Samsung outs Exynos 9 Series 9810

[Lodix]

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.

It says "around" not "up to". In mobile chips in general ( Huawei, Apple, Qualcomm, Samsung ) they are usually very honest when giving numbers of the performance increases.

 

[dawheat]

It says "around" not "up to". In mobile chips in general ( Huawei, Apple, Qualcomm, Samsung ) they are usually very honest when giving numbers of the performance increases.

I think the last couple years they've announced something like 25-30% increases in performance over the last gen in their releases. Either they've gone big like Apple or marketing has taken control, but their history doesn't support a ton of exaggeration in this area beyond the normal best case comparisons.

 

[Lodix]

I think the last couple years they've announced something like 25-30% increases in performance over the last gen in their releases. Either they've gone big like Apple or marketing has taken control, but their history doesn't support a ton of exaggeration in this area beyond the normal best case comparisons.

Last year with the Exynos 8895 they just claimed a 10% increase in CPU and 50-60% in GPU which they delivered. Of course they have gone wider they say that on the press release and one leaker said it is now a 6 issue cpu.

 

[StinkyPinky]

So why don't these chips make it to the North American market? Samsung is a huge company I'm sure they can roll them out to the models they sell in North America. Is it something to do with the modem?

 

[ZGR]

So why don't these chips make it to the North American market? Samsung is a huge company I'm sure they can roll them out to the models they sell in North America. Is it something to do with the modem?

I call it the Qualcomm monopoly. I had to go out of my way to get the international version of the Note. The Exynos chip will support GSM just fine, and I get 4G LTE with ATT, T-Mobile, etc. I am making sure my next Samsung phone will also be international. Their chips are fantastic. Qualcomm has a CDMA and a GSM variant of their modems, so it kinda makes it tough for Exynos to compete here I guess. I don't know the finer details, but I hate the way the market is divided up.

 

[CluelessOne]

Big question now: is it Meltdown or Spectre proof?

 

[krumme]

So why don't these chips make it to the North American market? Samsung is a huge company I'm sure they can roll them out to the models they sell in North America. Is it something to do with the modem?

I understand but seriously who can tell a a75 vs this fat monkey in a phone outside of benchmarks? Its 100% impossible. 100%.
If there is a difference its in battery life but as its the a55 cluster running 99% of the time anyway in both variants its probably about the same also.

 

[Hans de Vries]

Over 2 days have passed since the announcement and they still haven't retracted the "around two-fold" improvement of single core performance as a "typo" or so......

http://www.samsung.com/semiconductor/minisite/exynos/products/mobileprocessor/exynos-9-series-9810/

Now... That would be really nice after the apparent long delays in the ARM "Ares" and "Prometheus" core designs.
The TDP may hold it back a bit at 10nm LPP in the phone form factor, to improve a lot at their 8nm and 7nm process nodes.

 

[krumme]

Over 2 days have passed since the announcement and they still haven't retracted the "around two-fold" improvement of single core performance as a "typo" or so......

http://www.samsung.com/semiconductor/minisite/exynos/products/mobileprocessor/exynos-9-series-9810/

Now... That would be really nice after the apparent long delays in the ARM "Ares" and "Prometheus" core designs.
The TDP may hold it back a bit at 10nm LPP in the phone form factor, to improve a lot at their 8nm and 7nm process nodes.

Hans what kind of team size is needed to fork out such wide designs? Is this different from eg ryzen complexity?

 

[soresu]

Hans, isnt Prometheus the codename for A75?
If so, I'd assume that Ares will be the next high end core announced in May/June, unless they keep it back to affect a revision to counter the recently announced vulnerabilities like Spectre.

 

[Nothingness]

Hans, isnt Prometheus the codename for A75?

Indeed it is: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.100403_0200_00_en/lau1465905803414.html

 

[IntelUser2000]

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

2x faster @ 2.9GHz and 40% faster for multicore against 2.34GHz predecessor suggests:
-60% faster per clock
-2GHz multi core clock

I think ARM vendors including Apple are getting nice benefits from the "big.little approach". The idea starts to make sense when the advancements start being process limited. Before, you'd increase execution units to improve per core performance with little care for anything else - new process will take care of the transistor count and power consumption increase.

But its been few years where process improvements have slowed. And the CPUs are thermally limited. Hence all architects go with the 1% increase in power for 1% increase in performance approach, for example.

Traditionally, its said that per core performance increases at square root of core size(thus power use). 4x core size would end up in 2x the performance. If you are then process, thus power limited, that works against you. Rather than a 4x sized core being 2x faster, it needs to throttle and may not end up being faster. Also it means eventually your "little" cores aren't so little in terms of performance compared to your "big" cores.

With approaches like what ARM calls it DynamIQ(enhanced big.little basically), you can do fancy things with it. The "little" cores can be used to speed up multi-thread execution, as multiple cores scale near linearly in many applications. The "big" cores can be way bigger than before for Single thread, since smaller, and more power efficient cores can be used for everything else.

ARM's DynamIQ:
https://www.anandtech.com/show/11441/dynamiq-and-arms-new-cpus-cortex-a75-a55/2

Unfortunately for Intel and AMD they seem to be beset with serious execution issues and their implementation of such an idea is delayed. For Intel, their CTO Justin Rattner in 2008 talked about "Platform 2015", a vision that is being fulfilled today.

Around 2015-2016 we would have seen 10nm Intel architectures, had they not had issues with their process. I believe Intel's version, code-named Lake Field, will be achieved early as Ice Lake generation.

 

[Lodix]

I hope they give more details about the SOC at CES.

 

[Thala]

Before, you'd increase execution units to improve per core performance with little care for anything else - new process will take care of the transistor count and power consumption increase.

Leakage has been going up the last few shrinks, so the idea that process will take care of power is inherently flawed. For larger cores clock gating non utilized cells is not the solution anymore due to the leakage problem. In conclusion with smaller nodes the idea of big.LITTLE gains effectiveness - it is not avoidable.

 

[IntelUser2000]

Leakage has been going up the last few shrinks, so the idea that process will take care of power is inherently flawed. For larger cores clock gating non utilized cells is not the solution anymore due to the leakage problem. In conclusion with smaller nodes the idea of big.LITTLE gains effectiveness - it is not avoidable.

At one point, it wasn't big enough. In the 65nm generation and onwards, that's when it started skyrocketing.

In 1995 there were CPUs that barely had a fan on them. Desktop CPUs used power in the range of 10W. The big gain from 1995 to 2003 had to do with TDP increase, and massive process improvements. When going to a new process meant immediate 50% power reduction, or 50% clock speed improvements, you could do so much with the new design. The original Pentium for example, tripled transistor count over its predecessors.

Now, its rare to see 25% power reduction on a new process, and clock speed increases are eeked out. Voltage scaling is dead, frequency scaling is dead, thermals have hit a wall, and process scaling is on its deathbed.

 

[IntelUser2000]

Preliminary 9810 Geekbench results: https://browser.geekbench.com/v4/cpu/4866523
8810 result: https://browser.geekbench.com/v4/cpu/6007460

It's "only" 30% faster for cumulative ST score and 15% faster for MT score, but,

52% faster in ST Integer, 26% faster in MT Integer
51% faster in ST FP, 13% faster in MT FP

For the cumulative score the low memory score is holding the 9810 down. It doesn't matter as Integer and FP results are quite fast. It's possible it could get higher if the memory performance is on par/exceeds the 8810.

 

[Lodix]

Preliminary 9810 Geekbench results: https://browser.geekbench.com/v4/cpu/4866523
8810 result: https://browser.geekbench.com/v4/cpu/6007460

It's "only" 30% faster for cumulative ST score and 15% faster for MT score, but,

52% faster in ST Integer, 26% faster in MT Integer
51% faster in ST FP, 13% faster in MT FP

For the cumulative score the low memory score is holding the 9810 down. It doesn't matter as Integer and FP results are quite fast. It's possible it could get higher if the memory performance is on par/exceeds the 8810.

This score is fake.

 

[dark zero]

This score is fake.

Actually seems to be on par... Remember that is a preeliminary product. There are still things to fix.

Also I saw a rumor that ALL their cores of the Exynos are custom, even the low power ones.

If that so, I expect seeing a ST of 2800 and MT of 8.5K or even 9K but the problem of the chip will be the Graphics.

 

[Nothingness]

That Geekbench score looks fake:

Identifier ARM implementer 65 variant 1 part 3335 revision 0

That's the CPUID of a Cortex-A57: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0488c/CIHFAACD.html

 

[Lodix]

That Geekbench score looks fake:

That's the CPUID of a Cortex-A57: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0488c/CIHFAACD.html

It is the Exynos 7420, Andrei confirmed it.

Some people are claiming the modem inside this SOC supports CDMA... I don't see where they take this from.

 

[Andrei.]

It is the Exynos 7420, Andrei confirmed it.

Some people are claiming the modem inside this SOC supports CDMA... I don't see where they take this from.

The S359 modem supports CDMA and is less feature rich than the 9810 modem... we have no confirmation the latter though.

 

[Lodix]

The S359 modem supports CDMA and is less feature rich than the 9810 modem... we have no confirmation the latter though.

I see, thanks. Is this the modem inside the S8 family?

I hope they drope Qualcomm now that they are switching to TSMC at 7FF.

 

[Andrei.]

The S359 modem supports CDMA and is less feature rich than the 9810 modem... we have no confirmation the latter though.

It is the Exynos 7420, Andrei confirmed it.

Some people are claiming the modem inside this SOC supports CDMA... I don't see where they take this from.

First CDMA Exynos phone from Meizu: https://www.anandtech.com/show/12331/meizu-announces-m6s-with-exynos-7872

 

[french toast]

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

Wow, Samsung is getting serious, I also did not know M1& M2 are in order cores...but 6 wide at 2.9ghz...
Power consumption is going to be fruity.

 

[AkulaMD]

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

Wow, Samsung is getting serious, I also did not know M1& M2 are in order cores...but 6 wide at 2.9ghz...
Power consumption is going to be fruity.

Really nice sharing. I always thought that M1 and M2 are out of order.