News Intel offers new guidance on 13th and 14th Gen CPU instability — but no definitive fix yet

[TerryLaze]

I think the fix will increase efficiency by a lot. Win win situation, no?

No it won't, at the max it will reduce the range of the CPU, cutting off some clocks and/or watts at the top end. That's not the same thing as increasing efficiency.
You can't have increased efficiency at a certain clock if you can't reach that clock (anymore) ...


And there is always an efficiency curve for any CPU so going towards the sweet point of that curve will always "improve efficiency" and I put this in quotes because the efficiency is the whole curve so it's not improving it's just moving closer to the sweet point.

 

[TheHerald]

No it won't, at the max it will reduce the range of the CPU, cutting off some clocks and/or watts at the top end. That's not the same thing as increasing efficiency.
You can't have increased efficiency at a certain clock if you can't reach that clock (anymore) ...


And there is always an efficiency curve for any CPU so going towards the sweet point of that curve will always "improve efficiency" and I put this in quotes because the efficiency is the whole curve so it's not improving it's just moving closer to the sweet point.

Efficiency is work done for watts consumed. Clocks aren't a part of the equation. If with the new settings it gets more work done per watt then it's an increase in efficiency.

After all that the only thing people were complaining about with Intel, so we got a win. Let's all rejoice.

 

[thestryker]

The fake graph AGAIN? I mean come on now man.....how many times do we need to see this anandtech megafake graph, when the reviewer himself said it's problematic?

The RPL numbers are accurate which is all the post was about unless I'm reading it wrong (I'd have snipped the AMD results, but whatever).

The 7950X power figures certainly aren't because they were getting high PPT excursions. It's too bad that piece was never properly updated.

 

[bit_user]

I think the fix will increase efficiency by a lot. Win win situation, no?

Depends on what the user cares about. For a user who doesn't care about efficiency, then any loss of performance would be a losing proposition.

The fake graph AGAIN? I mean come on now man.....how many times do we need to see this anandtech megafake graph, when the reviewer himself said it's problematic?

You are not helping amd by doing this, imagine someone actually buying the 7950x - setting it to 65w and expecting to see 31k.

You're so wrapped up in your mindset that you missed how I was actually just referring to the Intel part of it and nothing to do with the AMD part.

P.S. Thanks @thestryker !

 

[TheHerald]

Depends on what the user cares about. For a user who doesn't care about efficiency, then any loss of performance would be a losing proposition.


You're so wrapped up in your mindset that you missed how I was actually just referring to the Intel part of it and nothing to do with the AMD part.

P.S. Thanks @thestryker !

I'm sorry for jumping the gun but I got tired of seeing this graph everywhere when talking about amds superior efficiency.

 

[TheHerald]

Regarding the first part, well most people complained about the power draw so this will be a fix for the majority.

Personally I really don't care, I have my 12900k locked to 75w,and it's not going to need more cause that 4090 is a major bottleneck, lol.

 

[TerryLaze]

Efficiency is work done for watts consumed. Clocks aren't a part of the equation. If with the new settings it gets more work done per watt then it's an increase in efficiency.

After all that the only thing people were complaining about with Intel, so we got a win. Let's all rejoice.

Clocks are the most important part of it, if the fix reduces clocks and gets better efficiency then you get the same amount of better efficiency by just reducing the clocks without a fix.
Higher clocks need way more power than the more performance they produce.

 

[TheHerald]

Clocks are the most important part of it, if the fix reduces clocks and gets better efficiency then you get the same amount of better efficiency by just reducing the clocks without a fix.
Higher clocks need way more power than the more performance they produce.

No they are not.

 

[TerryLaze]

No they are not.

Efficiency is work done for watts consumed. Clocks aren't a part of the equation.

So you are saying that any CPU will always do the same amount of work per watt, no matter the clocks it runs at...................................................................ok, sure!

 

[TheHerald]

So you are saying that any CPU will always do the same amount of work per watt, no matter the clocks it runs at...................................................................ok, sure!

I'm saying clocks aren't part of the efficiency equation.

 

[TerryLaze]

I'm saying clocks aren't part of the efficiency equation.

But you are saying that "Efficiency is work done for watts consumed. " so your statement has to either be for a specific clock or be valid for all clocks , there are only these two options. Unless you can come up with another option that doesn't involve clocks, but you haven't done that yet.

 

[TheHerald]

But you are saying that "Efficiency is work done for watts consumed. " so your statement has to either be for a specific clock or be valid for all clocks , there are only these two options. Unless you can come up with another option that doesn't involve clocks, but you haven't done that yet.

Clocks are part of the performance equation but we don't care about the clocks when measuring efficiency. It's irrelevant. We care about the output ie. the performance. With what clocks it managed that performance does not matter - doesn't change the efficiency.

Your argument reminds me of people talking about what nm is their CPU. Like who cares. "oh but smaller nm means it's more efficient". Great, then ignore the nm and measure the efficiency, why would you care about the nm themselves

 

[TerryLaze]

Clocks are part of the performance equation but we don't care about the clocks when measuring efficiency. It's irrelevant. We care about the output ie. the performance. With what clocks it managed that performance does not matter - doesn't change the efficiency.

But the amount of performance changes depending on the clocks.
The amount of watts used for that clocks (and with that the performance) changes depending on the clocks.
So performance/watt changes depending on clocks.

Everything depends on the clocks and you are saying that clocks are not important.

What you are saying is that reviews should lock in a performance target, like locking in 120FPS in a game for example or x score in cinebench, and then look at what amount of watts each one will be using to do it.
I completely agree with that by the way, but nobody is ever going to do that.

 

[TheHerald]

But the amount of performance changes depending on the clocks.
The amount of watts used for that clocks (and with that the performance) changes depending on the clocks.
So performance/watt changes depending on clocks.

Everything depends on the clocks and you are saying that clocks are not important.

What you are saying is that reviews should lock in a performance target, like locking in 120FPS in a game for example or x score in cinebench, and then look at what amount of watts each one will be using to do it.
I completely agree with that by the way, but nobody is ever going to do that.

Exactly, they should either lock a performance target or lock the power draw. That's the only proper way to measure efficiency.

 

[bit_user]

Exactly, they should either lock a performance target or lock the power draw. That's the only proper way to measure efficiency.

We've been through this: it's not. Each CPU has a perf/W efficiency curve. Comparing the curves tells you about the CPU architecture & microarchitecture, but it's not of much use to an end user.

What's meaningful to an end user is when you cluster products according tiers of price, performance, power, etc. and then make detailed comparisons on those points. That's because users don't buy one CPU and then dial in the settings to make it match the performance of another CPU, on a given benchmark. That's utter nonsense. Users will either run a CPU using out-of-the-box defaults or according to tuning specific to that CPU.

Those are the two best ways to compare this stuff. Anything else isn't necessarily invalid, but definitely less useful.

 

[TheHerald]

We've been through this: it's not. Each CPU has a perf/W efficiency curve. Comparing the curves tells you about the CPU architecture & microarchitecture, but it's not of much use to an end user.

What's meaningful to an end user is when you cluster products according tiers of price, performance, power, etc. and then make detailed comparisons on those points. That's because users don't buy one CPU and then dial in the settings to make it match the performance of another CPU, on a given benchmark. That's utter nonsense. Users will either run a CPU using out-of-the-box defaults or according to tuning specific to that CPU.

Those are the two best ways to compare this stuff. Anything else isn't necessarily invalid, but definitely less useful.

What other devices do you benchmark like that? Probably not a lot? Efficiency of a fan (noise to performance) you do it noise normalized. Efficiency of a TV (if anyone cares) you do it at ISO brightness. AC unit? ISO temp etc.

If people don't know you can change your CPU's power like you can change your TV's brightness then maybe some education is needed, not meaningless benchmarks. Just saying.

 

[bit_user]

What other devices do you benchmark like that?

We're talking about CPUs. Best not to try and reason by analogy. Focus on the problem at hand.

Efficiency of a fan (noise to performance) you do it noise normalized.

Not necessarily. People don't always treat noise as their limiting factor. Why wouldn't it be CFM-normalized? But then how much CFM to use? Wouldn't it be better to measure at multiple points, in case different users need different amounts of CFMs? Oh, but then you end up with a curve, like I said!

Even if you use iso-noise measurements, how do you know you picked the right noise threshold for everyone? As above, you'd want to measure at multiple points, because not everyone runs at the same noise level and the rankings will sometimes change based on where you set the noise threshold, as in this case:

Source: https://www.anandtech.com/show/21376/capsule-review-sunon-maglev-120mm-fan
​

The reason you see people do point measurements like that is merely out of convenience, not because it's the best way to do it!

Efficiency of a TV (if anyone cares) you do it at ISO brightness. AC unit? ISO temp etc.

Not everyone watches their TV at the same brightness or sets their AC to the same temp (or with the same outdoor temp). Efficiency needs to be measured on a curve, so people can see how they compare at the point that's right for them!

 

[TheHerald]

Not necessarily. People don't always treat noise as their limiting factor. Why wouldn't it be CFM-normalized? But then how much CFM to use? Wouldn't it be better to measure at multiple points, in case different users need different amounts of CFMs? Oh, but then you end up with a curve, like I said!

​

Because CFM is an absolutely useless value to use. Just like clockspeeds. The CFM is only valuable for the performance it comes with, but then you might as well just measure the performance itself, not the CFM. Unless you are talking about non PC applications for fans.

Even if you use iso-noise measurements, how do you know you picked the right noise threshold for everyone? As above, you'd want to measure at multiple points, because not everyone runs at the same noise level and the rankings will sometimes change based on where you set the noise threshold, as in this case:

You cant' pick the right noise for everyone but the most efficient fan at 35 dba is not going to suddenly turn the worst at 38 dba. Excluding of course something being completely messed up with the fan, in which case you wouldn't notice even when testing with your own methodology.

Efficiency needs to be measured on a curve, so people can see how they compare at the point that's right for them!

Then we are not in a disagreement? I don't know why you are quoting me, but we are saying the same thing. when you have an efficiency curve you can then proceed to compare at ISO power which im suggesting.

 

[TerryLaze]

What's meaningful to an end user is when you cluster products according tiers of price, performance, power, etc. and then make detailed comparisons on those points. That's because users don't buy one CPU and then dial in the settings to make it match the performance of another CPU, on a given benchmark. That's utter nonsense. Users will either run a CPU using out-of-the-box defaults or according to tuning specific to that CPU.

Those are the two best ways to compare this stuff. Anything else isn't necessarily invalid, but definitely less useful.

You are utter nonsense and you explain why yourself...there are tons of people that play with vsync on, because the system comes that way or because that's what they want, there are also tons of people that have systems with locked down mobos that can't do anything to change the PL.
So seeing the power draw at vcsync would be extremely helpful for many people, as would be seeing the performance on a locked down mobo, yes even for the halo CPUs, less so but still.
Extreme overclock results, which is almost every review these days, are much less useful for end users.

 

[bit_user]

Because CFM is an absolutely useless value to use.

It correlates directly with cooling capacity, and that's what someone with a certain CPU & heatsink needs.

You cant' pick the right noise for everyone but the most efficient fan at 35 dba is not going to suddenly turn the worst at 38 dba.

No, there probably won't be a big inversion of the rankings, but I'll give you the example of a Noctua NF-A9x14 PWM fan I have. Up to about 1k RPMs, it's almost inaudible. Then, by just like 1.1k RPMs, it develops an annoying whine. That sort of non-linearity could potentially cause it to drop several slots in the rankings.

Then we are not in a disagreement? I don't know why you are quoting me, but we are saying the same thing. when you have an efficiency curve you can then proceed to compare at ISO power which im suggesting.

That's only useful for CPU geeks, not end consumers.

What an end consumer wants to know is how the performance & efficiency of a few similar CPU models compare, when tested as they would actually run them. As nobody is trying to decide between an i3-14100F and an i9-14900KS, having them in the same efficiency ranking is of no practical value. Furthermore, nobody who cares about efficiency would buy a R9 7900X and try to configure it to match the Cinebench numbers of an i7-14700K, or vice versa. So, measuring them this way is absurd. It just creates fake numbers that don't really help anyone. If you're trying to provide benchmarks to help real people decide which product is the best for them, then either compare them with out-of-the-box defaults (which many people use) or sanely-tuned settings that are achievable on most setups.

You don't need iso-performance or iso-power. As I said above, it doesn't make sense to compare products of wildly different performance or power, because nobody is trying to decide between a R7 7600X vs. an i9-14900K. If you merely stick to products within the same market segment, they will naturally cluster near enough, in terms of performance and power, and then you just show people how they compare in those respects so they can make tradeoffs that align with their priorities.

 

[bit_user]

You are utter nonsense and you explain why yourself...

Nothing I said disagrees with your post. Also, calling people nonsense is an ad hominem attack. I addressed only what Harold has been saying.

there are tons of people that play with vsync on, because the system comes that way or because that's what they want, there are also tons of people that have systems with locked down mobos that can't do anything to change the PL.
So seeing the power draw at vcsync would be extremely helpful for many people, as would be seeing the performance on a locked down mobo, yes even for the halo CPUs, less so but still.

This argues for testing with out-of-the-box defaults, which I'm in favor of doing (among other things).

Extreme overclock results, which is almost every review these days, are much less useful for end users.

Agreed. Personally, I think reviews of non-K/non-X CPUs are generally more interesting.

 

[TheHerald]

It correlates directly with cooling capacity, and that's what someone with a certain CPU & heatsink needs.

Then ignore the CFM and measure directly the results of the cfm, which are the temperatures. That's all I'm suggesting.

No, there probably won't be a big inversion of the rankings, but I'll give you the example of a Noctua NF-A9x14 PWM fan I have. Up to about 1k RPMs, it's almost inaudible. Then, by just like 1.1k RPMs, it develops an annoying whine. That sort of non-linearity could potentially cause it to drop several slots in the rankings.

I've already said that can be the case, but you can't detect them using your methodology either, so it's a non sequitur. In order to detect that you have to test the whole curve, which I'm all for anyways.

What an end consumer wants to know is how the performance & efficiency of a few similar CPU models compare, when tested as they would actually run them. As nobody is trying to decide between an i3-14100F and an i9-14900KS, having them in the same efficiency ranking is of no practical value. Furthermore, nobody who cares about efficiency would buy a R9 7900X and try to configure it to match the Cinebench numbers of an i7-14700K, or vice versa. So, measuring them this way is absurd. It just creates fake numbers that don't really help anyone. If you're trying to provide benchmarks to help real people decide which product is the best for them, then either compare them with out-of-the-box defaults (which many people use) or sanely-tuned settings that are achievable on most setups.

You don't need iso-performance or iso-power. As I said above, it doesn't make sense to compare products of wildly different performance or power, because nobody is trying to decide between a R7 7600X vs. an i9-14900K. If you merely stick to products within the same market segment, they will naturally cluster near enough, in terms of performance and power, and then you just show people how they compare in those respects so they can make tradeoffs that align with their priorities.

Man I really don't care about the " what most people would do" argument. I really, really, really don't. All I'm saying is the only method to text efficiency is at iso.

 

[Silas Sanchez]

Efficiency is work done for watts consumed. Clocks aren't a part of the equation.

No. Here, efficiency is the work done for energy consumed. Watts are not energy, watts is the metric for work done per unit of time, a different concept to energy. In the ratio you need both expressions to be the same unit and concept. For example, ratio of output energy produced to input energy consumed or output power to input power. Of course clocks are not part of it because by definition they cant be.

Its not acceptable to say, yes by watts I mean watt hours, this is a rookie mistake and you must be clear as some people genuinely dont understand and as a result this lacking clarity say wrong things like my fridge is drawing 50amphour or my fridge consumed 50amps, it happens quite a bit. And for anyone else who points out, yes I know ah isn't energy, but in 12V fridges ah is still more prevalent than wh for battery runtime calcs and its very intuitive over the watthour, and its the same mistake in essence.

 

[bit_user]

Man I really don't care about the " what most people would do" argument. I really, really, really don't. All I'm saying is the only method to text efficiency is at iso.

Then you really, really, really don't care about producing numbers of actual consequence to real users. Then, I'm left to ask why bother? The only thing left to fight over would be Internet Points, which are even more worthless than they sound.

 

[Li Ken-un]

The only thing left to fight over would be Internet Points, which are even more worthless than they sound.

A solipsist would certainly be happy to imagine that they do 🙂