Question Proof that undervolting affects CPU's performance!

[Veloci]

I'm currently running an old Core 2 Duo E8400 and here is the result with and without undervolt:

It's not like I'm complaining, no sir. I'm very VERY happy with the results, actually. Before, the CPU was reaching 83c @ 1.20V. Now it remains at 65c @ 1.05V, is 100% stable and consumes 20W less power (65W vs. 45W).

I'm just a bit stumped by the results because people usually say that undervolting has no diverse effects on the performance, whatsoever.

This proves that they're... WRONG! Undervolting DOES affect performance, although the difference is quite minimal.

Mere 2.7% in my case!

 

[rgd1101]

people usually say that undervolting has no diverse effects on the performance

who said? anyone know OC would know .

 

[InvalidError]

How many times did you re-run the benchmark? Unless you did 5+ runs, tossed clear outliers out and averaged the rest, 5% is within the typical tun-to-run variance and is generally considered margin-of-error stuff.

 

[dudio]

Ofcourse undervolting affects performance. Truth is, undervolting is usually a poor fix for temperature issues. You shouldn't have to, and should consider other things to remedy higher than normal temps (ie: better cooling solutions).

 

[rigg42]

Ofcourse undervolting affects performance. Truth is, undervolting is usually a poor fix for temperature issues. You shouldn't have to, and should consider other things to remedy higher than normal temps (ie: better cooling solutions).

Tell that to people running ryzen X SKU's with PBO.

 

[TJ Hooker]

Try running the benchmark a few times at the same voltage and compare the results and you'll see why the above results don't actually prove anything.

 

[jsmithepa]

If OC and up performance involves +voltage.
-voltage, logically should do the opposite, is just pure logic.

by how much and your definition of adverse, well we can argue all day. Agree with ^him, u gotta do multiple readings and take the average.

 

[InvalidError]

If OC and up performance involves +voltage.
-voltage, logically should do the opposite, is just pure logic.

Lower voltages lowers leakage, lowers current, lowers power, lowers heat, lowers noise, etc. and on some lucky chips, this allows slightly higher clocks.

Every chip is different. Although more volts is USUALLY better, it isn't always better.

 

[Veloci]

How many times did you re-run the benchmark? Unless you did 5+ runs, tossed clear outliers out and averaged the rest, 5% is within the typical tun-to-run variance and is generally considered margin-of-error stuff.

Well, I ran the benchmark a total of 6 times, 3 times with the undervolt and 3 times at stock voltages; back to back with no background activity.

With the same voltage, I saw a minor difference of ~0.25% in each run. While with different voltages, I constantly saw a difference of 2.6-2.7%.

It's certainly not a margin of error, I believe!

 

[TJ Hooker]

With the same voltage, I saw a minor difference of ~0.25% in each run. While with different voltages, I constantly saw a difference of 2.6-2.7%.

Where'd you get the 2.6% from? In your two screenshots, the undervolt performs 0.5% lower in single core score, and about 0.9% worse in quad/multi core.

 

[Veloci]

Ofcourse undervolting affects performance. Truth is, undervolting is usually a poor fix for temperature issues. You shouldn't have to, and should consider other things to remedy higher than normal temps (ie: better cooling solutions).

Oh, please...

First of all, I live in a HOT climate (45c right now) and the room where I keep my PC doesn't have any air conditioning. My CPU idles at 52c (1.9Ghz @ 1.0V), even though I've just replaced the thermal compound and using a 95W TDP cooler with a 120mm fan!

Second of all, undervolting IS NOT a poor fix! Look this way; I'd to sacrifice <3% performance in exchange of 17c cooler core temperature and ~20W less power consumption! That's a great bargain, in all honesty!

Lastly, if my CPU had an unlocked multiplier, I'd have both overclocked AND undervolted it. This CPU seems to have a really good silicon so I'm pretty sure it'd have broken the 4Ghz barrier at stock voltages and with -100mV undervolt, ~3.5Ghz would be entirely possible. But too bad, I'm stuck with 3.0Ghz frequency...

 

[Veloci]

Where'd you get the 2.6% from? In your two screenshots, the undervolt performs 0.5% lower in single core score, and about 0.9% worse in quad/multi core.

I was quoting the overall benchmark score seen in the snapshot, mate.

At stock it's 26%, while it drops to 25.4% with the undervolt.

 

[TJ Hooker]

I was quoting the overall benchmark score seen in the snapshot, mate.

At stock it's 26%, while it drops to 25.4% with the undervolt.

The fact that the difference in the overall result is somehow more than double the difference of any of the component results might give an indication of how accurate and reliable this benchmark is.

 

[InvalidError]

The fact that the difference in the overall result is somehow more than double the difference of any of the component results might give an indication of how accurate and reliable this benchmark is.

He's comparing rankings in the result database. Being above 25.4% of other runs in the DB vs 26% doesn't mean anything performance-wise. Very weird way to look at it. The raw scores are what should be relevant here and 0.9% better at best is correct.

 

[Veloci]

The fact that the difference in the overall result is somehow more than double the difference of any of the component results might give an indication of how accurate and reliable this benchmark is.

User Benchmark is generally quite accurate in my experience but I can run Cinebench if you want!

 

[TJ Hooker]

Out of curiousity, I did by own user benchmark runs (6700K, turbo disabled):

[single core] [quad core] [multi core]
Run 1 (stock): 124 419 654
Run 2 (-100mV): 123 432 655
Run 3 (-100mV): 124 454 649

As you can see, not only does the undervolt outperform stock in some cases, the difference in scores from one run to the next without changing anything is as high as 5%. Like I've been saying, everything you've seen is essentially just noise.

Edit: In retrospect I probably could have disabled speed step, put windows in high performance power plan, etc. to try to make things as consistent as possible.

 

[Veloci]

Now that's confusing!

I'll run some Cinebenchs and report back. See what it has to say.

 

[geofelt]

45c. = 113f. OUCH!!!
My sympathies.

Of necessity, any cooler except peltier types can only run at some amount over ambient.

Add to that and the older tech in a E8400 is going to run hot.
The processor will monitor temperatures and will slow down or even shut off the chip if it detects a dangerous temperature.
Current chips limits is 100c, I do not know if a E8400 is different.

The best you can do is to use a decent cooler in a well ventilated case.
The stock Intel cooler is normally OK, but you might want to buy a better one.

Actually, apart from academic interest, this is much ado about nothing.
a 2% difference is all but undetectable.

 

[Veloci]

45c. = 113f. OUCH!!!
My sympathies.

Of necessity, any cooler except peltier types can only run at some amount over ambient.

Add to that and the older tech in a E8400 is going to run hot.
The processor will monitor temperatures and will slow down or even shut off the chip if it detects a dangerous temperature.
Current chips limits is 100c, I do not know if a E8400 is different.

The best you can do is to use a decent cooler in a well ventilated case.
The stock Intel cooler is normally OK, but you might want to buy a better one.

Actually, apart from academic interest, this is much ado about nothing.
a 2% difference is all but undetectable.

I need your sympathies, mate! It's agonizing...

You're right. The E8400 has 45nm cores hence it runs quite hot! As for the cooler; I'm using a 95W cooler with a huge 120mm fan whereas my Core 2 Duo is rated at 65W. This is as good as it gets as I'm so not interested in building a custom water block for my $8 CPU!


😊

 

[CountMike]

Tell that to people running ryzen X SKU's with PBO.

Well I'm using just that, It's not actually undervolting but limiting voltage from going higher than actually necessary for stable operation. Several of BIOS versions pushed voltage way too high if left to their own devices. While I recorded stable OC of 4.3GHz at voltage set to 1.4 and Llc pushed it occasionally to max recorded 1.416v BIOS pushed it to way too high 1.55v at same condition. Solution, I just set voltage manually at 1.4v and now it stays under 1.4v with PBO2 fully working, pshing some cores over 4.3 GHz. That's on this MB and newest BIOS, PBO was a bit lower but voltage is now same as before.

 

[USAFRet]

Proof that undervolting affects CPU's performance!

Actually, that's just proof that userbenchmark.com is a poor testing tool.

 

[CountMike]

Actually, that's just proof that userbenchmark.com is a poor testing tool.

Lol, yes, whole point of a benchmark is to compare against a stable and known mark not to hunt a moving target.

 

[Judespace]

Tell that to people running ryzen X SKU's with PBO.

That’s more of a Vega thing than a ryzen thing

 

[geofelt]

Liquid cooling is still essentially air cooling.
A radiator has only 45c. ambient to work with. Even a very good radiator can not do much better than 10c. or so over ambient.

The best benchmark is YOUR game or app.
See if you notice a difference in performance with either higher or lower voltage.
If you see a benefit, that is the way to go.

 

[rigg42]

That’s more of a Vega thing than a ryzen thing

Well I'm using just that, It's not actually undervolting but limiting voltage from going higher than actually necessary for stable operation. Several of BIOS versions pushed voltage way too high if left to their own devices. While I recorded stable OC of 4.3GHz at voltage set to 1.4 and Llc pushed it occasionally to max recorded 1.416v BIOS pushed it to way too high 1.55v at same condition. Solution, I just set voltage manually at 1.4v and now it stays under 1.4v with PBO2 fully working, pshing some cores over 4.3 GHz. That's on this MB and newest BIOS, PBO was a bit lower but voltage is now same as before.

Pretty much anybody running PBO should be running a -voltage offset if they want ideal performance. This is undervollting if you ask me. PBO throws way to much voltage at the CPU by default and causes higher temps which, in turn, directly affects peak boost and boost duration. It is the exact same concept as undervolting a Vega GPU.