6 ms·

AMD laptops have a hidden 10-second performance delay

36 points by buran77 6y ago

chaz6 6y ago
This is down to the system integrator as an option, not something that is inherient to AMD mobile processors.
achinese 6y ago
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.
- cratermoon 6y ago
  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?
  - xuhu 6y ago
    It's not so great when you change one cpp file and run make.
    - cratermoon 6y ago
      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?
- lostmsu 6y ago
  It will harm browsing. Pages typically load within 10 seconds, and people rarely binge-navigate at high rate.
kevin_thibedeau 6y ago
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?
bayindirh 6y ago
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).
- rainingcatndogs 6y ago
  Yes, that seems like a feature.
- vvanders 6y ago
  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.
  - bayindirh 6y ago
    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.
- ajnin 6y ago
  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.
mlyle 6y ago
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.
tsujp 6y ago
No they don't. They have the _option_ of an N second delay which AMD allows OEMs to set themselves for their own determined blend of say pure raw performance (no delay) or a more reserved ramp up to save battery. Here this is debunked: https://semiaccurate.com/2020/11/23/intel-looks-at-laptop-power-usage/ https://semiaccurate.com/2020/11/23/intel-looks-at-laptop-po... Another debunk: https://www.extremetech.com/extreme/317657-intel-is-spreading-fud-about-supposedly-huge-ryzen-4000-performance-drops-on-battery 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-DC-Power.jpg https://www.extremetech.com/wp-content/uploads/2020/11/AC-vs...
- antpls 6y ago
  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
  - mlyle 6y ago
    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.
- srazzaque 6y ago
  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.
  - tsujp 6y ago
    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).
a012 6y ago
Intel's performance on-battery also varies when battery level is 50% or 80%: https://www.phoronix.com/scan.php?page=article&item=tiger-renoir-battery&num=8 https://www.phoronix.com/scan.php?page=article&item=tiger-re...
StillBored 6y ago
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).
- bayindirh 6y ago
  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.
  - FeepingCreature 6y ago
    > "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?
    - 6y ago
quyleanh 6y ago
Benchmark on battery? It's so ridiculous from Intel marketing team.
- jhoechtl 6y ago
  OTOH benchmarking a Laptop CPU while running on battery is the regular use-case no?
pornel 6y ago
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.
nix23 6y ago
Arstechnica and Wired was once really good, now it fells they try to abort themselves...and are pretty successful in that matter.
dagrim1 6y ago
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)

36 points by buran77 6y ago

This is down to the system integrator as an option, not something that is inherient to AMD mobile processors.

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.

cratermoon 6y ago
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?
- xuhu 6y ago
  It's not so great when you change one cpp file and run make.
  - cratermoon 6y ago
    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?
lostmsu 6y ago
It will harm browsing. Pages typically load within 10 seconds, and people rarely binge-navigate at high rate.

kevin_thibedeau 6y ago

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?

bayindirh 6y ago

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).

rainingcatndogs 6y ago
Yes, that seems like a feature.
vvanders 6y ago
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.
- bayindirh 6y ago
  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.
ajnin 6y ago
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.

mlyle 6y ago

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.

tsujp 6y ago

No they don't. They have the _option_ of an N second delay which AMD allows OEMs to set themselves for their own determined blend of say pure raw performance (no delay) or a more reserved ramp up to save battery. Here this is debunked: https://semiaccurate.com/2020/11/23/intel-looks-at-laptop-power-usage/ https://semiaccurate.com/2020/11/23/intel-looks-at-laptop-po... Another debunk: https://www.extremetech.com/extreme/317657-intel-is-spreading-fud-about-supposedly-huge-ryzen-4000-performance-drops-on-battery 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-DC-Power.jpg https://www.extremetech.com/wp-content/uploads/2020/11/AC-vs...

antpls 6y ago
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
- mlyle 6y ago
  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.
srazzaque 6y ago
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.
- tsujp 6y ago
  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).

a012 6y ago

Intel's performance on-battery also varies when battery level is 50% or 80%: https://www.phoronix.com/scan.php?page=article&item=tiger-renoir-battery&num=8 https://www.phoronix.com/scan.php?page=article&item=tiger-re...

StillBored 6y ago

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).

bayindirh 6y ago
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.
- FeepingCreature 6y ago
  > "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?
  - 6y ago

quyleanh 6y ago

Benchmark on battery? It's so ridiculous from Intel marketing team.

jhoechtl 6y ago
OTOH benchmarking a Laptop CPU while running on battery is the regular use-case no?

pornel 6y ago

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.

nix23 6y ago

Arstechnica and Wired was once really good, now it fells they try to abort themselves...and are pretty successful in that matter.

dagrim1 6y ago

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)