Samsung's new M3 Cores: Super Wide, True Apple Competitors at Last?

[Inversion]

14 hours ago, hey_yo_ said:

I’d rather see Samsung make some Windows 10S devices using their own in house silicon. I wonder if the cache and IPC are on par with Apple as those are two of the reasons why Apple is ahead of Qualcomm. 

Screw 10S - If this is as good as it seems I'd like to see it run full windows 10. Could probably provide a decent alternative to Intel m3 and maybe up to Pentium on the laptop side. Correct me if I'm wrong but I'd still want to have use of Google Chrome which is why no Windows 10S

[Trixanity]

4 minutes ago, LAwLz said:

Now if Samsung could put out a good tablet with this chip I'd be very happy. It's amazing how no company except Sony seems to know how to make a tablet, and they gave up on that idea.

Possibly next year. Samsung gives tablets the smallest priority. They tend to let last year's tech trickle down while doing the bare minimum to design the thing. I'd say they release it just to release something in case the market takes off again for some reason (I mean where is the incentive right now? Google doesn't seem to give a fuck).

[LAwLz]

9 minutes ago, Trixanity said:

Possibly next year. Samsung gives tablets the smallest priority. They tend to let last year's tech trickle down while doing the bare minimum to design the thing. I'd say they release it just to release something in case the market takes off again for some reason (I mean where is the incentive right now? Google doesn't seem to give a fuck).

Considering the fact that Samsung hasn't released a good tablet since... ever? I am not holding my breath.

[Bouzoo]

40 minutes ago, LAwLz said:

It's amazing how no company except Sony seems to know how to make a tablet, and they gave up on that idea.

iPads. You can say whatever you want, but some models are awesome. My gf has an Air 2, and that thing still rocks. And I'm a no tablet person in general. It is one of few areas where I give Apple credit.

 

[Eroda]

15 hours ago, Airdragonz said:

Interesting. Yet at the same time, I don't think it will beat Apple in benchmarks or speeds.

1) Samsung has to throttle the cpu's from being too much faster than the Snapdragon 845 unless they want an outcry on their hands from their U.S. based customers.

2) Apple iOS is optimized purely for Apple iPhones and their bionic chipsets. The iOS system is much faster and lightweight than Samsung's custom OS which is a modified version of Android (forgot the name of the os). If Samsung starts shipping their phones with these processors and a completely stock and stripped down Android Oreo, then they will have a chance to actually get ahead Apple

well thats ok if its much faster and they have to lower the clock speeds substantionaly itll increase battery life 

[Drak3]

32 minutes ago, LAwLz said:

Considering the fact that Samsung hasn't released a good tablet since... ever? I am not holding my breath.

OG Tab S, Note 8.0, Note 10.1 2014. Those were great tablets

[79wjd]

5 minutes ago, Drak3 said:

OG Tab S, Note 8.0, Note 10.1 2014. Those were great tablets

I hated my Note 8.0, it felt very stuttery, plus the stylus was really too small to be comfortable to suit it's purpose of writing more than the occasional short note. I get that there are limits to the size of a stylus you can fit in the tablet itself, but still...

[Commodus]

3 hours ago, Trixanity said:

Apple's gains in core design are slowing. That's why they're working on many uncore things to make sure the gains are still there. Besides they're on different schedules so it's only natural for Apple to be ahead after the fact.

 

There's probably a bigger reason to upgrade from an S8 to S9 than there was from iPhone 7 to 8. There are a lot of changes here and there (slight design changes, cameras, software to name a few).

They are slowing down, but we'll see -- I'm curious to know whether or not Apple can wring more out of the A12 when it doesn't have the luxury of a new manufacturing process.  And yes, there's a definite leapfrog process.  To me, it's just amusing that there's a tendency on both sides to act as if CPU upgrades confer some kind of permanent lead, forgetting that the competing option is usually just half a year away.

 

I'll agree on the S8/S9 versus iPhone 7/8 split, although I actually see the S9 as Samsung's iPhone 7 upgrade: it's iterative, but it adds a bunch of sensible features people had wanted for a while, such as stereo speakers and a rear fingerprint reader in a better place.  Except there's a headphone jack, of course.

[D13H4RD]

I'll say this.

 

Don't get your hopes on the S9 having 2x the performance of the S8. While it is impressive, it's likely Samsung will release an SM-G960U/G965U with a Snapdragon 845 for the US market and the SM-G960F/G965F with the Exynos 9810 for the international market.

 

The reason I say this is because Samsung has to maintain feature and (theoretical) performance parity between the 2 so that one is not overly "superior" to the other. The Exynos 8895's DSP for instance can actually do UHD video at 60FPS, but Samsung chose to not implement it for the Exynos because it would make the SD version seem inferior.

 

In all manners of honesty, the Exynos Samsungs are still the ones to get IMHO. They just seem to run just that bit better.

[LAwLz]

25 minutes ago, Bouzoo said:

iPads. You can say whatever you want, but some models are awesome. My gf has an Air 2, and that thing still rocks. And I'm a no tablet person in general. It is one of few areas where I give Apple credit.

The iPad is fantastic, except for two huge issues I have with it.

1) It runs iOS. I am not exactly a fan of that but I beggers can't be choosers so I could get used to it. It shouldn't be too hard finding alternatives to the apps I use.

 

2) The screen is horrible for almost everything I want to use it for. Whoever decided that 4:3 was a good aspect ratio for a tablet deserves to be fired. It should have been 16:10 or 3:2 instead.

 

 

16 minutes ago, Drak3 said:

OG Tab S, Note 8.0, Note 10.1 2014. Those were great tablets

OK they have released some. Not in the last ~3 years though.

Can't agree with the Note 8.0 (too small) but the other ones were good at launch.

[DocSwag]

10 hours ago, Commodus said:

The problem if Samsung manages to catch up: congratulations, you're roughly even with Apple for 6 months, until the A12 comes out and puts Apple ahead again.  Most likely, the Exynos 9810 would need to have a clear lead over the A11 to make Apple nervous about the speed of its 2018 iPhones.

 

Having said all that: if there's a dramatic speed boost, it's still good news for Galaxy S fans.  The S9 sounds like it'll mostly be an iterative upgrade, so having a tangible speed difference will make it easier to justify... at least, if you're coming from an S7 or earlier.

IMO it's not meant to make apple sweat but rather other people like Qualcomm.

8 hours ago, Sniperfox47 said:

Does everyone forget that Denver V2/V3 exists? Or does it just not get counted because it's not *technically* using the ARM ISA?

Denver was interesting but it's no longer being used in smartphones plus the weird compiler shenanigans sometimes could reduce performance a fair bit in certain workloads.

1 hour ago, Sniperfox47 said:

They're in pretty much every self driving car being marketed in the past two years, and a number of higher end commercial drones. How ubiquitous are Apple's cores?

They aren't in smartphones anymore.

[Commodus]

10 minutes ago, LAwLz said:

The iPad is fantastic, except for two huge issues I have with it.

1) It runs iOS. I am not exactly a fan of that but I beggers can't be choosers so I could get used to it. It shouldn't be too hard finding alternatives to the apps I use.

 

2) The screen is horrible for almost everything I want to use it for. Whoever decided that 4:3 was a good aspect ratio for a tablet deserves to be fired. It should have been 16:10 or 3:2 instead.

On iOS, I look at it this way: tablets are ultimately app machines, and the iPad app ecosystem is exponentially superior to what you see on Android.  It doesn't matter if the Android slab has widgets or better notifications if half the apps you run are just stretched-out phone apps.

 

And I believe the 4:3 ratio is superior for a tablet.  16:10 is basically for people who treat their tablet like a dedicated Netflix viewer... and 3:2 means that page-sized documents don't really fit properly.  The iPad's ratio certainly isn't ideal for videos, but it works for browsing, document editing, books, magazines and the like.

[D13H4RD]

1 minute ago, Commodus said:

On iOS, I look at it this way: tablets are ultimately app machines, and the iPad app ecosystem is exponentially superior to what you see on Android.  It doesn't matter if the Android slab has widgets or better notifications if half the apps you run are just stretched-out phone apps.

 

And I believe the 4:3 ratio is superior for a tablet.  16:10 is basically for people who treat their tablet like a dedicated Netflix viewer... and 3:2 means that page-sized documents don't really fit properly.  The iPad's ratio certainly isn't ideal for videos, but it works for browsing, document editing, books, magazines and the like.

Unless that iPad is mega.

 

I own a 2015 12.9" Pro and it has gotten heaps of attention at campus almost solely due to its size.

[Drak3]

29 minutes ago, LAwLz said:

OK they have released some. Not in the last ~3 years though.

Can't agree with the Note 8.0 (too small) but the other ones were good at launch.

Yeah, the only great Android tablets released recently were the Surface Pro and Viao Z Canvas loaded with Android x86.

[themctipers]

15 hours ago, DocSwag said:

Or they can use exynos in all their phones, like they did with the S6. 

Samsung's "bloat" doesn't really affect benchmarks and stuff as evidenced by the S8, for example, not losing to the pixel 2. Rather, the bloat mainly impacts fps while doing stuff like scrolling through web pages, during which the SoC doesn't usually jump to max performance states. 

 

I don't think it really makes sense for an OS like ios to be able to improve processor performance. Rather, it would be down to optimizations on individual applications that would actually do this. If such optimizations at the OS level could improve performance, I think we'd see a bigger performance delta between the same SoC from different manufacturers. 

It's more of a priority thing I'm pretty sure. 

 

Focus more power on the user experience in springboard and it'll feel more fluid against having it be a FFA for cpu usage 

[Sniperfox47]

34 minutes ago, DocSwag said:

Denver was interesting but it's no longer being used in smartphones plus the weird compiler shenanigans sometimes could reduce performance a fair bit in certain workloads.

Denver2 (from Parker/the X2) was significantly better for that the the original Denver, and Carmel (from Xavier) is supposed to be pretty much identical to native performance according to what little I've seen of it. Do workloads without much repeating code still perform worse on Denver2? Yeah, but it's nowhere near as bad as it was with classic Denver on the Tegra K1. Not to mention you typically see Denver2 paired with ".Big" A57 cores which are perfectly suited for the kinds of code where Denver2 struggles.

45 minutes ago, DocSwag said:

They aren't in smartphones anymore.

Again, there's no reason they couldn't be. The only real con-side to them is their larger die size and more expensive price, which are the exact same issues Apple's SoCs have now. I mean if you're going to use "Well they're not used in that market!" as a way to say they're not comparable to other products, then we probably shouldn't be comparing *any* other chipsets to Apple's either, since they're never going to be put in a situation where they directly compete.

[LAwLz]

56 minutes ago, Commodus said:

On iOS, I look at it this way: tablets are ultimately app machines, and the iPad app ecosystem is exponentially superior to what you see on Android.  It doesn't matter if the Android slab has widgets or better notifications if half the apps you run are just stretched-out phone apps.

Well it depends on which apps you use. All the ones I use on my tablet scales very well for it, so for my use I don't think iOS is superior to Android because of the "app ecosystem". If anything I think switching to iOS would be inconvenient because of the lack of apps I use on Android. I could probably find alternatives or workarounds, but it would still be inconvenient for a while.

 

56 minutes ago, Commodus said:

And I believe the 4:3 ratio is superior for a tablet.  16:10 is basically for people who treat their tablet like a dedicated Netflix viewer... and 3:2 means that page-sized documents don't really fit properly.  The iPad's ratio certainly isn't ideal for videos, but it works for browsing, document editing, books, magazines and the like.

16:10 is better than 4:3 for:

Video watching - Less black bars.

Web browsing - Most websites are narrow and tall, not short and wide.

Gaming - Most games are made for widescreens (especially since most are ported from phones).

 

 

I don't think there is a big difference between the two for document, comic and book reading. 4:3 is slightly wider than the standard paper size, and 16:10 is slightly taller. 16:10 is further from the paper aspect ratio, but once you throw on the notification bar and the navigation bar it becomes very, very close.

 

I think 3:2 is a very good compromise.

 

 

25 minutes ago, Sniperfox47 said:

Again, there's no reason they couldn't be. The only real con-side to them is their larger die size and more expensive price, which are the exact same issues Apple's SoCs have now. I mean if you're going to use "Well they're not used in that market!" as a way to say they're not comparable to other products, then we probably shouldn't be comparing *any* other chipsets to Apple's either, since they're never going to be put in a situation where they directly compete.

The X1 has like 2-3 times higher heat output than high end ARM SoCs.

It would be interesting to see what power and performance would be like in a smartphone body, but I don't think it would be all that impressive.

[Sniperfox47]

26 minutes ago, LAwLz said:

The X1 has like 2-3 times higher heat output than high end ARM SoCs.

It would be interesting to see what power and performance would be like in a smartphone body, but I don't think it would be all that impressive.

Yeah, but most of that is the GPU. The X1 on the CPU side was 4 A57 cores and 4 A53 cores just like pretty much every other modern SoC at the time it came out.

 

And look at the GPU performance, the X1 still beats the Snapdragon 845 for pure GPU performance, and the 845 just came out 2 and a half years later. If you factory downclock and undervolt the GPU, it would totally be feasible to implement in a reasonable TDP for a high performance phablet (though likely not a small-factor phone).

 

The X2 (Parker) likewise could potentially be viable in a larger phone form factor and has Denver2 cores.

 

I doubt Xavier with it's massive Caramel cores and ridiculous GPU would though, I think it's targeting like the 30W range and ~300mm^2 die? That might be a little ridiculous for a phone or even a tablet.

[Doobeedoo]

It's a huge performance leap. Not long ago I was mentioning that I was sure someone will go with this route for future Android SoC soon and I expected Samsumg cause they were adamant with their custom designs. 

Really awesome to see parity with Apple A chips as far as core design in this way. Now it will be interesting to see how they handle power and they will probably make it closer to SD variant so gap is not that huge. Would be great if they could just use Exynos across the board. Lucky for me my area receives Exynos version. 

[Zodiark1593]

4 hours ago, LAwLz said:

 

That's not how OS schedulers work..

 

Actually, I think you're right. When the wide Core 2 Duo arrived, the benefits were immediately realized. If anything, how well the pipelines will be fed will probably be determined by the supporting hardware (OoO, Branch predictor, SMT, etc)

[DanielMDA]

19 hours ago, DocSwag said:

Source: https://www.anandtech.com/show/12361/samsung-exynos-m3-architecture

Ever since the A7 I think it's fair to say Apple has held a huge lead over everyone else on the market when it comes to CPU performance. Qualcomm and to a lesser extent Nvidia have remained competitive on the GPU side but due to Apple's use of very wide custom designed ARM CPU cores it has maintained a very large cpu performance advantage over everyone else, with estimates from benchmarks like geekbench 4 implying that Apple could be enjoying as much of a 2x performance lead compared to every single other major mobile CPU on the market.

 

However, it seems that this may be no more. With the new M3 cores Samsung is advertising as much as a 2x performance jump over the prior generation M2 cores, which would most likely put performance very close to that of the high performance cores in Apple's new A11 Bionic chip used in the iPhone 8, iPhone 8 Plus, and iPhone X. 

 

In order to do this they've not only raised clock speeds a fair bit (from 2.3 ghz to around 2.8, a 20% ish jump in and of itself). However, they've widened the core a lot to give a more AMD, Intel, and Apple like design that it sounds like will bring with it a HUGE performance jump when it comes to IPC: Perhaps over 50%. For reference, the Snapdragon 845 sounds like it will only get around a 30% performance jump compared to the 835, clock speed and IPC jumps all included.

 

This is really exciting, as it sounds like we may finally have a CPU that can compete with Apple in performance and power. If Samsung's claims are true I wouldn't be surprised if the S9 only uses Exynos chips as it sounds like the 9810 will hold a really big performance advantage over the Snapdragon 845. This might put more pressure on other CPU designers like Qualcomm to create wider CPU cores and could really push the market forward. I look forward to the next few years when it comes to mobile CPU performance and it sounds like we could see some real performance jumps now.

I want to see EverythingApplePro`s reaction to this kind of performance uplift, since we are talking twice the performance of the 8895/ Snapdragon 835, the 9810 will destroy the 845, the 8895 and the A11 with ease.

[Trixanity]

1 hour ago, Sniperfox47 said:

And look at the GPU performance, the X1 still beats the Snapdragon 845 for pure GPU performance, and the 845 just came out 2 and a half years later. If you factory downclock and undervolt the GPU, it would totally be feasible to implement in a reasonable TDP for a high performance phablet (though likely not a small-factor phone).

Source? From the data I've looked at X1 beats the 820 slightly but has a higher TDP making it pretty much worthless.

[Sniperfox47]

42 minutes ago, Trixanity said:

Source? From the data I've looked at X1 beats the 820 slightly but has a higher TDP making it pretty much worthless.

https://www.notebookcheck.net/Nvidia-Shield-Android-TV-Review.166556.0.html

 

Go the benchmarks section, select 3dmark and look at the graphics score. The X1 in the Shield TV got 7734 points vs the 4581 of the Snapdragon 820 for Slingshot.

 

Or a score of 57922 for the X1 vs 30061 for the 820 in Ice storm unlimited, which I think is a dumb test but is useful for comparison since according to the link here, the Snapdragon 835 only gets 57000-57415 ( https://www.cnet.com/news/samsung-galaxy-s8-qualcomm-snapdragon-835-speed/ ) and indications are the new Adreno 630 in the SD845 isn't much faster if at all in real world situations than the SD835's Adreno 540.

 

If you look at the physics score, or any other CPU limited benchmark it's basically the same performance as the Snapdragon 820, but that's because CPU wise it's pretty boring. 4x A57 Big cores and 4x A53 little cores like every other flagship SoC launched that year other than the SD820.

 

 

 

But this is getting super off topic so I'll drop it here.

[DocSwag]

6 hours ago, Sniperfox47 said:

Denver2 (from Parker/the X2) was significantly better for that the the original Denver, and Carmel (from Xavier) is supposed to be pretty much identical to native performance according to what little I've seen of it. Do workloads without much repeating code still perform worse on Denver2? Yeah, but it's nowhere near as bad as it was with classic Denver on the Tegra K1. Not to mention you typically see Denver2 paired with ".Big" A57 cores which are perfectly suited for the kinds of code where Denver2 struggles.

Again, there's no reason they couldn't be. The only real con-side to them is their larger die size and more expensive price, which are the exact same issues Apple's SoCs have now. I mean if you're going to use "Well they're not used in that market!" as a way to say they're not comparable to other products, then we probably shouldn't be comparing *any* other chipsets to Apple's either, since they're never going to be put in a situation where they directly compete.

Is it really that power efficient though? IIRC The K1 was only used in tablets and stuff where there's more thermal dissipation so I'm not sure if it's really good for use in smartphones.

4 hours ago, DanielMDA said:

I want to see EverythingApplePro`s reaction to this kind of performance uplift, since we are talking twice the performance of the 8895/ Snapdragon 835, the 9810 will destroy the 845, the 8895 and the A11 with ease.

It'll beat the 8895 and maybe the 845 by a fair bit, but I personally doubt it'll beat the A11.

[79wjd]

27 minutes ago, DocSwag said:

It'll beat the 8895 and maybe the 845 by a fair bit, but I personally doubt it'll beat the A11.

Well, double the performance will put the single core performance on par and multi core performance about 20% ahead. However, it was "up to" double the performance, so only time will tell.