to tower companies that agree to re-sell bandwidth on a competitive, non-exclusive basis.
Eg, the companies that now actually build & operate the towers would continue to do so; they would not let a given carrier monopolize that location.
I don't know how much spectrum goes idle at any time at any given location, but it seems like ordinary queueing theory would suggest, “a lot.”
Meanwhile, smaller competitor wireless carriers would not be locked out of competing with the oligopolistic incumbents. This would mean the tower companies would be selling actual usage, not a carrier's right to exclude competition. I can't imagine this would do anything other than encourage a better fit between cell size and demand, leading to less congestion.
So yes, you're on the right track. And it doesn't even involve any socialism; quite the opposite if you see oligopolies as socialism for business.
Upvote
4
(4
/
0)
I just want to see them expand the coverage and fix the packet loss.
Most of the time I'd be very happy with 5Mbps speeds...I was overall usually quite happy with 3G (which seemed to run around 2Mbps peak). My biggest complaint is the lack of 4G coverage combined with phones insisting on locking onto the weakest shred of signal often leaving me with 60-100% packet loss on the "super fast" connection.
When 4G is working properly and I'm in an area near a tower it has no issue pushing 50-60Mbps which is fast enough for anything I can think of even that I'd want to do on a wired machine. I think the record I've seen was around 120-140Mbps on a weekend when the area (which is mostly people who commute here to work) was mostly deserted and I probably had a whole tower to myself (I was in a parkinglot close to the tower that serves most of this area)
Upvote
2
(2
/
0)
Statistical
Ars Legatus Legionis
4G only support gigabit in theory. Not a single carrier globally has the frequency blocks wide enough to make that a reality. It is kinda like saying traffic in LA wouldn't be bad if they just put in a 80 lane highway. Sure with an 80 lane highway there would be no congestion but in any practical scenario you are going to have a lot less lanes and thus congestion.
5G supports a higher level of carrier aggregation (bonding multiple frequency bands together). It also supports a higher level of MU-MIMO (more spectral streams). That combined with the new wider sub 6 Ghz frequency bands being auctioned off and reusing the spectrum used for 2G/3G/4G as those are sunset means 5G networks should have more frequency to work with and should be able to use it more effectively. More Hz and more bits/Hz.
I mean since 5G-NR is simply an evolution of LTE in the same way 802.11ax is an evolution of 802.11ac which is an evolution of 802.11n you could add all these capabilities, and new frequency bands to LTE but in the end you would end up with "new LTE" hardware required both in the user handset and on the tower anyways so it would really just be the same 5G just called "super LTE" or something else.
Upvote
5
(5
/
0)