Aggressively Pursuing New Markets
ARM gives us a glimpse of upcoming products and their performance.
ARM has had a pretty fascinating history, but for most of its time on this Earth it has not been a very public facing company. After the release of the iPhone and ARM’s dominance in the mobile market, they decided to push their PR efforts up a few notches. Now we finally were able to see some of the inner workings of a company that was once a little known low power CPU designer that licensed cores out to third parties.
The company was not always as aggressive as what we are seeing now. The mobile space for a long time was dominated by multiple architectures that all have eventually faded away. ARM held steady with design improvements and good customer relations that ensured that they would continue into the future. After the release of the original iPhone, the world changed. Happily for us, ARM changed as well. In previous years ARM would announce products, but they would be at least three years away and few people took notice of what they were up to. I originally started paying attention to ARM as I thought that their cores might have the ability to power mobile gaming and perhaps be integrated into future consoles so that there would be a unified architecture that these providers could lean upon. This was back when the 3DS and PSP were still selling millions of units.
This of course never came to pass as I had expected it to, but at least ARM did make it into the Nintendo Switch. ARM worked hard to quickly put faster, more efficient parts out the door. They also went on a buying spree and acquired several graphics startups that would eventually contribute to the now quite formidable Mali GPU family of products. Today we have an extensive lineup of parts that can be bundled into a tremendous amount of configurations. ARM has a virtual monopoly in the cellphone market because they have been willing to work with anyone who wants to license their designs, technologies, and architectures. This is actually a relatively healthy “monopoly” because the partners do the work to mix and match features to provide unique products to the marketplace. Architectural licensees like Apple, Qualcomm, and Samsung all differentiate their products as well and provide direct competition to the ARM designed cores that are licensed to other players.
Today we are seeing a new direction from ARM that has never been officially explored. We have been given a roadmap of the next two generations of products from the company that are intended to compete in not only the cellphone market, but also in the laptop market. ARM has thrown down the gauntlet and their sights are set on Intel and AMD. Not only is ARM showing us the codenames for these products, but also the relative performance.
ARM claims that the push for always on, always connected 5G devices is transformational to the industry and the need for performance and long battery life will be accentuated in these use cases. ARM feels that they have the product stack that is more able to address the upcoming 5G rollouts. The latest Cortex-A76 is a significant jump in performance from the previous A75 and A73 products. It is able to achieve this through a lot of clever design work even while utilizing similar process nodes as the previous cores. The very latest process nodes will allow it to further clock higher, perform better, yet still retain the low power consumption that ARM has staked its brand upon.
The next two generations of products are codenamed “Deimos” which will debut at 7nm in 2019 and then followed by “Hercules” which will be produced on 7nm and eventually 5nm in a 2020 timeframe. While leaks happened in the past with ARM about codenames and upcoming parts, this is the first time that ARM has ever released a public roadmap of where they are going with their compute cores. Not only are they releasing the names and what likely nodes they will be produced on, but also are showcasing what actual performance may look like and how they compare to well known Intel mobile chips.
ARM compares the performance of a 3 GHz A76 against the previous generation Intel i5-7300U. At 3 GHz the chip sits just below the turbo boosted 7300U at 3.5 GHz. This is running SPEC2006INT. There are several caveats to this result. First off this is a previous gen mobile part from Intel rather than their latest 8000 series released this past year. This is also a single threaded workload result so it does not address multi-threaded performance. Next is that it is a simulated results from ARM rather than having it run on an actual chip (for now). While this is not exactly an accurate test, it does show the potential for ARM to compete in a landscape that it has only barely set foot into.
If ARM can back up these results with real hardware then Intel has a strong potential competitor on its hands. The Cortex-A76 result was projected to be in the sub 5 watt TDP range, and that includes the entire SOC that would be running. We would likely see configurations that could extend to 4 x A76 cores to hybrids like 2 x A76 and 2 x A55. These SOCs with integrated graphics could easily fit in the sub-5 watt TDP range while clocking the cores at the highest speeds. Consider that most cellphone SOCs are typically sub 1 watt TDP parts. In a larger chassis with better cooling options and a much larger battery, ARM SOCs could potentially provide near the performance of the latest Intel parts, but at under a third of the power dissipated. This means cooler running laptops with a far greater battery life than what the competition can provide.
We have seen ARM products in laptops as well as Windows 10 on ARM. With a virtual stranglehold on the cellphone market, ARM is looking to leverage its expertise and newly expanded marketing program on this very lucrative market. ARM is not currently announcing any new directives at this time, but they did hint that at the upcoming ARM TechCon they will be aggressively pushing the laptop platform.
ARM has a long way to go before they can get significant marketshare in the very competitive laptop market. Not only is Intel firmly entrenched, but AMD is also looking to make strides in this area with their latest mobile Zen based parts. Neither of those companies are standing still with Intel having a comprehensive roadmap leveraging improvements in design as well as advances on their 14nm processes. AMD will be releasing Zen2 based parts next year and will utilize advanced 7nm process nodes from their foundry partners as well.
If ARM’s performance estimates hold true and they can achieve a user experience competitive with Intel and AMD, then they have a very good chance of claiming a significant piece of the pie. The combination of competitive performance, better battery endurance, native Windows support, and the integration of 4G and 5G modems can give them an edge on the competition. It will be a challenge to change hearts and minds, but in the end features and price points can convince many to switch over. ARM is certainly entering a new era with their rapid expansions fueled by the SoftBank investment. The company certainly has its sights set on continuing this growth in many new directions.
Unless ARM can hugely
Unless ARM can hugely outprice AMD/Intel or there performance/power efficiency is significantly better, I’d be concerned that there is no particularly good reason for consumers/laptop manufacturers to switch from a large established PC/laptop x86 ecosystem to ARM.
What OS will they use? What software? shrug. It’s like intel trying to make mobile phone processors, they were alright, but there is just not enough reason to go with them. And it’s not like we haven’t had ARM processors in notebooks (chromebooks, etc).
I don’t know, feels like a weird move. Either way, would still be cool to have a high performance ARM laptop
I believe there is already a
I believe there is already a Windows 10 for ARM. It can do x86 emulation as well as natively run ARMv8 applications.
It exists but all accounts
It exists but all accounts its not very good and the emulation costs you no small amount of performance and power too.
The problem with ARM adoption has always been one of developer support and ultimately if all they can do is meet current Intel/AMD x86 CPU performance than they won’t get it.
Historically the old RISC HPC architectures like SPARC and Alpha n’ such were once able to beat x86 performance handily and they still got pushed off to the wayside and are now either defunct or borderline so.
If ARM wants to succeed as a PC it’ll have to beat current x86 CPU performance by large margins (ie. 40-50%+) in order to attract more developer support and even then it’ll be a real slog to gain any significant market share. ARM has been trying for years to compete on power usage and cost and it STILL hasn’t really made large in roads yet in the mobile PC, PC or server markets.
The only market place where ARM has done well is in markets where the devices are mainly limited by power or cost really and those aren’t markets where a high performance ARM ~5w chip will likely sell well in since they’ve had to get the power consumption too high up to get performance this high.
Actually, everything compiles
Actually, everything compiles nicely on ARM and the developer community nowadays doesn’t care much about the CPU vendor – as long as toolchains are available.
Which OS? Between iOS, android, chrome os, ubuntu and various linux distros, choice is much better than the microsoft’s upcoming subscription-based model for windows.
ARM has come a long way and has an excellent strategy. It will be interesting to see how this plays out over the next few years.
I didn’t say anything about
I didn’t say anything about compiling problems for a reason though. And just because something compiles doesn’t mean it’ll run well or efficiently. Or is all that bug free or stable.
Quite a lot of stuff can be easily made to run on ARM with tools if you don’t care about performance, power usage, stability, or security much. Which is quite typical with much of the x86 to ARM ports these days.
Windows ARM laptops and tablets have been around for years now but have gone nowhere for a reason.
And iOS, ChromeOS, etc all suffer from the same problem: they’re not windows.
ARM has a solid business approach for getting its CPU’s into highly power constrained and cost sensitive hardware. For any market beyond that they’ve had limited to no success despite over a decade worth of effort at this point. ARM might still get popular at some point of course…but there is no reason to be optimistic or think that day will be coming anytime soon.
“Apple, Qualcomm, and Samsung
“Apple, Qualcomm, and Samsung all differentiate their products”
Hell Apple’s custom cores that are designed fully by Apple and are engineered by Apple to execute the ARMv8A ISA(Apple acquired P.A. Semiconductor and got a that whole CPU design company’s allotment of CPU engineers including P.A. Semi’s VP of Engineering, at the time, Jim Keller). And Apple, with Its custom A7 Cyclone design, beat ARM Holdings to market with the first CPU that Ran/Executed Arm Holdings’ ARMv8A 64 bit ISA.
Samsung’s Mongoose M3 is just as wide order superscalar as the Apple A series CPU designs and that Mongoose core is a 6 decode and 12 wide instruction dispatch at 12 µops on its schedulers that schedule for the M3 core’s 12 execution ports(INT, FP, Branch, Load/store Etc.) The M3 has a reorder buffer size of 228 entries and looks even more like a desktop CPU than even Apple’s A series SKUs which used to be out ahead at one time before the Mongoose M3 was announced.
That’s also leaving out Cavium’s ThunderX2, which is not based on any ThunderX DNA, Cavium acquired Broadcom’s Vulkan CPU core design from Avago after Broadcomm was acquired by Avago. And that ThunderX2(Vulkan) custom ARM server core has SMT4 capability.
Really that Arm Holdings Refrence Design A76 is not even able to match Apple’s(A Series)/Samsung’s(M2 and M3) core performance and Intel’s and AMD’s x86 cores both support SMT2(2 processor threads pre core). That Deimos core better be at least a 6 instruction issue design just to match Apple’s and Samsung’s latest high Power custom cores.
And If the custom ARM CPU makers, outside of Cavium, do not look at even introducing at least SMT2(2 Processor threads per core) like both Intel and AMD then that’s a load of wasted execution pipline resources on any NON-SMT core ARM design that can not keep those CPU core execution piplines fully fed with work because they lack SMT capability.
Real Laptop/Desktop CPUs need CPUs with that SMT support what with the multi-threaded execution resources utilization demands of laptop/desktop OSs be they full Linux Distro or Windows/other Kernel based OSs. Laptops and desktop processors have to deal with heavily multi-tasking capable OSs compared to Phones and Tablets. The Linux Kernel will be fine for running Android on top for tablets and phones running simple apps but any real Linux Full DE(Desktop Environment) and Desktop Application software like Blender/Other applications and that’s a bit more multi-threaded than simple Android apps.
I’m sure Arm Holding’s refrence designs are great for low power Phones for smaller makers who do not have their own in-house CPU design capabilities, but Laptop performance is a bit of a streatch for Arm Holdings when that i5 7300u has a cTDP of between 15W-25W with SMT abilities and The A76 is most definitely going to beat any x86 on the Wattage/TDP and power usage front but laptop’s need at least 15+ watts or the Laptop Grade OS is not going to perform so well in heavy multitasking OS/software usage.
Apple’s Phone SKUs with the A series cores offload loads of work onto those specilized AI/DSP processor cores that Apple makes use of and Apple also makes use of Big/Little core parings, but at least Apple, and Samsung, have those Big cores there when needed while that Arm Holdings A76 is relatively small even by Apple’s small A series core standards. The A76 core may find a home paired with Samusung’s Powerful M3 cores in tablets/chromebooks but Apple and Samsung will have to begin thinking about adding SMT to their custom high power ARM core designs if they want to fully enter the Laptop SKU market.
Samsung’s Exynos 9810 uses 4 Mongoose-3 big cores operating at 2.9 GHz and 4 Cortex-A55 little cores operating at 1.9 GHz. Apple’s current A11 cores have a lot of specilized custom core IP that Apple specifically offloads workloads to in order to get the most processing done with as little power used as possible on its phones/tablets.
I do not see Apple or Samsung’s cores actually beating any x86 cores outright in muiti-threaded workloads until Apple or Samsung get SMT working for their respective custom ARM core SKUs. The ThunderX2 is a server core with SMT4, and AMD’s K12(SMT capable?, well Jim Keller was hinting at that) is probably in mothballs until AMD decides if they will have enough of a market to justify moving forward. Maybe a semi-custom client for AMD’s K12 will step forward for some K12 custom ARM Tablet/Portable Gaming design APU with Vega graphics. AMD appears to be letting it’s semi-custom clients fund a lot these days.
@serpico:
For ARM there are a
@serpico:
For ARM there are a number of operating systems with a good number of applications (more than just a few ten thousand) and for quite large portion of computer users today the main application they use is anyway just the browser.
I do see that microsoft already lost the “software wars”, they even begun to abandon their own OS to use the worlds most used OS, they are using OpenSource applications and just re-branding them to look like they have ported their own applications (just look at VS for Mac which is just old MonoDevelop which been around for Mac, Linux, BSD, QNX and even MS-Windows for ages). If microsoft wants to keep on with software, they better start to look into how to supply ARM native and soon also RISC-V native applications.
The big income for microsoft nowadays are services like Azuer, FUD-licenses, and Patent-Trolling.
Pretend you are a
Pretend you are a laptop/desktop user.
Pretend you are a non-techy windows user that maybe uses mac as well (this is probably a large chunk of users, agree?).
I present to your two laptops. They have the same performance (execution speed, battery life, etc). Laptop A has the software and OS you are used to.
Laptop B has an OS you are either not used, software you are not familiar with (maybe some software for it doesnt even exist, you dont know!), or you can run it in some emulation mode where you can use both the OS you want and the software you want but it runs all of your software 2x slower .
Let’s say you pretty much only use the browser on you computer.
But maybe you load up some other application here or there. Why would you choose Laptop B over A?
Let’s pretend Laptop B has 5% better overall performance, but still all of the other problems. Is this enough to go with Laptop B?
How much better performance do you think you need before you decide to switch to LaptopB as a typical user?
Maybe ARM can somehow switch it into some sort of marketing problem? I mean, how many users even know how there laptop performs? Forget performance, how can we out advertise our laptops? I dont know, still seems like a weird decision from ARM unless they have some awesome strategy/ performance that we are not aware of.
Just look at the Chromebooks,
Just look at the Chromebooks, they kind of match your laptop A/ laptop B scenario and still they sell far better than other machines in the same size/performance class, so this shows that it’s not any more necessary to have the same applications on a device you used to have.
Sure you can use emulation to make old applications to run, sure they will be running slower than native, but don’t forget that CPU’s are getting more powerful for each generation, so running an emulation may look as the same speed as on the old computer as it had an old generation of CPU which just is a lot slower.
If the ARM chips will manage to stay low energy using with increased computing power, then you will see increase of battery life and it will even make that American company in Washington to focus more on ARM, sure with something similar to what OSX had in the beginning which allowed it to run both PowerPC and amd64 based applications at the same time. With time amd64 based application may just be some legacy applications, who knows…
I have to say that Microsoft ain’t as powerful as the Porn-industry and they didn’t manage to make HDDVD to replace DVD as they managed with replacing BetaMax with VHS, so how do you think Microsoft can stop ARM when the majority of worlds consumer computer devices uses Unix or Unix-like OS?
Don’t think from a North
Don’t think from a North American view. This will be pushed by russia, china, and anyone else that doesn’t trust silicon with NSA back doors in their respective markets.
The problem is everyone else
The problem is everyone else is putting back doors into all the x86 CPU competitors too. And shoddy ones at that apparently if the recent news about the VIA C3’s hidden “god mode” are anything to go by.
Other countries that are developing chips like this are doing so as much for the obvious potential economic benefits as to have back doors in that they control for their own national security interests.
Or at least that is how it seems to me. I have no special information and the above is mostly speculation but I don’t think you can be cynical enough about anyone’s govt. and those govt.’s views on privacy and citizen’s rights in this day and age.
Yes but in the case of that
Yes but in the case of that mode was in the documentation even if the ISA was undocumented(Infromation provided by request probably under NDA).
look at page 82 Titled “ALTERNATE INSTRUCTION EXECUTION”
of the Datasheet(1). Not eaxctly hidden even if the undocumented ISA is not published and can be obtained from Via on a need to know basis.
There are also the Standard JTAG ports for testing with that under NDA also.
So Yes there may be back doors but there are plenty of front doors also. Thate UEFI/Secure Boot has supplanted the old BIOS and some very SuperFish-ial spyware baked in at way below ring zero of the OS, in fact the UEFI is first the get control before the OS even is loaded.
(1)
“VIA C3 Nehemiah
Processor
Datasheet
Revision 1.13
September 29, 2004
VIA TECHNOLOGIES, INC.”
http://datasheets.chipdb.org/VIA/Nehemiah/VIA%20C3%20Nehemiah%20Datasheet%20R113.pdf
Vague mentions of features
Vague mentions of features are pretty far from detailed explanations of said features. And a “god mode” that allows you bust any protections is, uh, a pretty goddamn bad idea in general no matter what.
Getting at JTAG ports generally requires physical access to the machine. If an attacker gets physical access to your machine you’ve got bigger problems to worry about.
Also quibbling about this stuff doesn’t actually address or refute my broader point mentioned in that post either.
Nintendo has used ARM all the
Nintendo has used ARM all the way back to the Game Boy Advance (ARM7).
The Nintendo DSi contained an ARM7 and ARM9.
The Nintendo 3DS also sports a dual ARM processor architecture, an ARM11 and an ARM9.
Sure, Nintendo Switch contains an ARM processor within its Nvidia Tegra X1 SOC, but as a company, Nintendo has a long history with ARM, dating back to the early 2000s (I’d say late 90s if you count R&D!)
Yes, they have… but I was
Yes, they have… but I was talking more about ARM in consoles. It was all PowerPC there for a while, but I expected them to skip over x86 and go ARM up and down the line so their portables match up with the overall software ecosystem that they were developing.
I think cuz’ the devs still
I think cuz’ the devs still want to be able to sell games on the PC’s with a minimum of expense in porting things and with x86 since the performance and cost are actually quite good all things considered.
Yeah power consumption is much higher than a similarly performing ARM chip would be but unless they’re trying to do something like Nintendo’s Switch they won’t care too much since they’ll have something like a 150-250W power budget to work with.
I assume x86 won the console
I assume x86 won the console business mostly due to the GPU attached..
So ARM can defy physics and
So ARM can defy physics and make a powerful CPU that only uses 5W. I presume all these years Intel and AMD have been sitting on their thumbs while ARM will release revolutionary new technology. It may be able to keep up in the lowest end part of the laptop market, but at 5W it won’t compete with anything used for serious gaming or computing.
They’re not defying physics
They’re not defying physics but they might be juking the stats or cherry picking their benchmark somehow.
I don’t see any reason to believe that ARM couldn’t meet or beat x86 performance while using less power. The ARM ISA does have some advantages.
I also don’t however see any reason to believe that will allow them to get much PC/server market share since other factors play a much bigger role in x86’s dominance in those markets.