TR Forums

Figured I'd post some results from recent experience undervolting an FX-8320. If there's interest, I'll go through the whole process for people, but right now I figure results are going to be what most people wanted to see.

So, here's some data:





------------------------------------------------------------------------------

The next bit are power numbers. I measured using a Kill-A-Watt at the wall before my UPS. That includes my monitors, computer box, speakers, and such. It's not ideal, but my UPS didn't like providing power if I didn't have the box in the main spot. Deltas are what matter, so that'll do.




Not sure that I'll leave it this way, but it is interesting, to say the least. It is especially interesting to consider how it would affect people's opinions of AMD chips if AMD was more aggressive on voltage and lowered their TDPs.

I did end up having a couple of freezes (not BSODs, just freezes) after I set these to run in normal operation, which occurred in under 15 minutes in every case. This, despite having proven each P-state for over 12 hours on Prime95 (some were well over 12 hours, but 12 was the minimum for some transition states). By upping the NB frequency from the default 2200 MHz to 2400 MHz (as I'd had it before I did any tweaking, for memory optimization), I seem to have resolved that issue.

Good stuff. I almost always under volt AMD processors since AMD almost seems paranoid about it. I guess it's one way for them to save costs on verification? In the Athlon/Phenom II days it was trivial for me to bring the stock voltage from 1.425 (I think) to 1.325.

I'm not sure how much Intel would benefit from under volting since their power figures have been superior for quite awhile now, though. Might be a fun review for TR to do.

I was also able to undervolt my 8350 as well. Can't remember the exact stock voltages it was running at during idle and load (I used CPUID), but I seem to have lopped 8% to 10% off the stock voltages in both idle and load. I was able to go down one notch lower in the BIOS but Any Video Converter would hang up when all cores are loaded, so I had to raise the BIOS setting by a notch. I had no problems with other apps though. AVC seems to be a little fussy.

If the mobo supports it, some combination of undervolt and load line calibration might offer the best combination of power/heat/noise and stability.

I should probably play around with this stuff, since I have an FX-8320 that is left powered on 24x7.

Interesting. How do you underclock?

Considering building an FX-8350 rig since I had a free one given to me (vs. buying the gear to build a G3258-based system). One of my main concerns was power draw on the AMD system, and I'd been curious how things would go if I gave undervolting a try. So, awesome data - thanks for the post!

Not bad for power savings... particularly at 3.5ghz and up.

Here is what I got:

All core loaded using Prime95

1.112v at 3.5ghz
1.150v at 3.7ghz
1.225v at 4.0ghz
1.265v at 4.2ghz

Unrelated but, 1.27v is also the voltage I had to use to make a G3258 stable at 4.2ghz.

edit: Deleted first post ranting about wanting to building an FX-8350 system, but no MB being available.

@Sschaem:

As i see it, AMD made 2 mistakes:

1. They didn't continue the FX line through to Steamroller, so OEMs knew there and then that it's a dead platform. 990FX had no successor either, making it even more obvious that AM3+ is dead. No OEMs wanna touch AM3+ now, so boards are becoming increasingly scarcer, killing any prospects of selling existing FX chips in the channel further.

2. They should've encouraged/required board makers to produce mATX boards. How hard can it be?

But you know what? Where I live there are some stores that will refuse to sell you a CPU (Intel or AMD) unless you buy your board from them too. They say the reason is that their suppliers are making them do it. It's an illegal business practice but nonetheless it may be their way to make sure people don't just buy the CPUs and leave the board makers with excess inventory. 1:1 I guess. Still, when my previous AM3+ board gave up the ghost back in January I had a lot of trouble looking for a decent replacement so I hope this Gigabyte board I'm using now won't quit for many years as I plan to hold on to my FX until AMD's next x86 core comes out and its platform matures enough, assuming it moves things forward, of course.

Cant spend to much time. But I plugged my watt monitoring tool on the FX workstation.

FX-8320, 1.265v 4.2ghz, 16GB of ram, SSD + GTX 680

System Idle (desktop / Cinebench R15 loaded) 90w (1.4ghz - .888v)cpuZ
Peak power usage running Cinebench R15 CPU test: 226w (4.2ghz - 1.28v)couZ

(Score 670)

Delta idle to full load: 136w

Note: this is at the wall from a 1000w power supply.

edwpang wrote:

Interesting. How do you underclock?

Underclocking is pretty simple, especially with an unlocked chip. It's the same as overclocking, but you drop the multiplier.

Undervolting, on the other hand, is a very different process, and can take a bit more work. I found a couple of resources that helped me out. AMD's PSCheck

Here: https://www.youtube.com/watch?v=cK3HTtztg9g

And here: http://www.overclock.net/t/1499562/undervolting-editing-p-states-for-piledriver-using-amd-mrstweaker-mini-guide

The youtube video does a good job of giving a basic run down of how to use PS Check, and can give you a way of doing things.

The bit from the "Overclock.net" forums actually gives a more elegant solution for making it "permanent" once you've done the testing necessary to call it good.

sschaem wrote:

Here is what I got:

All core loaded using Prime95

1.112v at 3.5ghz
1.150v at 3.7ghz
1.225v at 4.0ghz
1.265v at 4.2ghz

Nice numbers. I was only able to hold voltages that low for about 15 minutes on Prime. And thanks for the second post. Answered some of my other questions.

ronch wrote:

@Sschaem:

As i see it, AMD made 2 mistakes:

1. They didn't continue the FX line through to Steamroller, so OEMs knew there and then that it's a dead platform. 990FX had no successor either, making it even more obvious that AM3+ is dead. No OEMs wanna touch AM3+ now, so boards are becoming increasingly scarcer, killing any prospects of selling existing FX chips in the channel further.

2. They should've encouraged/required board makers to produce mATX boards. How hard can it be?

But you know what? Where I live there are some stores that will refuse to sell you a CPU (Intel or AMD) unless you buy your board from them too. They say the reason is that their suppliers are making them do it. It's an illegal business practice but nonetheless it may be their way to make sure people don't just buy the CPUs and leave the board makers with excess inventory. 1:1 I guess. Still, when my previous AM3+ board gave up the ghost back in January I had a lot of trouble looking for a decent replacement so I hope this Gigabyte board I'm using now won't quit for many years as I plan to hold on to my FX until AMD's next x86 core comes out and its platform matures enough, assuming it moves things forward, of course.

Sry, I delete my post that you reply to...

But AMD certainly shot itself in the leg on this one.

First they should have binned the FX serie early on with lower voltage, as the bulldozer chip where not in demand for servers.
They could have introduced a 4ghz FX-8350T a long time ago with a 95w TDP, and if they had a decent mATX motherboard it would have help their image.

ronch wrote:

They should've encouraged/required board makers to produce mATX boards. How hard can it be?

The problem is availability of IGPs that don't suck. AMD has little incentive to produce an updated northbridge-based GPU, since they are pushing their APUs for that segment of the market now. Because of this, mATX Socket AM3+ boards tend to use the horribly dated 760G/SB710 chipset, which is fully a half decade old. This relegates mATX AM3+ boards to the bargain basement segment of the market. There are actually a handful of them out there, from the usual suspects (Asus, MSI, ASRock, etc.); but if you think about it, they are strangely unbalanced products.

FightingScallion wrote:

edwpang wrote:

Interesting. How do you underclock?

Underclocking is pretty simple, especially with an unlocked chip. It's the same as overclocking, but you drop the multiplier.

Undervolting, on the other hand, is a very different process, and can take a bit more work. I found a couple of resources that helped me out. AMD's PSCheck

...
The youtube video does a good job of giving a basic run down of how to use PS Check, and can give you a way of doing things.

The bit from the "Overclock.net" forums actually gives a more elegant solution for making it "permanent" once you've done the testing necessary to call it good.

[/quote]


Yeah, I knew about PSCheck, but it's a bit tedious to keep it permanent, ie. after rebooting and standy. Actually standby is more difficult to handle. I'll check the the guide.

edwpang wrote:

Yeah, I knew about PSCheck, but it's a bit tedious to keep it permanent, ie. after rebooting and standy. Actually standby is more difficult to handle. I'll check the the guide.

Hmm...I see your point. I just popped open Overdrive after a "hibernate/wake" cycle and it seems to have reverted. I'm going to check and see if there's a good way to fix that. Will report back with more results as I resolve this. I have a related problem (keyboard macro causes "hibernate" rather than "sleep", despite a sleep command being given). I'll see if it's just hibernate/standby that causes the reversion or if sleep also does that, after I get the hibernate to work.

Okay, I found a solution.

When creating the scheduled task, we're going to add an additional trigger. So, after you've followed the steps in the other guide, do this (I'm in Win8.1, but I expect it will work in Vista/7 similarly):

1) Double click on the task (whatever you named it).
2) Choose the "Triggers" tab
3) Click "New"
4) On the "Begin the task:" dropdown, choose "On an event"
5) For settings, the "Basic" radio should be selected
6) On the "Log" dropdown, go to "System"
7) For the "Source" dropdown, choose "Power-Troubleshooter"
8) For Event ID, enter "1" (no quotes) - (1 is the event ID for resuming from a low-power state)

This seems to work with both hibernate and sleep.

FightingScallion wrote:

Okay, I found a solution.

When creating the scheduled task, we're going to add an additional trigger. So, after you've followed the steps in the other guide, do this (I'm in Win8.1, but I expect it will work in Vista/7 similarly):

1) Double click on the task (whatever you named it).
2) Choose the "Triggers" tab
3) Click "New"
4) On the "Begin the task:" dropdown, choose "On an event"
5) For settings, the "Basic" radio should be selected
6) On the "Log" dropdown, go to "System"
7) For the "Source" dropdown, choose "Power-Troubleshooter"
For Event ID, enter "1" (no quotes) - (1 is the event ID for resuming from a low-power state)

This seems to work with both hibernate and sleep.

Actually, I have done the same procedure too. But sometimes it's not reliable. After wake up, I sometimes see the default voltage comes back. I only wish it's included in BIOS setting. It can be an AMD killer feature for power users.

So far undervolting's worked a treat on both of my AM3+ systems. The ASRock Extreme4's UEFI made it easy - core voltage at stock speeds for my FX-8320 is down to 1.2125v from a stock value of 1.35v. A TDP calculator online suggested I knocked effective TDP down to 100 watts, and the difference has been enjoyably apparent. The FX-4100 I'm using for a server seems to have come from a good batch... My initial guess of 1.2v (down from a stock value of 1.375v) has been stable and much quieter. Idle temps with the stock cooler went from 30-35C to a degree or two above room temperature. I'm pleased.