Samsung outs Exynos 9 Series 9810

[Dayman1225]

With the benefits of the industry’s most advanced 10nm process technology, the Exynos 9810 will enable seamless multi-tasking with faster loading and transition times between the latest mobile apps. The processor has a brand new eight-core CPU under its hood, four of which are powerful third-generation custom cores that can reach 2.9 gigahertz (GHz), with the other four optimized for efficiency. With an architecture that widens the pipeline and improves cache memory, single-core performance is enhanced two-fold and multi-core performance is increased by around 40 percent compared to its predecessor.

Source

 

[FIVR]

They're gonna need better than 100% to beat A11 Bionic in Single thread performance or worse yet, A11X

 

[raghu78]

There is no way single core performance is two fold of the predecessor. I would be impressed if they cross 3000 in geekbench 4.

 

[Dayman1225]

Anandtech article on it

https://www.anandtech.com/show/12212/samsung-announces-new-exynos-9810-soc

 

[HurleyBird]

I wouldn't be surprised if they boosted single core performance close to 2x in geekbench. I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance. They'd almost be insane not to when this single metric has somehow become the defacto benchmark of processor performance. Imagine what would inevitably happen if the GPU-buying market assessed the speed of GPUs almost entirely based on their 3DMark scores. GPUs would get very good at 3DMark in very short order.

We desperately need a wider variety of benchmarks for mobile that reliably isolate processor performance.

 

[Nothingness]

I wouldn't be surprised if they boosted single core performance close to 2x in geekbench. I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance. They'd almost be insane not to when this single metric has somehow become the defacto benchmark of processor performance. Imagine what would inevitably happen if the GPU-buying market assessed the speed of GPUs almost entirely based on their 3DMark scores. GPUs would get very good at 3DMark in very short order.

We desperately need a wider variety of benchmarks for mobile that reliably isolate processor performance.

High-end ARM CPU designers take care of Geekbench performance but not only. Many other benchmarks are run ranging from micro-benchmarks up to SPECCPU 2017, browsers, Javascript and so on.

Back to the topic, if indeed Samsung achieves 2x speedup I'd be impressed and that'd be quite an achievement!

 

[Lodix]

They're gonna need better than 100% to beat A11 Bionic in Single thread performance or worse yet, A11X

It would be very close to the A11 in raw CPU performance ( and better in real world multitasking performance because it has 4 big cores instead of 2 ) if this is true. And why do you compare it to a tablet SOC with higher TDP ?

I wouldn't be surprised if they boosted single core performance close to 2x in geekbench. I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance. They'd almost be insane not to when this single metric has somehow become the defacto benchmark of processor performance. Imagine what would inevitably happen if the GPU-buying market assessed the speed of GPUs almost entirely based on their 3DMark scores. GPUs would get very good at 3DMark in very short order.

We desperately need a wider variety of benchmarks for mobile that reliably isolate processor performance.

Taking in mind that their subtest scale similarly to SPEC benchmarks I would not mind this in absolutely. But no, they don't do this.

 

[ZGR]

I wouldn't be surprised if they boosted single core performance close to 2x in geekbench. I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance. They'd almost be insane not to when this single metric has somehow become the defacto benchmark of processor performance. Imagine what would inevitably happen if the GPU-buying market assessed the speed of GPUs almost entirely based on their 3DMark scores. GPUs would get very good at 3DMark in very short order.

We desperately need a wider variety of benchmarks for mobile that reliably isolate processor performance.

Oh crap, Geekbench is now the tech equivalent of the Nurburgring....

 

[Yakk]

7/8th marketing, 1/8th substance.

Seems about right.

Although Samsung really seems to be on the ball ARM processor wise considering they also make most of everything also.

 

[Thala]

There is no way single core performance is two fold of the predecessor. I would be impressed if they cross 3000 in geekbench 4.

Indeed. A 3000 score in Geekbench would be surprising considering that A75 should be "only" around 2600. In the past Exynos SoCs would either use ARM cores directly (A57) or their custom cores were very close to ARMs latest and greatest.

On the other hand it puzzles me, why the multicore score only increases by 40%? Assuming the same number of cores the multi-core score should faithfully follow the single core score.

 

[Thala]

I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance.

Name only one reasonable micro-architectural improvement, which would only affect say the Geekbench integer score while degrading other applications. Thats close to impossible and much less reasonable.

However with compiler optimizations you can target very specific code sequences.

So the biggest issue with benchmarks are not micro-architectural optimizations but compiler optimizations. Intel for example has been for years tuning their compilers for SPEC.

 

[HurleyBird]

Name only one reasonable micro-architectural improvement, which would only affect say the Geekbench integer score while degrading other applications. Thats close to impossible and much less reasonable.

Degradation would be a bit of a reach. Optimising towards a specific (but generally well rounded) benchmark could only reasonably be expected to exaggerate true performance.

 

[krumme]

4 wide ?

 

[Lodix]

Indeed. A 3000 score in Geekbench would be surprising considering that A75 should be "only" around 2600. In the past Exynos SoCs would either use ARM cores directly (A57) or their custom cores were very close to ARMs latest and greatest.

On the other hand it puzzles me, why the multicore score only increases by 40%? Assuming the same number of cores the multi-core score should faithfully follow the single core score.

We don't know yet the true potential of the Cortex A75 it should score more than that if ARM were correct when they presented it.

Samsumg custom cores have been on par or slightly ahead of IPC compared to the ARM stock at the same time.

The increase of "only" 40% in multi core must be because the 2'9GHz are only for 1-2 cores load and for 3-4 will clocks lower.

4 wide ?

A chinese leaker said it is 6 issue CPU.

 

[dark zero]

If Mongoose is superior than A75 it will mark the Samsung own improvements.... I guess they should talk with Microsoft to make their processors compatible too with Windows.

 

[raghu78]

We don't know yet the true potential of the Cortex A75 it should score more than that if ARM were correct when they presented it.

Samsumg custom cores have been on par or slightly ahead of IPC compared to the ARM stock at the same time.

The increase of "only" 40% in multi core must be because the 2'9GHz are only for 1-2 cores load and for 3-4 will clocks lower.

A chinese leaker said it is 6 issue CPU.

https://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

Yeah it looks like Samsung really went for a massively wide Apple like 6 micro-op issue core. I think the 1 core turbo is 2.9 Ghz and for 2-4 cores the clocks will probably be around 1.8 - 2.2 Ghz. I think for the first time in a long while we will see Samsung compete aggressively with Apple in single core performance. Multi core as usual will be very competitive.

 

[Andrei.]

If Mongoose is superior than A75 it will mark the Samsung own improvements....

Well it better damn be after 3 generations otherwise the whole custom core thing is a massive waste of money. SLSI is clearly targeting Apple here, not trying to be equivalent to ARM.

People also need to stop fussing about GeekBench. It's the best public cross platform benchmark out there. I'll have an article soon with SPEC2006 figures across last few gens of SoCs and to be fair the relative % positioning between SoCs and their GB4 scores doesn't change much at all.

If Samsung can reign in the power consumption of their un-core memory subsystem I have no doubts that they can reach those claims. If the 2.9GHz clocks are below 3W then that's absolutely fine. And yes obviously the multi-core perf is lower as they won't be running all 4 cores at 2.9GHz.

 

[Lodix]

https://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

Yeah it looks like Samsung really went for a massively wide Apple like 6 micro-op issue core. I think the 1 core turbo is 2.9 Ghz and for 2-4 cores the clocks will probably be around 1.8 - 2.2 Ghz. I think for the first time in a long while we will see Samsung compete aggressively with Apple in single core performance. Multi core as usual will be very competitive.

I bet it is 2'9GHz for 2 cores and 3-4 cores will be 2'3-2'5GHz.

 

[jt7]

I bet it is 2'9GHz for 2 cores and 3-4 cores will be 2'3-2'5GHz.

At 4 x 2.3 Ghz, the big cores should score higher than the estimated ~9000 in GB, even allowing for less than ideal scaling and leaving out the smaller cores all together. I think they might be clocked sub 2 Ghz if all cores are active.

It's too early to say, but It seems strange they went with a 4 + 4 design. If the big cores are so good, couldn't they have got to ~9000 with a 2 + 4, or even a 2 + 2 as Apple has shown?

 

[krumme]

At 4 x 2.3 Ghz, the big cores should score higher than the estimated ~9000 in GB, even allowing for less than ideal scaling and leaving out the smaller cores all together. I think they might be clocked sub 2 Ghz if all cores are active.

It's too early to say, but It seems strange they went with a 4 + 4 design. If the big cores are so good, couldn't they have got to ~9000 with a 2 + 4, or even a 2 + 2 as Apple has shown?

The midrange samsung phones gets 2xa73 plus 4xa55.
I seriously dont know who wants more cpu grunt than that in a phone. And can tell in a blind test.
And now this slim and lean a73 makes much sense in this market of fat highend cores.
We drivel over a doubling of st gb score for the highend but the midrange gets more than that and performance difference that is needed and noticiable. From a53 to a73 !

 

[krumme]

Amd and Intel needs to up their game.
Samsungs big fat monster R&D budget is starting to materialize.
Euv and 6 issue design...

 

[DeletedMember377562]

Am I the only one who sees the clear issue here? Single core performance effects multi core performance as well. If you increase your CPU with great single core performances, it will reflect on the multithreaded performance. Exynos chips have 4 efficiency and 4 performance cores. So supposing SC perf improved 2x on just the 4 perf cores, that should increase multithreaded performance by at least 50% overall (even more so, seeing as the efficency cores, are noticably weaker/smaller). We should also assume their efficiency cores have improved, which clearly puts multithreaded performance even higher. There's of course the possibility that they have decreased the amount of cores, or reduced the size of their efficiency cores compared to the performance ones. Apple manages to have such a fantastic single core performance because of their massive performance cores. Surely, Samsung can't match their performance with 4 performance cores?

In any case, it' highly unlikely SC has improved 2x. Such an increase is unprecedented in the mobile world (the A7, which was a massive innovation, improved perf by 70-80%), and Samsung would market the shit out of it. Not write a single, modest line about it on their site.

If it actually turns out to be true that the Exynos 9810 has improved by 80-100%, i'd hate to be a US costumer with Snapdragon S9's. Because then it would obliterate the Snapdragon 845 in performance by 70-80%.Their current Mali GPU have already surpassed Snapdragon's Adreno solutions in performance.

I wouldn't be surprised if they boosted single core performance close to 2x in geekbench. I also wouldn't be surprised to find out that the various ARM SOC vendors are now optimising their designs around getting high geek bench scores at the expense of actual performance. They'd almost be insane not to when this single metric has somehow become the defacto benchmark of processor performance. Imagine what would inevitably happen if the GPU-buying market assessed the speed of GPUs almost entirely based on their 3DMark scores. GPUs would get very good at 3DMark in very short order.

We desperately need a wider variety of benchmarks for mobile that reliably isolate processor performance.

Lol, what are you smoking? Geekbench is very representative of actual performance. Also, there's various other CPU tests being used on other sites as well. You also discredit 3DMark, even though even that is very representative of actual performance: take a look at how various GPUs rate in 3DMark compared to gaming benchmarks. The differences are similiar.

But in the end it's not about raw performance, but good software. Which is why the Pixel 2, despite having a chip that has half the SC of the A11, provides just as fast of an experience as the iPhone X, and a smoother experience than iOS overall.

 

[Lodix]

Am I the only one who sees the clear issue here? Single core performance effects multi core performance as well. If you increase your CPU with great single core performances, it will reflect on the multithreaded performance. Exynos chips have 4 efficiency and 4 performance cores. So supposing SC perf improved 2x on just the 4 perf cores, that should increase multithreaded performance by at least 50% overall (even more so, seeing as the efficency cores, are noticably weaker/smaller). We should also assume their efficiency cores have improved, which clearly puts multithreaded performance even higher. There's of course the possibility that they have decreased the amount of cores, or reduced the size of their efficiency cores compared to the performance ones. Apple manages to have such a fantastic single core performance because of their massive performance cores. Surely, Samsung can't match their performance with 4 performance cores?

As I have mentioned earlier they have very likely implemented the same solution as they did on the Exynos 8890, when just 1-2 big cores are loaded the frequency can go higher, like intel or AMD do. When 3-4 cores are loaded the frequency for the big cores will be lower, that's why. But Geekbench doesn't scale that well with cores anyway.

 

[french toast]

People expecting multi core scaling to to be the same increase as the single thread performance of one of the big cores is missing a number of things.

1, Not knowing anything about Samsung's custom fabric and memory interfaces means we don't know how much bandwidth is available to those cores, android SOC's with multiple cores have not scaled very well and this is likely not just down to Amdahl's law.

2, As mentioned, those quoted frequency and performance numbers are likely single core turbo..for short bursts, using all cores would drop the frequency down significantly...so why have 8 cores in the first place then?..well perhaps efficiency..4 slower cores is likely more efficient than 2 faster ones...also power of marketing .

3, Even if the big cores miraculously all clock at 2.9ghz (they won't) and Samsung's claims about IPC hold up...the little cores (A55s) have not doubled in performance...(20%?).
Pretty obvious from what little info we have that multi core performance can't double.
The GPU looks a little on the shy side, I'm not sure how area efficient Mali GPU is? ..my assumption is it is not great when compared to adreno and apple solutions, but not much public material?

 

[Mopetar]

I think people are missing the important words in these claims. It’s not the 100% that’s important, it’s the up to that you need to focus on, but naturally everyone immediately goes for the double even though that’s not what Samsung claimed.

It’s the same story with every new chip. Look back at Apple (or NVidia or Intel for that matter) launches where they claim up to 30%/50%/80% improvements. What they mean is that one particular sub-test in Geekbench or some other benchmark gets that huge increase, but general or average performance increase is never that good. If it were that’s what they’d use (assuming some individual aspect wasn’t 200%) instead, but it’s rarely the case.