AMD laptops have a hidden 10-second performance delay(arstechnica.com)
arstechnica.com
AMD laptops have a hidden 10-second performance delay
https://arstechnica.com/gadgets/2020/11/amd-laptops-have-a-hidden-10-second-performance-delay-heres-why/
30 comments
Great links. Whilst I'll admit a confirmation bias on my side, benchmarks aside, I'm pretty happy with my AMD X13. This is after literally 15 years of only purchasing Intel laptops - aside from thunderbolt, I dont feel like I've lost out on anything.
An interesting little excerpt from the first link:
> this tuning Intel does is in software and for Windows only. If you change OSes or even if the OS is upgraded, chances are this tuning will vanish.
An interesting little excerpt from the first link:
> this tuning Intel does is in software and for Windows only. If you change OSes or even if the OS is upgraded, chances are this tuning will vanish.
My first ever PC, that was mine mind you, was some cheap AMD Athlon back in 2008, after that when the i7 came out I was on that train until very recently when I picked up a 3700X instead -- so essentially 11 years of Intel for me.
I'm not really attached to either brand just what's currently the best, can run Linux, and whose owning company/companies are behaving themselves (stability/support/development).
I'm not really attached to either brand just what's currently the best, can run Linux, and whose owning company/companies are behaving themselves (stability/support/development).
I thought the "race-to-sleep" method was the most energy efficient one? Why adding such delay?
https://en.wikichip.org/wiki/race-to-sleep
https://en.wikichip.org/wiki/race-to-sleep
With really aggressive boost, the pendulum has swung the other way: the amount of performance per watt when heavily boosted nosedives as voltage to the part is cranked up and leakage increases.
Intel is cherrypicking benchmarks where the boost-on-battery behavior is worse. If you look at Cinebench, Handbrake, etc, they don't show a comparative deficit. And there's even benchmarks that show the opposite: Intel waits too long to boost on battery for e.g. Blender and performance suffers in those benchmarks.
And, of course, short benchmark runs suffer disproportionately for those integrators who chose to put a delay-before-boost timer in to improve battery life.
Basically, Intel is grasping at any straws to show performance parity to AMD at this point... and now I will be even more hard pressed to take anything coming out of INTC marketing seriously.
And, of course, short benchmark runs suffer disproportionately for those integrators who chose to put a delay-before-boost timer in to improve battery life.
Basically, Intel is grasping at any straws to show performance parity to AMD at this point... and now I will be even more hard pressed to take anything coming out of INTC marketing seriously.
Not jumping to full throttle mode without some delay while using battery is a good thing in my book.
Frequent throttling to full power wastes too much energy in the long run.
IIRC Linux CPU frequency scaling can achieve the same effect with hysteresis settings (which I used to tame my jumpy CPU back in the day).
Frequent throttling to full power wastes too much energy in the long run.
IIRC Linux CPU frequency scaling can achieve the same effect with hysteresis settings (which I used to tame my jumpy CPU back in the day).
I'd have thought the opposite : run the CPU at full power so you can terminate short tasks quickly and go back to sleep. Then if the task takes longer, throttle down as needed to regulate heat dissipation. Since there are a lot of small tasks when using a computer that would be also better for responsiveness. It's not like a car that takes a lot of energy to accelerate, you can just change the CPU clock speed as you want. What I'd like to see is a benchmark of the actual amount of energy used to perform tasks in various conditions.
Yup, if I recall from the article the actual power numbers were NDA'd but pointed towards AMD destroying Intel on a battery usage basis.
Sadly very few companies are integrating the higher end Zen chips.
Sadly very few companies are integrating the higher end Zen chips.
IIRC lately, Intel's real TDP under serious load is almost twice as the public value. I guess it's also valid to their server parts to a certain degree.
Intel has a habit of hiding some ugly things about their chips under NDA until it has competition or the chip is last-gen. The uglier thing is they're adding these values back to their documents later as a revision and telling look, it was already there! (cough, AVX base frequencies, cough!).
Competition and being a very big company changed them... a lot.
Intel has a habit of hiding some ugly things about their chips under NDA until it has competition or the chip is last-gen. The uglier thing is they're adding these values back to their documents later as a revision and telling look, it was already there! (cough, AVX base frequencies, cough!).
Competition and being a very big company changed them... a lot.
Yes, that seems like a feature.
This is down to the system integrator as an option, not something that is inherient to AMD mobile processors.
I expected Ars Technica to be better than this. They took Intel's bait and spread the rumor Intel wants to spread. They've even copied Intel's slides instead of publishing their own benchmark.
So why did the "on battery" benchmark end at the same time as the full power case? Was it completing the same amount of work with less total energy expenditure?
The problem is that setting "performance" mode doesn't really recover the perf in the machine I tested. Building kernel's when the device is unplugged seems to indicate its stuck in battery save mode despite the OS requesting a higher perf profile.
Ugh, and certain OEMs (HP for one) don't let the user tweak these settings in the BIOS because they apparently want to charge people to change things like whether the laptop supports S3 mode (for linux). If one spends a bit of time reversing the bios, a lot of options show up that aren't accessible by default, including the precision boost options.
I get HP/etc doesn't want random "leet" pc users messing with the power/thermal settings, but hiding them with alt-a/fn-f1/etc advanced menus probably took care of 99.9% of the support requests while leaving people with an out. Changing the password/key-combo on the recent machine is mostly incompetence, as is failing to follow the OS's performance requests (which is seemingly a large part of the problem here).
Ugh, and certain OEMs (HP for one) don't let the user tweak these settings in the BIOS because they apparently want to charge people to change things like whether the laptop supports S3 mode (for linux). If one spends a bit of time reversing the bios, a lot of options show up that aren't accessible by default, including the precision boost options.
I get HP/etc doesn't want random "leet" pc users messing with the power/thermal settings, but hiding them with alt-a/fn-f1/etc advanced menus probably took care of 99.9% of the support requests while leaving people with an out. Changing the password/key-combo on the recent machine is mostly incompetence, as is failing to follow the OS's performance requests (which is seemingly a large part of the problem here).
Supporting more options generally need more engineering and test work, hence it drives prices up.
These settings and multi-os compatibility is generally reserved for enterprise level devices (EliteBook, XPS/Precision, Thinkapd) where these options both will be used and result in more sales.
From end-user perspective it's not nice but, it makes sense from a business perspective since most of the people doesn't care about their disposable laptops' Linux compatibility. "If it works, it works. If it doesn't, I'll buy a new one" they think.
These settings and multi-os compatibility is generally reserved for enterprise level devices (EliteBook, XPS/Precision, Thinkapd) where these options both will be used and result in more sales.
From end-user perspective it's not nice but, it makes sense from a business perspective since most of the people doesn't care about their disposable laptops' Linux compatibility. "If it works, it works. If it doesn't, I'll buy a new one" they think.
I don't buy that, sure they might actually have to test their device for conformance with the requisite standards. OTOH most of the far-eastern brands manage to provide those options and they work. That is because the core parts are provided by intel/amd. How is it that msi/asus/supermicro/etc manage to afford all those options on devices that cost 1/2 of what the average HP device does?
HP/etc can't even get windows working correctly on a device that ships with windows. That is fundamentally what the article is about (the OS pref requests not being honored properly).
Its because HP/Dell/etc aren't really providing any real validation above the underlying ODM anymore. So whether the S3/whatever mode actually works is more a function of whether intel/amd and their bios provider insyde/ami/etc managed to get the option hooked up at all. So if you happen to enable S3 with some funny device and it doesn't come out of standby properly, then sure its a bug, you have to disable that device or disable S3.
But what really happens with the HP/etc laptops is you don't have the choice. Which is why you see endless complaints about modern standby issues on their "windows" devices despite running the supported OS. If the end user could just go and ignore the MS recommendations and disable modern standby on a laptop class device they likely would be much happier. Having the machine work reliably vs download email when the lid is closed is an easy choice.
In summary, the "value add" the big OEM's are adding is product segmentation by removing options the underlying technology providers ship by default.
HP/etc can't even get windows working correctly on a device that ships with windows. That is fundamentally what the article is about (the OS pref requests not being honored properly).
Its because HP/Dell/etc aren't really providing any real validation above the underlying ODM anymore. So whether the S3/whatever mode actually works is more a function of whether intel/amd and their bios provider insyde/ami/etc managed to get the option hooked up at all. So if you happen to enable S3 with some funny device and it doesn't come out of standby properly, then sure its a bug, you have to disable that device or disable S3.
But what really happens with the HP/etc laptops is you don't have the choice. Which is why you see endless complaints about modern standby issues on their "windows" devices despite running the supported OS. If the end user could just go and ignore the MS recommendations and disable modern standby on a laptop class device they likely would be much happier. Having the machine work reliably vs download email when the lid is closed is an easy choice.
In summary, the "value add" the big OEM's are adding is product segmentation by removing options the underlying technology providers ship by default.
> "If it works, it works. If it doesn't, I'll buy a new one" they think.
Nobody thinks that. Laptops are expensive.
> From end-user perspective it's not nice but, it makes sense from a business perspective
Lots of deplorable things make sense from a business perspective. That's a very low bar to clear. "Intel's actions aren't literally counterproductive to their short-term interests". Yes, and so? They're not literally motivated by spite for their users. Is that supposed to make it better?
Nobody thinks that. Laptops are expensive.
> From end-user perspective it's not nice but, it makes sense from a business perspective
Lots of deplorable things make sense from a business perspective. That's a very low bar to clear. "Intel's actions aren't literally counterproductive to their short-term interests". Yes, and so? They're not literally motivated by spite for their users. Is that supposed to make it better?
> Laptops are expensive
My non sw developer friends look for Windows laptops in the 300-400 Euro range. Things with flexible keyboards, maybe still 768p screens, a large HDD plus possibly a small SSD, 8 GB RAM, i3 or i5, a couple of USB ports, one HDMI. That's all they need.
If I have to explain them why my laptop costs plus than 1000 more than that I only tell them that it's for work, every day all the time.
My non sw developer friends look for Windows laptops in the 300-400 Euro range. Things with flexible keyboards, maybe still 768p screens, a large HDD plus possibly a small SSD, 8 GB RAM, i3 or i5, a couple of USB ports, one HDMI. That's all they need.
If I have to explain them why my laptop costs plus than 1000 more than that I only tell them that it's for work, every day all the time.
> Nobody thinks that. Laptops are expensive.
For people who know and buy what they need, they're indeed expensive but, some consumers see computers as consumable entities. They buy a machine, use it 2-3 years and buy a new one then. This is my experience at the least. On the contrary, I tend to use every device I own 7-10 years before replacing.
> Lots of deplorable things make sense from a business perspective. That's a very low bar to clear. ... Is that supposed to make it better?
Saying that doesn't mean I support business interests over legitimate user needs but, businesses are greedy entities which always want more money so, they always follow the path which brings more money. It doesn't make any better either. I want to be able to buy any laptop and/or device I want and change the software stack and tune it the way I wish however, as we see today, this freedom is attacked more and more.
Unless consumer choice generates more profits (which is unlikely), we won't see that trend change much.
For people who know and buy what they need, they're indeed expensive but, some consumers see computers as consumable entities. They buy a machine, use it 2-3 years and buy a new one then. This is my experience at the least. On the contrary, I tend to use every device I own 7-10 years before replacing.
> Lots of deplorable things make sense from a business perspective. That's a very low bar to clear. ... Is that supposed to make it better?
Saying that doesn't mean I support business interests over legitimate user needs but, businesses are greedy entities which always want more money so, they always follow the path which brings more money. It doesn't make any better either. I want to be able to buy any laptop and/or device I want and change the software stack and tune it the way I wish however, as we see today, this freedom is attacked more and more.
Unless consumer choice generates more profits (which is unlikely), we won't see that trend change much.
hmmm, this actually sounds good: got the same gaming experience and can same some battery when browsing web pages. Wondering how to configure my Intel laptop to do it.
Agreed, this actually sounds like spot-on for a common use case: you've got your laptop with you, you've been working a while or gaming and you close it as your battery starts to ebb. But a little while later, you realize you need to check your calendar or email for something for your upcoming meeting/date. Open up the laptop, find the info, close it again. You really didn't need 100% poer to read an address, did you?
It's not so great when you change one cpp file and run make.
Can you open up the editor, edit the file, save it, then exit or bg the editor (or go to another window) and type 'make' in 10 seconds or less?
It will harm browsing. Pages typically load within 10 seconds, and people rarely binge-navigate at high rate.
Owning a ryzen 4800u laptop I can confirm and it's actually an annoyance for me.
My main issue is responsiveness is worse on battery, opening apps and pages is noticably slower on my 4800u compared to my i7-8550u on battery.
Changing performance and power modes has no effect on this, the 10s delay remains intact and only power used after that changes.
This was the case on my S540-13ARE, my current Slim 7 and also have heard it from matebook users.
For many these shorter task uses is what they actually use their laptops for, not for encoding and running cpu stress benches. So actually intel does have a point here (and it happens to be something I noticed when buying these laptops pretty fast and is indeed an issue for me)
My main issue is responsiveness is worse on battery, opening apps and pages is noticably slower on my 4800u compared to my i7-8550u on battery.
Changing performance and power modes has no effect on this, the 10s delay remains intact and only power used after that changes.
This was the case on my S540-13ARE, my current Slim 7 and also have heard it from matebook users.
For many these shorter task uses is what they actually use their laptops for, not for encoding and running cpu stress benches. So actually intel does have a point here (and it happens to be something I noticed when buying these laptops pretty fast and is indeed an issue for me)
Intel's performance on-battery also varies when battery level is 50% or 80%: https://www.phoronix.com/scan.php?page=article&item=tiger-re...
Arstechnica and Wired was once really good, now it fells they try to abort themselves...and are pretty successful in that matter.
Here this is debunked: https://semiaccurate.com/2020/11/23/intel-looks-at-laptop-po...
Another debunk: https://www.extremetech.com/extreme/317657-intel-is-spreadin...
AMDs own slide on performance tuning OEMs have access to: https://www.extremetech.com/wp-content/uploads/2020/11/AC-vs...