FREE performance for Ryzen!

[warmongerp]

3 hours ago, ageekhere said:

Is this like ASUS AI Overclocking? I wonder how it compares. 

Motherboard manufactures auto OC features are notoriously horrendous, (Hi MSI and Asus)

 

I will bet my left pinky that this program is leagues above them.

[williamcll]

Looking at my broken 2700X and now to my 3600, is there any way I can know how well my motherboard can do overclock with its 6 VRMs?

[MageTank]

8 hours ago, ShrimpBrime said:

There's no such thing as "Free" performance. 

IE: More clocks Same voltage = hotter chip. period. 

 

On a technical note, since this is a "tech" forum.....

Transistors stop working when under-volted. (A very small percentage)

When enough transistors stop working, the Cpu comes to a halt state and shuts down.

Many of you may have experienced this. So what's the solution? More v-core.

 

Free overclocking? Free performance? 

I think not.

 

Total click bait title for a tech forum. Total lie.

 

 

 

 

It's not a lie, nor is your explanation entirely accurate. I am sure you understand that modern processors often have power/thermal limitations in-place. Undervolting can help provide additional headroom in these situations, which is why it's extremely popular on mobile devices that throttle their hardware based on those aforementioned limitations. In desktop hardware, we use the same principles in undervolting to achieve lower thermals when voltage scaling is nearly non-existent or detrimental to our overclocking.This reduction in thermals often gives us slightly higher overhead when overclocking, despite using less voltage. I doubt that I need to explain the concept of electrical resistance to you or the benefits undervolting has on mitigating electromigration (assuming current isn't increasing as voltage is lowered).

 

You can get "free" performance by undervolting, if you don't consider your time valuable when stress testing for stability. The lowered voltage will lower your thermals, and if you were limited by thermals rather than voltage, you can potentially get an increase in clock speed. Since we all know Ryzen doesn't OC all that well, simply undervolting would likely allow it to maintain a higher turbo for a longer duration, which is likely where most of these gains will come from.

 

If you'd like, we can develop a testing methodology to compare against and easily prove/disprove the impact undervolting has when overclocking. We can use CPU's, GPU's or memory as the principle is pretty universal across most hardware, though memory would lean more towards IMC voltages instead of VDIMM.

[Guest]

1 hour ago, MageTank said:

It's not a lie, nor is your explanation entirely accurate. I am sure you understand that modern processors often have power/thermal limitations in-place. Undervolting can help provide additional headroom in these situations, which is why it's extremely popular on mobile devices that throttle their hardware based on those aforementioned limitations. In desktop hardware, we use the same principles in undervolting to achieve lower thermals when voltage scaling is nearly non-existent or detrimental to our overclocking.This reduction in thermals often gives us slightly higher overhead when overclocking, despite using less voltage. I doubt that I need to explain the concept of electrical resistance to you or the benefits undervolting has on mitigating electromigration (assuming current isn't increasing as voltage is lowered).

 

You can get "free" performance by undervolting, if you don't consider your time valuable when stress testing for stability. The lowered voltage will lower your thermals, and if you were limited by thermals rather than voltage, you can potentially get an increase in clock speed. Since we all know Ryzen doesn't OC all that well, simply undervolting would likely allow it to maintain a higher turbo for a longer duration, which is likely where most of these gains will come from.

 

If you'd like, we can develop a testing methodology to compare against and easily prove/disprove the impact undervolting has when overclocking. We can use CPU's, GPU's or memory as the principle is pretty universal across most hardware, though memory would lean more towards IMC voltages instead of VDIMM.

Right, so many points to touch on.

 

We could start by lower the thermals FIRST which reduces leakage and then as a result a lower vcore and be used.

 

There's no need to test a theory on this while we are talking about minor differences in an average system and being temp dependant and than of course with a variety of boards just straight power limited.

 

Either way, not totally accurate testing will come from this, not terribly worried about it honestly.

[MageTank]

1 hour ago, ShrimpBrime said:

Right, so many points to touch on.

 

We could start by lower the thermals FIRST which reduces leakage and then as a result a lower vcore and be used.

 

There's no need to test a theory on this while we are talking about minor differences in an average system and being temp dependant and than of course with a variety of boards just straight power limited.

 

Either way, not totally accurate testing will come from this, not terribly worried about it honestly.

How exactly would you lower thermals first? A better cooling solution? Let's assume you threw the best ambient cooling solution you can at it and thermals are still high, then what? When using ambient cooling methods, there is no substitution for the thermal reduction brought about by undervolting. When we talk thermals, voltages impact on thermals scales quadratically compared to clock speeds alone. You'd see a dramatically lower drop in temperature by undervolting than by what you would from underclocking, that is extremely easy to test and prove. Just like you'd see far higher thermals simply by raising voltages while keeping clock speeds the same, than what you would by increasing clocks and keeping voltage the same. 

 

My point is simple and easy to test. If you undervolt correctly, you can see a reduction in temperature without compromising stability. This reduction in temperature can lead to more stable clock speeds and as a result, a slight improvement in performance. If we are talking mobile platforms, this improvement is far more pronounced due to the aforementioned power/thermal limitations from my previous post.

 

We use this exact same principle when overclocking modern GPU's. Voltage curve overclocking is far superior to the old methods of brute forcing voltage and hoping it doesn't crash. I've seen far greater returns on my GPU overclocking using more tame voltage values, finesse over brute force. We see this with memory overclocking when talking IMC voltages. Why does the concept all of a sudden not apply to CPU overclocking? 

 

My offer still stands if you want to develop a methodology for me to test against, or let me prove my point with my own methodology. Either way, it's not impossible to come up with a decent test that can be reproduced.

[WereCatf]

11 hours ago, ShrimpBrime said:

IE: More clocks Same voltage = hotter chip. period.

Clocks don't increase temperature, power (ie. voltage * current) does. If the power-consumption stays the same, the temperature stays the same.

[Guest]

13 minutes ago, MageTank said:

How exactly would you lower thermals first? A better cooling solution? Let's assume you threw the best ambient cooling solution you can at it and thermals are still high, then what? When using ambient cooling methods, there is no substitution for the thermal reduction brought about by undervolting. When we talk thermals, voltages impact on thermals scales quadratically compared to clock speeds alone. You'd see a dramatically lower drop in temperature by undervolting than by what you would from underclocking, that is extremely easy to test and prove. Just like you'd see far higher thermals simply by raising voltages while keeping clock speeds the same, than what you would by increasing clocks and keeping voltage the same. 

 

My point is simple and easy to test. If you undervolt correctly, you can see a reduction in temperature without compromising stability. This reduction in temperature can lead to more stable clock speeds and as a result, a slight improvement in performance. If we are talking mobile platforms, this improvement is far more pronounced due to the aforementioned power/thermal limitations from my previous post.

 

We use this exact same principle when overclocking modern GPU's. Voltage curve overclocking is far superior to the old methods of brute forcing voltage and hoping it doesn't crash. I've seen far greater returns on my GPU overclocking using more tame voltage values, finesse over brute force. We see this with memory overclocking when talking IMC voltages. Why does the concept all of a sudden not apply to CPU overclocking? 

 

My offer still stands if you want to develop a methodology for me to test against, or let me prove my point with my own methodology. Either way, it's not impossible to come up with a decent test that can be reproduced.

You can test anything you desire and you dont have to "prove" anything to me really. I know who you are and your capabilities. 

 

That said, why would you think I havent tested?

Granted not on a 3990x but on Zen and Zen+... never the less it's the point of under-volting.

 

And under what such amount? 0.01v? 0.025v?

Will all this take in account "average systems" most people leave LLC on auto?

 

Again, you can prove something on your system, make an average an perhaps some percentage of people may or may not have the same results.

 

Now any testing I've done wouldn't matter cause your interested in your own testing that will ultimately prove something that you system does but mine does not. So you testing will apply to those running the exact same hardware with the exact same cooling ambient temps and so forth.

 

15 minutes ago, WereCatf said:

Clocks don't increase temperature, power (ie. voltage * current) does. If the power-consumption stays the same, the temperature stays the same.

I've never once had this experience. Not on any cpu on any platform. The increase, as minimal as it is, increase thermal output. Never bothered to measure consumption actually.

 

But 3.7ghz to 4ghz use the same vcore, at 4ghz thermals will be higher. That much I'm aware of with my systems... besides the pstate voltage wont handle that 300mhz OC, but if it did....

[WereCatf]

4 minutes ago, ShrimpBrime said:

I've never once had this experience. Not on any cpu on any platform. The increase, as minimal as it is, increase thermal output. Never bothered to measure consumption actually.

Heat is energy, ergo if there is an increase in heat-output, there must also be an increase in power-consumption. Energy doesn't magically appear out of nowhere.

[MageTank]

7 minutes ago, ShrimpBrime said:

You can test anything you desire and you dont have to "prove" anything to me really. I know who you are and your capabilities. 

 

That said, why would you think I havent tested?

Granted not on a 3990x but on Zen and Zen+... never the less it's the point of under-volting.

 

And under what such amount? 0.01v? 0.025v?

Will all this take in account "average systems" most people leave LLC on auto?

 

Again, you can prove something on your system, make an average an perhaps some percentage of people may or may not have the same results.

 

Now any testing I've done wouldn't matter cause your interested in your own testing that will ultimately prove something that you system does but mine does not. So you testing will apply to those running the exact same hardware with the exact same cooling ambient temps and so forth.

 

I've never once had this experience. Not on any cpu on any platform. The increase, as minimal as it is, increase thermal output. Never bothered to measure consumption actually.

 

But 3.7ghz to 4ghz use the same vcore, at 4ghz thermals will be higher. That much I'm aware of with my systems... besides the pstate voltage wont handle that 300mhz OC, but if it did....

Without going into much detail about my employment, I have access to an absurd amount of hardware ranging from "soon to be released" all the way back to Nehalem and Phenom II. I am pretty confident I can put together a comprehensive testing methodology that factors in chip to chip variance, but that wouldn't really be the point. I don't think we are really arguing against the degree in which undervolting helps, but rather the simple point of "does it help?". At least, that is what I am trying to convey and would like to demonstrate if need be. Again, you can argue that it's not "free" performance if one has to devote a decent amount of time to stress test and assure stability as time is money, but I disagree with the notion that undervolting doesn't offer an increase in performance, especially on modern hardware with boost tables tailored to both power and thermals.

 

As for the question regarding LLC, I don't believe it matters. LLC overshooting or undershooting for others wouldn't really make a difference to the grand point that is being made, which is: lower voltage values and the reduced thermals brought about by the lower voltage can aid in clock stability and thus an improvement in performance. Regardless of board or LLC value, the same principle applies to everyone that reduces their voltage relative to what it is currently residing at. The only real difference per person is how low of a voltage your hardware configuration will tolerate, and the level of instability you are willing to settle with. Gaming stable? AVX stable? This will be subjective in the end.

 

Hopefully that makes sense of what I am trying to say, or what I am offering. 

[Guest]

15 minutes ago, MageTank said:

Without going into much detail about my employment, I have access to an absurd amount of hardware ranging from "soon to be released" all the way back to Nehalem and Phenom II. I am pretty confident I can put together a comprehensive testing methodology that factors in chip to chip variance, but that wouldn't really be the point. I don't think we are really arguing against the degree in which undervolting helps, but rather the simple point of "does it help?". At least, that is what I am trying to convey and would like to demonstrate if need be. Again, you can argue that it's not "free" performance if one has to devote a decent amount of time to stress test and assure stability as time is money, but I disagree with the notion that undervolting doesn't offer an increase in performance, especially on modern hardware with boost tables tailored to both power and thermals.

 

As for the question regarding LLC, I don't believe it matters. LLC overshooting or undershooting for others wouldn't really make a difference to the grand point that is being made, which is: lower voltage values and the reduced thermals brought about by the lower voltage can aid in clock stability and thus an improvement in performance. Regardless of board or LLC value, the same principle applies to everyone that reduces their voltage relative to what it is currently residing at. The only real difference per person is how low of a voltage your hardware configuration will tolerate, and the level of instability you are willing to settle with. Gaming stable? AVX stable? This will be subjective in the end.

 

Hopefully that makes sense of what I am trying to say, or what I am offering. 

I think such information can be really useful.

 

Just not at this forum. Where most of these people dont even think to under volt unless they have a thermal issue. At least that what it seems from all the reading and assistance.

 

Recently, a young man clocked down to 3.6ghz while a slight under-volt did not help thermals enough. The solution was simple, take care of the cooling issue. Then perhaps shoot for an under volt.

 

By all means, I'd be one super interested in some experiments for a good read and additional general knowledge. 

 

 

 

Here's a screen shot of undervolting.

1.188v 4ghz zen+, I'd expect any cpu with cold thermals to leak much less. 

 

Example only to convey my point and nothing more. This is not a challenge screen shot by any means.

You can view the thread at our team site, but must create an account to view all the pictures.

 

 

 

 

[Guest]

1 hour ago, WereCatf said:

Heat is energy, ergo if there is an increase in heat-output, there must also be an increase in power-consumption. Energy doesn't magically appear out of nowhere.

Respectfully a transistor creates heat from passing current and I/O operation.

The faster the operation, the more heat. Again small amount. Nothing night and day. 

It might actually consume more AMP at the same voltage as a result. As I said never measured it.

[Results45]

You guys should really test this on Microsoft Flight Sim 2020........

 

[Guest]

53 minutes ago, Results45 said:

You guys should really test this on Microsoft Flight Sim 2020........

 

See that's an issue for testing. 

The game utilizes a few cores at best and very GPU bound + memory performance, IF and so forth all make some sort of performance difference.

 

We are talking about "FREE" performance gains while under-volting a cpu down to just before instability and then overclocking the cpu.

 

It's just really not worth using a game to accurately describe an under-volt + higher over clock ( or higher boosting "duration"??) While measuring this free performance deal. 

 

WPrime, Cinebench, full load Prime95 or OCCT and so forth to directly reflect the free performance. (CPU only + therms) (examples)

 

However, I dont think (yet another) utility to accomplish this is accurate to the word "free" while this program will use system resources and drive space. All of what is being mentioned can be done in the bios. That would be closer to "free" performance than anything.

 

All just the angle I see it from basically. 

[BTGbullseye]

23 hours ago, ShrimpBrime said:

However, I dont think (yet another) utility to accomplish this is accurate to the word "free" while this program will use system resources and drive space.

It's a set and go, not a background task from what I can see.

[Guest]

8 minutes ago, BTGbullseye said:

It's a set and go, not a background task from what I can see.

Cool cool. Hopefully this brings some real overclocking head room though.

 

Damn AMD got these things tweaked pretty well from the box as it is. 

 

Never the less a gain is a gain, and I acknowledge that whole heartedly. 

[BTGbullseye]

5 minutes ago, ShrimpBrime said:

Cool cool. Hopefully this brings some real overclocking head room though.

Agreed.

6 minutes ago, ShrimpBrime said:

Damn AMD got these things tweaked pretty well from the box as it is. 

Which is extremely good for the general consumer, even if it is disappointing for overclockers.

 

[Guest]

1 hour ago, BTGbullseye said:

Agreed.

Which is extremely good for the general consumer, even if it is disappointing for overclockers.

 

Well it would be nice to have at least 200mhz to work with lol. Even for the gamers and OCers.

 

On my daily ( HTPC) I use QuickCPU. Ever heard of that one? I use it mainly for core parking and power savings.

[BTGbullseye]

1 hour ago, ShrimpBrime said:

On my daily ( HTPC) I use QuickCPU. Ever heard of that one? I use it mainly for core parking and power savings.

Never heard of it before... I'll give it a gander.

[BTGbullseye]

2 hours ago, ShrimpBrime said:

On my daily ( HTPC) I use QuickCPU. Ever heard of that one? I use it mainly for core parking and power savings.

Interesting, but it looks like my standard Windows install customizations do pretty much all the same things this program does. lol

[Guest]

Just now, BTGbullseye said:

Interesting, but it looks like my standard Windows install customizations do pretty much all the same things this program does. lol

Right, but a 3rd party utility for those that dont know how to customize an OS. I actually like the app on my Intel setups better honestly.

 

 

 

[BTGbullseye]

10 minutes ago, ShrimpBrime said:

Right, but a 3rd party utility for those that dont know how to customize an OS. I actually like the app on my Intel setups better honestly.

It definitely would be useful to speed up the settings adjustments, and make it a lot easier for non-power users.

[Guest]

13 minutes ago, BTGbullseye said:

It definitely would be useful to speed up the settings adjustments, and make it a lot easier for non-power users.

I stumbled across this app while looking for some legacy software. Figure give a shot, what the heck. The appealing part for me was the core parking control over most anything. Saves some time digging in regedit to turn off the setting in the OS. 

 

 

[BTGbullseye]

24 minutes ago, ShrimpBrime said:

I stumbled across this app while looking for some legacy software. Figure give a shot, what the heck. The appealing part for me was the core parking control over most anything. Saves some time digging in regedit to turn off the setting in the OS. 

Yeah, it's not bad.

 

It's not quite as useful on AMD systems as it is on Intel though.

[Guest]

4 minutes ago, BTGbullseye said:

Yeah, it's not bad.

 

It's not quite as useful on AMD systems as it is on Intel though.

Heck I didnt know LTT had a hwbot team. Lol. 

Are you getting into the competitive benchmarking more?

[BTGbullseye]

16 hours ago, ShrimpBrime said:

Heck I didnt know LTT had a hwbot team. Lol. 

Are you getting into the competitive benchmarking more?

Not really. I was just upset that there were no 5700XT benchmarks in GPUPI at all... I'm still the only one apparently.