GPU Performance Benchmarks
We will begin with results from GFXBench, and the venerable T-Rex benchmark. This is an older test which uses OpenGL ES 2.0, and results from the 1080p offscreen test are charted below:
Here the Galaxy S9+ is neck-and-neck with the iPhone X. But what about an OpenGL ES 3.0 test? For that we move up to the Manhattan 1080p offscreen test:
Here we see the power of the Adreno 630, which should provide better real-world performance than any other Android phone, and in this case better than the GPU cores from Apple's A11 Bionic as well. Another ES 3.0 test follows:
ALU 2 is too close to call, effectively a tie at 1080p. For the next result we will look at just the GPU portion of the Basemark OS II 2.0 test:
Apple has an advantage here with this older test, which is somewhat surprising given the strong overall showing of the Galaxy S9+ in Basemark OS II 2.0 – though one synthetic benchmark is not enough to paint the full picture.
We now move on to 3DMark, starting with the unlimited version of the Sling Shot test. For this we cannot have a true comparison as Apple does not use OpenGL ES 3.0 for this test as the Android version does, opting for their proprietary Metal API instead.
A sweep for the Adreno 630 in the GS9+, with a better overall score thanks to higher graphics and physics sub-scores.
Finally, we look at results from the unlimited version of Sling Shot Extreme, with Apple's Metal API up against OpenGL ES 3.1 with the Android phones:
Another win for the Adreno 630, which is quite a bit more powerful than Apple's A11 Bionic in these more demanding graphics benchmarks.
On the next page we'll take a look at general usage and battery life, and then cover some final thoughts.
Why the Kryo 385 cores when
Why the Kryo 385 cores when Samsung has the M3 Mongoose-3 cores(1) that are just as powerful as Apple’s A series core designs.
Really is Samsung Being limited because of the other radio IP that it has to license from Qualcomm but that Samsung M3 core is much wider than any Arm Holdings cores even the ones that are semi-customizied by Qualcomm.
Why does the US market not get the Really Wide Samsung Cores.
(1)
“Mongoose 3 (M3) – Microarchitectures – Samsung”
https://en.wikichip.org/wiki/samsung/microarchitectures/mongoose_3
because the Exynos variant
because the Exynos variant sucks balls and is overhyped. Check out the Anandtech review where they do a deep dive on both variants. The Exynos variant is almost as bad as last years SOCs
Maybe that’s more to do with
Maybe that’s more to do with the Apple Bionic’s other processor IP compared to the M3 in that Samsung Phone SKU that makes use of the M3 mongoose cores in that Exynos 9810. And Qualcomm has It’s DSP IP that’s available through its device API for apps to make use of. The GPU IP on the Samsung phone may not be there but the Single core benchmarks on the M3 based Exynos 9810 are not that bad.
I wish that Anandtech would take the Time do a REAL deep dive Apple’s later cores(Not Phone SKUs) after the A7/Cyclone and even wikichip lacks more complete Apple A series cores beyond The Apple A7/Cyclone series cores that Anand Lal Shimpi did when he did that last detailed Deep Dive for Anandtech before he left the publication.
Where are the Wikichip entries for all of Apples A series cores i’m not finding the and It’s like Apple has gone all double top secret with it’s CPU cores specifications.
The M3 based Mongoose M3 cores in the Exynos 9810 are just a little behind the Apple A10 cores in single threaded performance and that’s not the fault of the M3’s cores as most of those Phone SKU bemchmarks are testing GPU/other IP and I’m only talking about that M3 core wikichip info as it relates to a wide order superscalar design that’s has 6 wide instruction decoders just like Apple’s designs since the A7 and the M3′ execution engine if fatter than Apple A7 designs.
There is not enough information on Apple’s later A series CPU core designs as Apple keeps that under tight wraps and Apple did hire Anand Lal Shimpi to keep him quiet concerning any of Apple’s later core designs. The Apple A7 cores hardware features where very well sussed out by Anand using the available software testing tools in addition to some tools Anand wrote himself.
That Wikichip on the M3 is a real Deep dive and even Wikichip’s Apple series information is lacking and your definition of deep dive differs because any deep dive for the Apple A series cores after the Apple A7 Cyclone cores is still lacking to this day. CPU core Deep dives go into great detail into the decoders, Execution engine, and INT, FP, branch units, load store units, reorder buffer, and other units that make up the CPU’s core execution resources and Deep Dives about Phone SKUs are not Deep Dives about a CPU’s cores. That M3 Mongoose custom core has a 12 micro-op instruction issue amd that’s damn wide.
I’m not interested in any Phone SKU I’m interseted in the CPU core’s detailed breakdown and that M3 core looks more like a desktop CPU core no matter the Phone SKU it gets crammed into along with all that other GPU IP, DSP IP and others AI IP that Apple’s A11 gets use of. Apple’s A11 has a dedicated AI processor also but that’s a defferent discussion, I talking about only CPU cores in my posts and not that other IP.
Edit: if fatter than
to: is
Edit: if fatter than
to: is fatter than
Exynos garbage plus they run
Exynos garbage plus they run hot and thermal throttle unlike snapdragon that can sustain performance for prolong periods.
“Exynos garbage”, which
“Exynos garbage”, which version and in what form factor. The M3/mongoose core is not bad and I’d like to see that in some talet form factors and maybe even laptops.
And snapdragon has had its share of thermal issues also. You still appear to thinking about phones whereas I’m more interested in the M3(Mongoose) core in Tablets and Laptops/cromebooks etc.
I really do not care about Phones as I want to see wider custom ARM designs like the M3/Mongoose that can easily be used on laptops. Apple and Qualcomm have other IP on those phones that offload some tasks from the CPU’s cores to the DSP/AI/GPU cores so I’ll still not Judge the M3 Mongoose cores cramed inside a phone form factor. Apple and Qualcomm mostly have their excess IP(DSPs, AI, other specilized procesors) on their SOCs/modules that is not related to any CPU cores and that’s why Apple and Gualcomm can use less power.
You still want to talk about mostly useless smart phones, and I’ll talk wider order superscalar Arm Cores like Apple’s A series cores(A7 later) and that Samsung M3/mongoose cores in laptops and tablets instead of only phones.
I’m wanting those Samsung M3 mongoose cores to start showing up in Tablets and laptops/chromebooks that have a bit more thermal headroom. Feature Flip-phones are good enough for me but that Samsung M3/mongoose in a tablet/laptop form factor with more thermal headroom is going to be better than that Qualcomm/Windows 10 SOC SKU and those narrow semi-costom ARM cores that are still basically Arm Refrence cores more than they compare to Apple A7-A11 cores or Samsung’s M3/mongoose cores.
Exynos(the Exynos 9810) is just one SKU that makes use of the Samsung M3 mongoose cores, the first SOC SKU to do so. There will be other SOCs that make use of the M3/Mongoose cores so I’m not about to judge the M3 in only one tapeout for a phone SKU. Tablets and Laptops are making use of much weaker cores than Samsung’s M3/Mongoose and looking at the single core IPC mertics on the M3/Mongoose core in anandtech tells me that it’s very close to Apple’s A10 and not too far behing the Apple A11. Now all Samsung has to do with the M3/Mongoose is maybe add SMT capabilities and that would be the next step towards a nice ARM based laptop SKU from Samsung if they make use of more GPU cores.
That Exynos 9810 lacks the GPU/AI/DSP functionality that Apple and Qualcomm have so that’s not the M3/mongoose core fault and Samsung needs to double down on pairing the M3/Mongoose cores with some better GPU IP and AI/DSP processor IP ato better compete with Apple and Qualcomm.
Anandtech has been doing its
Anandtech has been doing its best to dispel certain misunderstandings about mobile SoCs. One of the big takeaways from their more recent chip analyses is high peak performance for certain SoCs is not a reliable guide to sustained performance. The SD845, for example, appears to outperform both the Apple A11 and Exynos 9810 when we look beyond peak performance to sustained performance. Mobile chips work with a ‘governor’ engaged much of the time. That is necessary to stay within thermal limits that must be observed (if users can reasonably expect to hold those devices without discomfort). And, in fact, the ‘lower performing’ SD845 strikes a better design balance than ‘higher performing’ A11 and Exynos 9810 parts that delivers higher sustained performance without exceeding thermal limits. The governor on the SD845 has a lighter touch, so to speak, than the the more brutal governors of the A11 and Exynos 9810 chips.
The SD845 also is the better chip in terms of energy efficiency and the balanced design has a lot to do with that. Still, for certain classes of devices, that can tolerate the SoCs running at a level closer to their peak performance capability (without being held back by an invasive governor) the A11 and Exynos 9810 would be better choices. In a direct comparison the A11 seems to have a small peak performance edge over the Exynos 9810. It probably also manages bursts of performance better. That can already be attested by the way the A11 frugally manages battery power in the smartphone context compared to the Exynos chip. It is unclear whether this indicates a definite win of the A11 over the Exynos part in terms of energy efficiency, though. Anandtech haven’t explored that question yet, but they have noted that the Exynos chip seems to be weighed down by crummy software controlling the operation of the SoC. So, further optimization might yield better SoC performance or a better performance/energy efficiency balance. Only Anandtech seem to be looking into these matters. Other sites are reporting benchmark numbers that are largely meaningless.
Edit: if fatter than
to: is
Edit: if fatter than
to: is fatter than