IMPORTANT Explaining the AMD Ryzen "Power Reporting Deviation" -metric in HWiNFO
- Thread starter The Stilt
- Start date
Shamino has unleased a guinea pig bios that allows the CPU and SOC telemetry to be changed. Says it goes by milliAmperes... But It incriments in 1000's and I have it set to -6000 for this result
So its working, but its more like 60 milliamps, not 6000 :3 I'll see if I can old-taichi mode it.
Took a setting of - 50000 to make it match that Taichi result :3 didnt even get to 142w on the reading but the temperature and the scores are increased.
So its working, but its more like 60 milliamps, not 6000 :3 I'll see if I can old-taichi mode it.
Took a setting of - 50000 to make it match that Taichi result :3 didnt even get to 142w on the reading but the temperature and the scores are increased.
Last edited:
The CPU has also some other means of knowing how much certain parts should be consuming and this is expected to be quite accurate. Sorry, but we can't go into further details.
bogdanbggg
New Member
Ryzen 3900x , asus x570 tuf gaming plus => load optimized settings in bios and enable DOCP , memory @3200 mhz.
Is this ok? 100.1% at full load @cinebench r20
Is this ok? 100.1% at full load @cinebench r20
Perfectly. Values between 95 - 105% are well OK.
Here is a complementary post to the OP, I made to Reddit few moments ago.
Just to be perfectly clear, since there appears to be some unfounded concerns related to this, especially after the a bit of a popcorn fart of a headline Tom's Hardware pulled yesterday.
First of all, while this exploit CAN reduce the life-span of the CPU in the same manner as intentional overclocking would, there really is no risk in outright burning the CPU like Tom's Hardware article appears to be suggesting.
Despite this exploit effectively increases the power limit of the CPU, it doesn't affect the built-in functionality (FIT), that is there to monitor and to protect the silicon from excessive wear and damage. I assumed this would have been apparent from the original write-up, where I wrote:
" With 150A setting (50% of the actual), the average HWiNFO "Power Reporting Deviation" during Cinebench R20 NT is 50.2%. With this setting, the average CPU core frequency is 4106.6MHz, power consumption seen by the CPU 91.553W (of 142W limit) and peak CPU temperature of 79°C. This setting is already limited by maximum voltage allowed by the silicon fitness (FIT), so there were pretty much no additional performance gains, or ill-effects for that matter to be had. "
So when the allowed voltage is still being limited in the same exact way, and at the same exact point with and without this exploit, it is pretty apparent that there is no real risk in outright killing the CPU involved here. Increasing the power limit WILL increase the voltage per the normal voltage-frequency curve (higher frequency == higher voltage), but only within the bounds set by the silicon fitness monitoring feature (FIT).
This exploit is nothing but a clandestine way to increase the power limit.
In order to produce comparable results, please follow the workflow specified in the OP.
- Use Cinebench R20 NT (multithreaded only).
- Part load and idle reading are completely irrelevant and should be ignored (no need to state them here either, for clarity sake).
- Please test at stock settings. These include voltage offsets (which can and will cause clock stretching), load-line adjustments and manual overclocking. The main idea here is to allow the CPU to stay in control of as many of its parameters as possible.
Originally this feature was intended only for the 3rd gen. Ryzen CPUs and I've personally only tested it on them. Martin (Mumak) tested this briefly on 2nd gen. Ryzen. and at least in that case it appeared to work fine. However, based on the couple of user reports here the 2nd and especially first gen. Ryzen deviation reporting looks somewhat suspicious. Because of that we'll need to look more into that, to ensure that the reporting also on 1st and 2nd gen. parts is working correctly.
Just to be perfectly clear, since there appears to be some unfounded concerns related to this, especially after the a bit of a popcorn fart of a headline Tom's Hardware pulled yesterday.
First of all, while this exploit CAN reduce the life-span of the CPU in the same manner as intentional overclocking would, there really is no risk in outright burning the CPU like Tom's Hardware article appears to be suggesting.
Despite this exploit effectively increases the power limit of the CPU, it doesn't affect the built-in functionality (FIT), that is there to monitor and to protect the silicon from excessive wear and damage. I assumed this would have been apparent from the original write-up, where I wrote:
" With 150A setting (50% of the actual), the average HWiNFO "Power Reporting Deviation" during Cinebench R20 NT is 50.2%. With this setting, the average CPU core frequency is 4106.6MHz, power consumption seen by the CPU 91.553W (of 142W limit) and peak CPU temperature of 79°C. This setting is already limited by maximum voltage allowed by the silicon fitness (FIT), so there were pretty much no additional performance gains, or ill-effects for that matter to be had. "
So when the allowed voltage is still being limited in the same exact way, and at the same exact point with and without this exploit, it is pretty apparent that there is no real risk in outright killing the CPU involved here. Increasing the power limit WILL increase the voltage per the normal voltage-frequency curve (higher frequency == higher voltage), but only within the bounds set by the silicon fitness monitoring feature (FIT).
This exploit is nothing but a clandestine way to increase the power limit.
In order to produce comparable results, please follow the workflow specified in the OP.
- Use Cinebench R20 NT (multithreaded only).
- Part load and idle reading are completely irrelevant and should be ignored (no need to state them here either, for clarity sake).
- Please test at stock settings. These include voltage offsets (which can and will cause clock stretching), load-line adjustments and manual overclocking. The main idea here is to allow the CPU to stay in control of as many of its parameters as possible.
Originally this feature was intended only for the 3rd gen. Ryzen CPUs and I've personally only tested it on them. Martin (Mumak) tested this briefly on 2nd gen. Ryzen. and at least in that case it appeared to work fine. However, based on the couple of user reports here the 2nd and especially first gen. Ryzen deviation reporting looks somewhat suspicious. Because of that we'll need to look more into that, to ensure that the reporting also on 1st and 2nd gen. parts is working correctly.
Tom's Hardware cannot recognize a**holes from elbows so anything written there is popcorn fart, not just this article. CPU is not printer toner who will run out if you print bold letters instead Arial Narrow Italic.
I will re-post my score, 97,9-99,7 during Cinebench R20. But my R5 3600 boosts even to 4400MHz.
I will re-post my score, 97,9-99,7 during Cinebench R20. But my R5 3600 boosts even to 4400MHz.
hello
Trying to figure out whats going on with my system,
So mine goes way over 100% to 700%.
In games I have great temps. Hovering mid 40's.
That means however that my Cpu will end his life faster by staying at manual CCX OC ?
Thank you.
3900x
B450 Tomahawk Max
2080ti
2x8 Gskill 4266@ 3800cl16
IF 1900
Trying to figure out whats going on with my system,
So mine goes way over 100% to 700%.
In games I have great temps. Hovering mid 40's.
That means however that my Cpu will end his life faster by staying at manual CCX OC ?
Thank you.
3900x
B450 Tomahawk Max
2080ti
2x8 Gskill 4266@ 3800cl16
IF 1900
Attachments
So everyone here is talking about all of these newer motherboards. The X470 and X570 boards and mostly the Ryzen 3000 / Zen2 processors. I went to look and I tried the latest v6.27-4185 version of HWINFO and you aren't reporting the "CPU Power Reporting Deviation" for older systems? Why not? I would like to know this info too. I'm using a AsRock X370 Taichi board with a Ryzen 5 2600 processor. Manual overclock to 4.2 Ghz all-Core @ 1.35v. I put it under load at 100% with WPRIME 1024m and there's nothing listed in hwinfo. Also HWINFO is reporting significantly less power draw than other programs. AIDA64 shows my processor in this situation drawing 150 watts for CPU Package Power, yet HWINFO here only shows 94W max power draw. So which program do I trust to be accurate? See screenshot below, no "CPU Power Reporting Deviation" listed:
So everyone here is talking about all of these newer motherboards. The X470 and X570 boards and mostly the Ryzen 3000 / Zen2 processors. I went to look and I tried the latest v6.27-4185 version of HWINFO and you aren't reporting the "CPU Power Reporting Deviation" for older systems? Why not? I would like to know this info too. I'm using a AsRock X370 Taichi board with a Ryzen 5 2600 processor. Manual overclock to 4.2 Ghz all-Core @ 1.35v. I put it under load at 100% with WPRIME 1024m and there's nothing listed in hwinfo. Also HWINFO is reporting significantly less power draw than other programs. AIDA64 shows my processor in this situation drawing 150 watts for CPU Package Power, yet HWINFO here only shows 94W max power draw. So which program do I trust to be accurate? See screenshot below, no "CPU Power Reporting Deviation" listed:
There are 2 reasons:
1. Your system doesn't seem to report all parameters required for this metric to be available. A BIOS update might help and/or resetting to stock settings.
2. There seems to be an issue on some 1st/2nd generation Zen families with this metric as posted here: https://www.hwinfo.com/forum/threads/hwinfo-v6-27-4185-beta-released.6455/post-25091
I'm already on the latest bios that Asrock has released publicly for this board, which uses AMD AGESA Combo-AM4 1.0.0.4 Patch B. I'll try stock settings here in a little bit if I get time today.
Whiskeyomega
Member
You might want to send them an email. Asrock are known not to release bios' for some reason. Im on 2.03 on my Aqua and have been since Febuary and its never been released despite being their best bios for memory oc at least
The Aqua is a newer X570 board. This is an older X370 board. I've had a look around the internet and the majority of X370 boards from almost all manufacturers are all stopped at about 1.0.0.4 Patch B it seems. So this is not just an Asrock issue. These boards won't be getting support for the 4000 / Zen3 series of processors anyway so it's not surprising they aren't getting future bios updates. I will email them anyway and ask, thanks for the suggestion.
EDIT: All of Gigabyte's X370 boards are on 1.0.0.4 B. The same goes for all AsRock X370 boards, and Asus's X370 boards. Poor MSI X370 owners are stuck on 1.0.0.3 ABBA and don't even get the 1.0.0.4 B update. It seems all around that no one on X370 is getting anything newer than 1.0.0.4 B at the moment.
Last edited:
alucardyos
New Member
I just read about this and went to test a new budget pc I recently build.
MOBO: Gigabyte B450M S2H (BIOS F50) [I just saw that there's a new F51e version, but only affects RAM compatibility]
CPU: Ryzen 5 2600
Cooler: Aerocool VERKHO 2 (really low end, but keeps temps under 80ºC on stress)
RAM: Kingston HyperX 3000MHz (2x4GB)
PSU: Mars Gaming MP700
I normally use Ryzen Master to OC, but I left the control mode on Auto, but the RAM is still overclocked to 3200MHz.
The lowest deviation I got was 126% when running Cinebench R15 and R20.
I get about 147% with Prime95 full load.
What does this mean exactly? Will I be able to overclock it without any risk? I'm kind of new to the OC world also...
MOBO: Gigabyte B450M S2H (BIOS F50) [I just saw that there's a new F51e version, but only affects RAM compatibility]
CPU: Ryzen 5 2600
Cooler: Aerocool VERKHO 2 (really low end, but keeps temps under 80ºC on stress)
RAM: Kingston HyperX 3000MHz (2x4GB)
PSU: Mars Gaming MP700
I normally use Ryzen Master to OC, but I left the control mode on Auto, but the RAM is still overclocked to 3200MHz.
The lowest deviation I got was 126% when running Cinebench R15 and R20.
I get about 147% with Prime95 full load.
What does this mean exactly? Will I be able to overclock it without any risk? I'm kind of new to the OC world also...
Last edited:
Whiskeyomega
Member
So are X470/570. they're all on 1.0.0.4B that is the most upto date AGESA code. Not the latest bios.
Similar threads
