MediaTek might have pulled an AMD with its new ARM based SoC competing against Qualcomm best offering

[GoodBytes]

MediaTek is a Taiwanese fabless semiconductor company who has for years made entry level and med range SoCs for phone and tablets. They are massive company, just unknown by most, as their work tend to cater the low and med range market, and so has rarely made any news splash. There is a good chance that your budget friendly smartphone and tablet is powered by a MediaTek chip. Their SoC is also used in TVs, Amazon Fire stick, Chromebooks, modems and routers, electronic sport equipment's, smart speakers, and many more.

 

 

 

Today, MediaTek makes a splash with it's new ARM SoC. The company claims that its new Dimensity 9000 flagship chip can compete against Qualcomm best offering

Built-on TSMC’s 4nm process, and using ARM’s new v9 architecture, where a single Cortex-X2 performance core is set to peek at 3.05GHz, in addition has 3x Cortex-A710 cores running at 2.85GHz, and four Cortex-A510 efficiency cores running at 1.8GHz. As for GPU, it runs on it's 10-core Arm Mali-G710. And features 6 cores for "AI processing". 

 

 

It's previous generation flag ship chip the were reported to be close to Snapdragon 888 or Samsung’s Exynos 2100 performance, but the 9000 model is set to be more competitive.

 

The Verge reports:

Quote

Where previous top-tier Dimensity chips (like last year’s Dimensity 1000) were still less powerful than contemporaries like Qualcomm’s Snapdragon 888 or Samsung’s Exynos 2100, the new Dimensity 9000 is coming out of the gate to make a case for 2022 Android flagships.

 

The new Dimensity 9000 is the first mobile chip to be built on TSMC’s 4nm process, in addition to using Arm’s new v9 architecture. It’s also the first announcement CPU to use Arm’s new core designs: a single Cortex-X2 performance core clocked at 3.05GHz, three Cortex-A710 cores at 2.85GHz, and four Cortex-A510 efficiency cores at 1.8GHz.

 

The GPU, meanwhile, is a 10-core Arm Mali-G710, along with MediaTek’s fifth-generation APU with six total cores for AI processing (which the company says offers four times the performance and power efficiency compared to its previous generation).

The new 18-bit Imagiq Gen 7 ISP claims to be the world’s first chip to be capable of capturing a 320-megapixel image (assuming your phone has a sensor that can shoot at that level), capable of transferring data at 9 gigapixels per second.

 

Has a 5G modem (which support 3GPP’s Release 16 specification), and the company claims to be the first SoC on the market that supports Bluetooth 5.3 and Wi-Fi 6E

 

Sources

https://www.theverge.com/2021/11/18/22790189/mediatek-dimensity-9000-flagship-chip-qualcomm-snapdragon-competition-arm

 

 

My thoughts

Of course, we have to see actual benchmarks to confirm, but whether it is true or near true, for a company like MediaTek to start pumping out SoC close to or the same level of performance as Qualcomm best offering, even if efficient isn't a match, shows how much, in my opinion, how Qualcomm has been complacent for years. This is especially true when a company like Apple, which aren't in  the CPU/SoC industry (should not appear as experts in a domain that isn't their focus, I mean) surpassed substantially Qualcomm best offering to a point where it's not even funny.

 

The worst part, is that Microsoft has been working with Qualcomm for years to make Windows on ARM a solid purchase for most consumers, and Qualcomm clearly could not see the potential revenue that it could bring to them. Clearly the company has no vision, and is just enjoying its crown, which seems that it will be detrowned very soon.

 

While we don't know the price of this SoC, there is a good chance, logically speaking, that it will be cheaper than Qualcomm (overpriced, in my opinion) SoC, as all prior SoCs of MediaTek have been much cheaper than Qualcomm's

 

If Qualcomm doesn't wake up, I think we are looking at another RIM/Blackberry.

 

Now, to be fair, Qualcomm has an event on Nov 30th, so we will see what they will bring. But assuming it your typically marginal performance improvement, it shows how MediaTek is very close to Qualcomm.

[Arika]

If true: about damn time.

[Fnige]

Now will phone manufacturers use those chips? I remember the Dimensity 1000 chip which iirc was actually powerful but i dont remember many phones really using them

[Levent]

Quote

MediaTek might have pulled an AMD with its new ARM based SoC

 

Lol more so than you might think. Competition is always good BUT knowing Mediatek and how bad their android sources and documentations are, I personally would always take Qualcomm SoCs over Mediatek any day. It wasnt that long ago they wouldn't even release source code for their hardware. 

[dizmo]

Don't they just use ARM cores? So, wouldn't it be more of Mediatek finally catching up with their chip design, rather than Snapdragon being complacent?

While it might be the same performance wise, what you have to look at when you look at an SoC is everything else; camera capabilities, charging capabilities, AI, etc etc. Kind of like how you can have 2 cars that go 0-60 at the same speed, but one can handle turns and the other can't, making it far more useful. Not to mention, support, I haven't heard much about how long Mediatek supports it's SoCs for. Something that will affect Android versions. The fact it's efficiency isn't the same is also a massive detractor, as having a phone that's fast but dies far more quickly really isn't having something that's on par at all, and is in fact far less useful.

 

I think it's far, far too soon to be making any kind of RIM/Blackberry claims.

[SorryBella]

51 minutes ago, GoodBytes said:

Today, MediaTek makes a splash with it's new ARM SoC

Freedom Phone 2 Confirmed? 

[igormp]

I believe that's the first SoC that's both using ARM's newest Cortex-X2 and TSMC's 4nm, so I wouldn't doubt that would be the fastest consumer chip out there (let's see if it can match Apple's A15).

[lostcattears]

The oldies are losing ground quick they better invent something new and fast your old previous 5-7% performance boost is slacking.

[suicidalfranco]

Would have preferred a new kirin chip, mediatek still sucks for custom roms

[Kisai]

10 hours ago, GoodBytes said:

 

 

While we don't know the price of this SoC, there is a good chance, logically speaking, that it will be cheaper than Qualcomm (overpriced, in my opinion) SoC, as all prior SoCs of MediaTek have been much cheaper than Qualcomm's

 

If Qualcomm doesn't wake up, I think we are looking at another RIM/Blackberry.

 

Now, to be fair, Qualcomm has an event on Nov 30th, so we will see what they will bring. But assuming it your typically marginal performance improvement, it shows how MediaTek is very close to Qualcomm.

 

In my opinion is basically the same as in the Qualcomm thread. Even a high-end ARM SoC will not be competitive with Apple's full vertical integration.  

 

These companies need to either release their own ARM OS, or adopt one not based on Linux or Windows before there will be any meaningful competition in the space. Because relying on Microsoft to release a build for their SoC ensures that the end-user will get locked into a device that becomes worthless when Microsoft inevitably gives up on the ARM version for their SoC. Linux is a bit better and worse in this regard as the SoC support might be in the kernel, but if your SoC goes kaput and you need to switch to another ARM SoC, you're pretty much hosed since you won't be able to just pull the drive out and boot it on another ARM SoC. ARM chips are if anything less backwards-compatible than x86-64 chips, entirely because the instruction set changes from version to version, and there can be cases where one SoC will not have the right configuration of V7/V8/V9 cores to be able to boot.

 

And then there is the SoC's GPU. These companies are completely unwilling to open source their GPU drivers for their SoC's, and because nobody has their SoC's in desktop units, nobody is willing to reverse engineer the SoC GPU's to build an open source driver to keep the device operating beyond it's 2-year life.

 

Honestly, what ARM-as-a-PC needs is a stable reference platform, which none exist. All Linux versions for ARM only support ARM V8 presently, and the compiler has to support the specific SoC. V9 support only started appearing in compilers last month. https://www.phoronix.com/scan.php?page=news_item&px=Armv9-GNU-Toolchain-Starting

 

https://intel.github.io/llvm-docs/clang/ReleaseNotes.html

 

So a V9 CPU might be able to boot a V8 OS, assuming it wasn't compiled to use any features unique to another SoC. These SoC vendors need to get together and say they will support all X features in their SoC's so that OS's and software will be compiled to use those features and only those features.

 

[LAwLz]

10 hours ago, FnigePython said:

Now will phone manufacturers use those chips? I remember the Dimensity 1000 chip which iirc was actually powerful but i dont remember many phones really using them

Yep, the Dimensity 1000 had a really competitive CPU as well. It was the other stuff that was kind of lacking, like the GPU and ISP.

But it seems like MediaTek are aiming to fix those issues with the 9000. 

I am really excited for this chip. It seems like they have rectified almost all of their issues and might actually come out on top this generation.

 

10 hours ago, igormp said:

I believe that's the first SoC that's both using ARM's newest Cortex-X2 and TSMC's 4nm, so I wouldn't doubt that would be the fastest consumer chip out there (let's see if it can match Apple's A15).

It's a first for many things.

• First SoC announced with X2 CPU cores.
• First SoC announced with A710 CPU cores.
• First SoC announced with A510 CPU cores.
• By extension, the first Armv9 SoC.
• First SoC with the Mali-G710 GPU (I think?)
• First SoC with LPDDR5X.
• First SoC announced that is being built using TSMC N4.

At least if we are looking at consumer stuff that has been announced. 

[RejZoR]

Mediatek is actually the largest mobile chip maker by volume. They just don't make flashy headlines like Qualcomm or Apple do. Their chips are in everything from TV's, tablets, smartphones, digital photo frames, cars etc.

 

Dimensity 1000 wasn't bad, but it just didn't end up in many phones and apparently it was 1000L variant which was underpowered compared to "regular" Dimensity 1000. I can count like 5 phones that came with it and most of them didn't even end up in Europe...

[Doobeedoo]

Well, I look forward to see new phone SoCs performance from both, then again, they need to offer something equal like Apple does eventually.

[Sauron]

57 minutes ago, Kisai said:

Honestly, what ARM-as-a-PC needs is a stable reference platform, which none exist. All Linux versions for ARM only support ARM V8 presently, and the compiler has to support the specific SoC. V9 support only started appearing in compilers last month. https://www.phoronix.com/scan.php?page=news_item&px=Armv9-GNU-Toolchain-Starting

 

If there were real industry interest for Linux on V9 it wouldn't take long for it to be fully supported. It's just a question of whether someone wants to take the plunge. More than OS support, what would be lacking is application support; there is software for macOS on ARM that can take advantage of Apple's SoCs' strengths, not so much for Linux.

[igormp]

2 hours ago, Kisai said:

Linux is a bit better and worse in this regard as the SoC support might be in the kernel, but if your SoC goes kaput and you need to switch to another ARM SoC, you're pretty much hosed since you won't be able to just pull the drive out and boot it on another ARM SoC

Except you can boot in new drivers for the new SoC. Don't even need to wait for it to be upstream, like most manufacturers used to do.

2 hours ago, Kisai said:

ARM chips are if anything less backwards-compatible than x86-64 chips, entirely because the instruction set changes from version to version, and there can be cases where one SoC will not have the right configuration of V7/V8/V9 cores to be able to boot.

The problem with compatibility has nothing to do with the instruction set (in fact, you can run an armv4-compiled binary onto an armv9 chip with aarch32 enabled without problems).

The problem lies within the peripherals, which usually require appropriate drivers for each SoC.

2 hours ago, Kisai said:

And then there is the SoC's GPU. These companies are completely unwilling to open source their GPU drivers for their SoC's, and because nobody has their SoC's in desktop units, nobody is willing to reverse engineer the SoC GPU's to build an open source driver to keep the device operating beyond it's 2-year life.

Freedreno for the Adreno and Lima/Panfrost for Mali are a thing.

 

2 hours ago, Kisai said:

and the compiler has to support the specific SoC

No, it doesn't, it just needs to know which ISA it's targeting and not the whole SoC.

 

2 hours ago, Kisai said:

V9 support only started appearing in compilers last month. https://www.phoronix.com/scan.php?page=news_item&px=Armv9-GNU-Toolchain-Starting

 

https://intel.github.io/llvm-docs/clang/ReleaseNotes.html

Now I'm just nitpicking, but Clang started supporting ARMv9 in Clang14 since september: https://www.mail-archive.com/cfe-commits@lists.llvm.org/msg246683.html

 

36 minutes ago, Sauron said:

If there were real industry interest for Linux on V9 it wouldn't take long for it to be fully supported. It's just a question of whether someone wants to take the plunge. More than OS support, what would be lacking is application support; there is software for macOS on ARM that can take advantage of Apple's SoCs' strengths, not so much for Linux.

Vendors usually target those features/abstractions on Android, not on linux. There's no equivalent to something like the NNAPI on linux, sadly. Heck, even hardware video acceleration isn't available in browsers on linux

[LAwLz]

4 hours ago, RejZoR said:

Mediatek is actually the largest mobile chip maker by volume. They just don't make flashy headlines like Qualcomm or Apple do. Their chips are in everything from TV's, tablets, smartphones, digital photo frames, cars etc.

 

Dimensity 1000 wasn't bad, but it just didn't end up in many phones and apparently it was 1000L variant which was underpowered compared to "regular" Dimensity 1000. I can count like 5 phones that came with it and most of them didn't even end up in Europe...

I'd say the Dimensity 1000 was an alright chip.

It was basically a mid range SoC, but the CPU portion was high end in 2020. Then they dropped the ball and didn't have a real high end offering for 2021.

Now it seems like they are not only coming back for 2022, but they also are focusing on making the entire SoC high end, not just the CPU. 

 

 

3 hours ago, igormp said:

Now I'm just nitpicking, but Clang started supporting ARMv9 in Clang14 since september: https://www.mail-archive.com/cfe-commits@lists.llvm.org/msg246683.html

And let's not forget Armv9 was released was released march 30 2021. So it's only been a couple of months and we are already seeing software and hardware support for it. 

[hishnash]

On 11/20/2021 at 12:00 AM, Kisai said:

 

Honestly, what ARM-as-a-PC needs is a stable reference platform, which none exist. All Linux versions for ARM only support ARM V8 presently, and the compiler has to support the specific SoC. V9 support only started appearing in compilers last month. https://www.phoronix.com/scan.php?page=news_item&px=Armv9-GNU-Toolchain-Starting

 

https://intel.github.io/llvm-docs/clang/ReleaseNotes.html

 

So a V9 CPU might be able to boot a V8 OS, assuming it wasn't compiled to use any features unique to another SoC. These SoC vendors need to get together and say they will support all X features in their SoC's so that OS's and software will be compiled to use those features and only those features.

From falling the Linux on M1 progress I would not be surprised if the best ARM64 machines to run linux will be Macs.  As you point out every single ARM SOC has its own way of handling things, there is no such thing as the IBM PC standard or UEFI standard that abstracts this out nicely, in particular when your thinking about non-server stuff like GPUs, Video encoders/decoders Co-proseosrs for display or auxiliary compute (Tensor or matrix math) or security stuff for crypto and disk encryption.  While apple is also doing everything custom at least it appears they have opted to abstract out most of the lower level implementation of this to firmware that is loaded pre-boot for example the challenge of getting the display controller to work is figuring out how to talk to that co-prosorss OS rather the figuring out the raw instructions, registers etc that need to be poked to get proper display port handshakes to work.

in the x86 space we are used to just needing to support intel/amd and both actively contribute to the linux kernel all be it sometimes with some lag for non server features such as midea engines etc.  Maybe if we end up with an arm vendor that sells into the server space and the consume space and opts to use the same arcituture for both we will get this level of support, for free. 


 

[LAwLz]

17 minutes ago, hishnash said:

As you point out every single ARM SOC has its own way of handling things, there is no such thing as the IBM PC standard or UEFI standard that abstracts this out nicely

There are standards for it. They are just not being used in consumer hardware. 

https://www.basicinputoutput.com/2021/08/arms-systemready-specifications.html?m=1

 

[Jaesop]

On 11/18/2021 at 7:53 PM, dizmo said:

Don't they just use ARM cores?

There's more to designing a SoC than just plopping in ARM cores, though.

[hishnash]

6 minutes ago, LAwLz said:

There are standards for it. They are just not being used in consumer hardware. 

https://www.basicinputoutput.com/2021/08/arms-systemready-specifications.html?m=1

 

Yer this standard is really just for server hardware and does not cover the consume side of things, like talking to the integrated gpu or anything else. 

 

[LAwLz]

8 hours ago, hishnash said:

Yer this standard is really just for server hardware and does not cover the consume side of things, like talking to the integrated gpu or anything else. 

The standard itself isn't just for servers. It's mostly used by servers though because the makers of consumer electronics don't care. 

[williamcll]

promising, but I'll wait for the benchmarks.