MediaTek Dimensity 820 unveiled: higher CPU clock speeds, extra GPU core and dual SIM 5G
- s
- scoelho
- JsT
- 12 Jul 2021
Anonymous, 18 May 2020Your knowledge on ARM CPUs is rather shallow. Those Kryo co... moreYes I agree with you, idk if AMD together with Samsung can change a little about that.
- r
- r31ya
- sEG
- 23 Oct 2020
BNet07, 28 May 2020Can anyone please tell me about heatting issue of mediatek ... moreFor more recent series like G90, the heat is much more managable.
With heat during high work load usually at 42~44*c which recent SD732 also reach similar heat level.
It's kinda overblown in youtube reviews tough and annoyingly some youtuber did somewhat false comparison with checking the heat level directly above chipset for G90 and then compared with some snapdragon by checking the heat level at the center of the phone.
- ?
- Anonymous
- Hku
- 16 Jun 2020
BNet07, 28 May 2020Can anyone please tell me about heatting issue of mediatek ... moreIts 7 nm so thermal shouldn't be a problem.
- B
- BNet07
- 6p}
- 28 May 2020
Can anyone please tell me about heatting issue of mediatek processor?
Is this Dimensity 820 out of this claim?
- Dolanescu
- G6p
- 20 May 2020
Aryan, 19 May 2020Dimensity 820 comes packing a powerful punch. It definitely... moreCheck antutu cpiset benchmarks... is on par with SD 8235...
- c
- callmeakshat
- vGe
- 19 May 2020
Ankit Gupta, 19 May 2020Of late, I have noticed that Mediatek is really paying atte... moreYes they need to keep upgrading if they need to give a head on competition to Snapdragon 5G chipsets. Atleast in the 5G space, mediatek doesn't want to leave any stones unturned
- A
- Ankit Gupta
- rJs
- 19 May 2020
Of late, I have noticed that Mediatek is really paying attention towards its GPU and this is good enough for the future chipsets.
- Kangal
- X{T
- 19 May 2020
Anonymous, 19 May 2020I can't believe an out-of-order design can be more energy e... moreYeah, like I said, it's crazy good. It's definitely the best architecture available at the moment.
Another great point, are that Apple uses a Semi-Native coding language in the operating system: Objective-C and Swift. Not to mention they also have Metal, which was in-design before the open-standard Vulkan (though the proprietary Mantle was first) and released around the same time.
"is Apple's soc are specifically designed for benchmarks?"
Not really.
Apple cares not for benchmarks. When they're losing, so be it. When they're winning, not too bad. Their marque is the brand/marketing, that means the cohesive experience, recognisable design, and status symbol... this goes back to the iPod, iMac, Lisa, and even the Apple II. They definitely shy away from specifications, on their website and even during unveiling. Improvements are usually vague, but never not-accurate.
The reason why Apple chips, starting with the Apple A7, usually blast through benchmarks is because generally they're great hardware. The GPUs pretty beefy, as is the CPU, and the other co-processors. I already pointed towards how they benefit from vertical integration. They might be shy on RAM, but (even better) the chipset has lots and lots of cache... and that's just expensive. Qualcomm would struggle to do the same thing, as it has to make a profit too. Only Samsung and Huawei have the luxury of experimenting with cache and large chips on their flagships, yet even they've been rather tame until now. Apple sells the iPhone at a high cost, so they can recoup the cost of the chipset on each sale (plus a hefty profit margin).
- A
- Aryan
- XS%
- 19 May 2020
Dimensity 820 comes packing a powerful punch. It definitely has a lot of potential.
- A
- Arjunagarwal
- U@W
- 19 May 2020
Dual sim 5G is an incredible addition here. 5G is all set to deliver the best.
- ?
- Anonymous
- 08B
- 19 May 2020
Kangal, 18 May 2020Thanks for the summary, this is helpful to those that aren'... moreI can't believe an out-of-order design can be more energy efficient compare to in-order...
Pipelines are far more complexe and it leads to multiply transistors by 4 to give the same perfs. And when you multiply transistors, you multiply energy consumption.
Of course, out-of-order had flexibility in instructions queue, but this flexibilty is useless when you have out-of-order big cores in addition.
Plus I'm always surprise when I see a device with crazy Geekbench scores and notice very tiny differences in real life use (exemple : Redmi Note 8 (2500pts single core in Geekbench 4) Pro vs iPhone XR (4800pts single core in Geekbench 4). XR isn't impressive compare to 8 Pro...speed is very close...absolutely not the double!
As you can't do task monitoring on iPhone, I ask me if Apple's soc are specifically designed for benchmarks
- M
- Mediatek sux
- ypc
- 19 May 2020
Kangal, 18 May 2020Thanks for the summary, this is helpful to those that aren'... more"against a Cortex-A55 implementation such as on the Snapdragon 855, the new Thunder cores represent a 2.5-3x performance lead while at the same time using less than half the energy"
Going by geekbench5 scores of a55, thats better performance than cortex a75 on the sd845.
- ?
- Anonymous
- fI{
- 19 May 2020
Kangal, 18 May 2020Thanks for the summary, this is helpful to those that aren'... moreThe worst 64-bit chips are the first gen 64-bit Snapdragon chipsets. Other than that, there's nothing bad about 64 bit ARM Cortex cores. Cortex-A57 is still relevant to this day, even the Nintendo Switch uses it.
- L
- Lele
- tDS
- 19 May 2020
How about Battery consumption??
- Kangal
- X{T
- 18 May 2020
Anonymous, 18 May 2020Last real Kryo was used on Snapdragon 820...all the others ... moreThanks for the summary, this is helpful to those that aren't knowledgeable in this area.
The A76 is the top recommendation, highly recommend it on the 8nm node.
The A73 has aged very well, highly recommend it on the 16nm node.
....The A73 is not as bad as the custom Mongoose M2, the custom Mongoose M1, the custom Kryo-100, the Standard A72, or even the (much) worse the A57. They kinda sucked even on good nodes. The A75 and A77 architectures are a bit of a mixed bag.
I am hoping the A78 bumps things up properly and significantly. Allows 8 instructions per cycle, much larger cache, and lower latency and higher bandwidth with proper DDR5 memory. It really needs to catch up to Apple's Architecture prowess in the Performance cores, ever since they were introduced in the 2018 Apple A11-Bionic (and have been iteratively improved on since).
Though I have to say, the most advanced architecture is actually the new Efficiency cores (Thunder CPU) in the A13-Bionic. It's an out-of-order processor that is slightly faster than the Cortex A73, and yet it uses slightly less power than the Cortex A53. Truly unbelievable. I'm hoping the A14-Bionic might utilise a 3+5 design. I do wonder why Apple has not developed their server with it, like Amazon has done with the AWS Graviton1 (A72 CPU) and the latest Graviton2 (A76 CPU). The Graviton2 is already (slightly) ahead of AMD's Latest Ryzen-based servers, and much ahead of Intel's Aging Core-based servers... Apple would dominate if they made a server based on the Thunder CPU architecture!
- A
- AhmedSLL
- rme
- 18 May 2020
Can anyone explain why Mediatek keeps overclocking they're efficiency cores? And how does it affect the performance?
- Shellyman Infinity
- B1E
- 18 May 2020
zaryan, 18 May 2020looks like mediatek started the same battle with Qualcomm a... moreYeah, this is getting interesting.
- ?
- Anonymous
- 08B
- 18 May 2020
My bad...
Some corrections :
Kryo 385 is Cortex 4A75+4A55 with 4x16bits lpddr4x memory support
- ?
- Anonymous
- 08B
- 18 May 2020
digitaldevil, 18 May 2020sd got Kryo not cortex , cortex is the trash cpu which s... moreLast real Kryo was used on Snapdragon 820...all the others are more rebranding
Kryo 250 Gold+Silver (SD632) is Cortex 4A73+4A53 with lpddr3 memory support
Kryo 260 is Cortex 4A73+4A53 with 2x16bits lpddr4x memory support
Kryo 280 is Cortex 4A73+4A53 with 4x16bits lpddr4x memory support
Kryo 360 is Cortex 2A75+6A55 with 2x16bits lpddr4x memory support
Kryo 385 is Cortex 4A75+4A55 with 2x16bits lpddr4x memory support
Kryo 460 is Cortex 2A76+6A55 with 2x16bits lpddr4x memory support
Kryo 485 is Cortex 4A76+4A55 with 4x16bits lpddr4x memory support
Kryo 585 is Cortex 4A77+4A55 with lpdr4x and lpddr5 memory support
Kryo 2xx Gold and Cortex A73 can process 2 instructions per cycle
Kryo 3xx Gold and Cortex A75 can process 3 instructions per cycle
Kryo 4xx Gold and Cortex A76 can process 4 instructions per cycle
Kryo 585 Gold and Cortex A77 can process 6 instructions per cycle
On Dimensity 820, a Cortex A76 @2.6ghz can process up to 2.6 x 4 = 10.4 x 10^9 instructions per second
On Snapdragon 845, a Kryo 385 Gold @2.8ghz can process up to 2.8 x 3 = 8.4 x 10^9 instructions per second
On Dimensity 1000, a Cortex A77 @2.6ghz can process up to 2.6 x 6 = 15.6 x 10^9 instructions per second
Any other question?