iMore's reply to Linus on why macs are slower than PCs

[Stefan Payne]

1 hour ago, mr moose said:

Why do people still tout this?  There is absolutely nothing erroneous with Intel's TDP figures. 

When people knowledgable about stuff bash Intel for the TDP, it is wrong.

It became similar to AMD's ACP back in the day, wich Intel bashed. I've shown you the Document a couple of times.

1 hour ago, Morgan MLGman said:

They're designed to be misleading and are measured differently to what AMD does with their TDP ratings.

Its not just AMD its the Industry Standard, how it was used and understood for decades

 

The Industry Standard for TDP for _ALL COMPONENTS_ is either average maximum under full load or the absolute maximum rating.

 

Intel is the odd one out that redefined their Specification for "base clock" and exempted the Boost Clock from the TDP.

I don't know when they changed it, couldn't find any information about that on the Internet.

What I do know is that back in 2011 with the first generation i7 on LGA1366 and probably also LGA1156, the TDP meant average consumption under load. Including the Boost Clock (wich was negligable at the time anyway. Just around 200MHz more or so).

It was after that. Wich means at the earliest 2012 with Sandy Bridge, when they changed it. BUT they still used in practice the industry standard!

They really violatd it in the last 2 years!

The i7-7700K under max load, barely violates the TDP at all! As you would expect.

THe reviews I've found about the 7700K show it around 90-100W for max. Load - what you expect.

 

It really got weird with the 8k and 9k series...

 

 

Before that, Intel used the Industry Standard wich means TDP = average maximum power consumption, even for the Turbo!

 

 

[Morgan MLGman]

3 minutes ago, Stefan Payne said:

SNIP

I agree, if you look at my latest post I actually elaborated on that as well

[mr moose]

1 minute ago, Morgan MLGman said:

Intel actually used to depict TDP pretty much the same way as AMD does. They changed it not for OEMs, but for the fact that it looks better to customers that don't really know what that rating is

Say who?  why do you think that?

1 minute ago, Morgan MLGman said:

- this is my problem with their TDP and why I said it's misleading. It's not misleading to me or you, because we actually have any kind of knowledge about the topic, that doesn't mean that it's not misleading at all.

Um yes it does, not understanding something does not equal it being misleading.   And to be quite honest, who the hell buys a CPU based on the TDP rating?  Are you going to design your own cooler? did you not read the reviews on the CPU?

1 minute ago, Morgan MLGman said:

I can give you several examples of politicians using precisely the same tactic to mislead people about data that they find only partially good for their agenda. They're not wrong, and you can't call them out for being wrong, but they also didn't give you the bigger picture on purpose that would change the way you interpret that data.

Except this is just a cut and dried product spec.  there is nothing flash or major about it,  if you thought it meant power draw then blame whatever internet idiot told you that,  because Intel never said that.

1 minute ago, Morgan MLGman said:

Fun fact: Intel actually tried to do TDP ratings a better way not so long ago (in the times of mobile Ivy Bridge & Haswell Y-Series CPUs), they created something called SDP (Scenario Design Power) which according to Intel represented "the average power consumption of a processor using a certain mix of benchmark programs to simulate "real-world" scenarios". But they're Intel so they ditched it.

Fun fact, companies drop stuff when it doesn't work all the time, insinuating a nefarious reason doesn't mean anything.  I don't see any reason why you would make such a casuistic link between something so common as being for nefarious purposes.

 

 

1 minute ago, Morgan MLGman said:

Intel's definition of SDP was:

And IMO it's a more realistic metric. Or useful at all, unlike current TDP
 

 

There are lots of specs surrounding power and thermals.

Here is a great example of how these ratings are different depending on the manufacturer and the unit design and end use.

 

http://www.ti.com/lit/ds/symlink/na555.pdf

 

This is a chip I used to use all the time on various products.  If you scroll down to 7 Specifications, you'll see about seven different parameters for temperature including thermal impedance and thermal resistance.  Items 3,4 and 5 show how to calculate the thermal and power dissipation conditions for reliability is not a single straight up figure.  This is because circuit design, materials, power requirements, end use conditions are all different for every product.  The entire EE industry is littered with different criteria for defining TDP or a variant on that.

 

If Intel wanted to they could list theirs in exactly the same way TI has for the 555, just make a few minor adjustments so it covers all the necessary attributes of their specific product and away it goes.  But I suspect something so engineering oriented would be interpreted as intentionally trying to mislead the public.  Have TI mislead the public because you have to have a decent educatiuon to understand their specs?

[mr moose]

5 minutes ago, Stefan Payne said:

When people knowledgable about stuff bash Intel for the TDP, it is wrong.

It became similar to AMD's ACP back in the day, wich Intel bashed. I've shown you the Document a couple of times.

Its not just AMD its the Industry Standard, how it was used and understood for decades

 

The Industry Standard for TDP for _ALL COMPONENTS_ is either average maximum under full load or the absolute maximum rating.

 

Intel is the odd one out that redefined their Specification for "base clock" and exempted the Boost Clock from the TDP.

I don't know when they changed it, couldn't find any information about that on the Internet.

What I do know is that back in 2011 with the first generation i7 on LGA1366 and probably also LGA1156, the TDP meant average consumption under load. Including the Boost Clock (wich was negligable at the time anyway. Just around 200MHz more or so).

It was after that. Wich means at the earliest 2012 with Sandy Bridge, when they changed it. BUT they still used in practice the industry standard!

They really violatd it in the last 2 years!

The i7-7700K under max load, barely violates the TDP at all! As you would expect.

THe reviews I've found about the 7700K show it around 90-100W for max. Load - what you expect.

 

It really got weird with the 8k and 9k series...

 

 

Before that, Intel used the Industry Standard wich means TDP = average maximum power consumption, even for the Turbo!

 

 

See my last post.  A gave you a good example of an industry standard that was completely different again.

[leadeater]

2 minutes ago, mr moose said:

See my last post.  A gave you a good example of an industry standard that was completely different again.

Also industry standard is to not have "boost" or similar. My MOSFETs in my amps don't "turbo". Other than CPUs and GPUs the rest of the industry does not have this so it can't be a standard used in engineering because it's just not a thing.

[Stefan Payne]

1 minute ago, leadeater said:

Also industry standard is to not have "boost" or similar. My MOSFETs in my amps don't "turbo". Other than CPUs and GPUs the rest of the industry does not have this so it can't be a standard used in engineering because it's just not a thing.

They sometimes have a peak rating...

 

Besides:

Intel had Turbo for Years yet they still used the TDP = max. Power under normal use, including Turbo.

Just look at the i7-7700K. It is around the 90W +/- 10%.

[mr moose]

3 minutes ago, leadeater said:

Also industry standard is to not have "boost" or similar. My MOSFETs in my amps don't "turbo". Other than CPUs and GPUs the rest of the industry does not have this so it can't be a standard used in engineering because it's just not a thing.

I just don't know why this is such a hard concept to understand.

 

EDIT:

I get the feeling I could list every single reference to power dissipation and thermal design  characteristics across a thousand products and someone would still argue the bog stock Wikipedia explanation or "but Intel did this" as if that settles it. 

[leadeater]

9 minutes ago, Stefan Payne said:

Intel had Turbo for Years yet they still used the TDP = max. Power under normal use, including Turbo.

Just look at the i7-7700K. It is around the 90W +/- 10%.

Things change, products evolve. Older CPUs lacked the capability to actually have many C-States and P-States so had very minimal power and frequency differences. Now we can, and it's better because we can have more compact devices that will boost for the short times we need it to. All other times the CPU is in a lower power state so why should the TDP spec be unnecessarily higher than it needs to be to account for a 2 minute spike that can be handled by a standard cooler and a bump in fan speed as required.

 

Did you know Xeons use a different TDP spec to desktop CPUs? Because it's based off a different reference workload and run time.

[Stefan Payne]

1 minute ago, leadeater said:

Things change, products evolve.

No, competition happened.

THey had to increase the Core Count and had two possibilitys
a) stay honest and call the 8700K a 125-140W TDP CPU

b) fiddle around with the TDP and be creative with the use of it and make it look like 95W.

 

1 minute ago, leadeater said:

Older CPUs lacked the capability to actually have many C-States and P-States so had very minimum power and frequency differences. Now we can, and it's better because we can have more compact devices that will boost for the short times we need it to. All other times the CPU is in a lower power state so why should the TDP spec be unnecessarily higher than it needs to be to account for a 2 minute spike that can be handled by a standard cooler and a bump in fan speed as required.

Because in the Small Enviroments you mentioned, 80W PSU can be used, wich would be overloaded when the CPU boosts to whatever it feels like it.

 

1 minute ago, leadeater said:

Did you know Xeons use a different TDP spec to desktop CPUs? Because it's based off a different reference workload and run time.

Xeons are Xeons, and a different thing.

 

THe thing is that it wasn't until very recently that Intel really changed the meaning of TDP. The i7-7700K, from what I could find in 5min googling, show that it stays within TDP.

The i7-8700K is where it gets violated and the 9900K seems like the TDP is not worth the paper its written on.

[Morgan MLGman]

1 minute ago, mr moose said:

Say who?  why do you think that? 

What would be another reason to ditch the standard method? This way they can create a CPU with let's say 64 cores, advertise its base clock as 1GHz or 0,5GHz, even though it's actually running at 3,5GHz for everyone because of "Turbo" and say that the TDP is less than half of what it actually should be. What prevents Intel from doing that? They'd still be technically correct within their own current TDP definition.

2 minutes ago, leadeater said:

All other times the CPU is in a lower power state so why should the TDP spec be unnecessarily higher than it need to be to account for a 2 minute spike that can be handled by a standard cooler and a bump in fan speed as required.

I agree, provided the standard cooler can actually handle that spike. Apple's didn't, so there's the problem. I personally put the blame for that 100% on Apple, however there may be cases where regular people purchase inadequate coolers based on the TDP ratings and end up with throttling 9900K's
Not to mention about what happens when AVX instructions are being used... Skylake-X products should have another "TDP"-like rating that applied specifically when running AVX-512 workloads, haha

[leadeater]

Just now, Stefan Payne said:

No, competition happened.

I was referencing 2011 era CPUs up until now. Back then there was very little difference between SKUs for power draw and Intel largely applied the same TDP across an entire range without regard for power usage differences, top SKU was the TDP for all. Intel still largely does that but the difference now is that the base clocks are controlled to it matches TDP. There's also no more i3-560s with locked clocks (no boost) with the same TDP as an i5-680 with much higher base and can boost even more on the same architecture and node. The old TDP was not better, in fact in many ways it was worse.

[GoldenLag]

1 hour ago, Morgan MLGman said:

If you bought a 95W cooling solution to use with a 9900K, you'd have a pretty bad time

AMD wraith spire on a 9900k would be an interesting test. 

 

1 hour ago, mr moose said:

Also I don't know any coolers that are sold with watt ratings,  probably a very good reason for that.

https://www.bequiet.com/en/cpucooler/1378

 

so how would a Drp4 vs 9990XE fair? (an yes i am aware the 9990XE has a 5 watt higher TDP, but that shouldnt be a major factor)

 

or 9980xe vs shadow rock slim

https://www.bequiet.com/en/cpucooler/479

 

both sold with TDP spec. 

[mr moose]

3 minutes ago, Stefan Payne said:

No, competition happened.

THey had to increase the Core Count and had two possibilitys
a) stay honest and call the 8700K a 125-140W TDP CPU

b) fiddle around with the TDP and be creative with the use of it and make it look like 95W.

 

Because in the Small Enviroments you mentioned, 80W PSU can be used, wich would be overloaded when the CPU boosts to whatever it feels like it.

 

Xeons are Xeons, and a different thing.

 

THe thing is that it wasn't until very recently that Intel really changed the meaning of TDP. The i7-7700K, from what I could find in 5min googling, show that it stays within TDP.

The i7-8700K is where it gets violated and the 9900K seems like the TDP is not worth the paper its written on.

 

Just now, Morgan MLGman said:

What would be another reason to ditch the standard method? This way they can create a CPU with let's say 64 cores, advertise its base clock as 1GHz or 0,5GHz, even though it's actually running at 3,5GHz for everyone because of "Turbo" and say that the TDP is less than half of what it actually should be. What prevents Intel from doing that? They'd still be technically correct within their own current TDP definition.

I agree, provided the standard cooler can actually handle that spike. Apple's didn't, so there's the problem. I personally put the blame for that 100% on Apple, however there may be cases where regular people purchase inadequate coolers based on the TDP ratings and end up with throttling 9900K's
Not to mention about what happens when AVX instructions are being used... Skylake-X products should have another "TDP"-like rating that applied specifically when running AVX-512 workloads, haha

 

 

Oh my,  there is no "standard method".  Every product (literally every single product) is different and requires a different method for determining thermal specifications.    There is nothing standard about this and never has been.   You know what is a standard?  Jedec,  they are standards, MIL-STD-883 is a standard.  TDP is not.  TDP is specific only to each product as each manufacturer sees fit. If the manufacturer can not guarantee a specific thermal dissipation requirement after specific conditions have been passed then they are not going to give it one. 

 

The specs are not false, wrong, erroneous, fiddled, creatively produced or anything like it.    They are specifically determined and specifically listed.  if someone buys a cooler that is too small for their CPU it is because they didn't bother to educate themselves first.  Not because the spec is misleading.

[leadeater]

21 minutes ago, Stefan Payne said:

The i7-8700K is where it gets violated and the 9900K seems like the TDP is not worth the paper its written on.

Well no because I can 100% guarantee you were looking at reviews using a static workload to give you an average power usage for that application where the Intel spec isn't that. Intel doesn't run a synthetic benchmark, they use a workload simulation and take the average of that. In Intel's tests the power state and draw of the CPU is changing throughout the test. Edit: Capped to Base frequency.

 

The spec isn't violated.

 

Edit:

Also as we should all know by now motherboard makers were defaulting to MCE On.

[mr moose]

2 minutes ago, GoldenLag said:

AMD wraith spire on a 9900k would be an interesting test. 

 

https://www.bequiet.com/en/cpucooler/1378

 

so how would a Drp4 vs 9990XE fair? (an yes i am aware the 9990XE has a 5 watt higher TDP, but that shouldnt be a major factor)

 

or 9980xe vs shadow rock slim

https://www.bequiet.com/en/cpucooler/479

 

both sold with TDP spec. 

It appears bequiet do it.  Happy to accept that, but  if you want to know how it would go you need to look up reviews on a power draw or you need to buy one and measure it as you test the coolers abilities.   The thing here is you need the specific definition of bequiets TDP rating (it might not be designed to match a CPUs listed TDP).  then you need to determine the amount of heat the 9990XE will need to have dissipated to maintain the thermals you require (inside Intel limits) at the load you want.   I know it sounds like more work than a consumer wants, but the TDP rating is not for general consumers. 

 

Or you could just look up reviews for the best performing cooler for the CPU you want to buy and make sure you don't buy one that is too small for your desired outcome.  This way you don't have to confuse yourself with a spec that is largely misunderstood.

 

 

[GoldenLag]

9 minutes ago, mr moose said:

TDP is specific only to each product as each manufacturer sees fit.

Except there is a Pretty standard way to calculate heat transfer

 

https://www.engineeringtoolbox.com/conductive-heat-transfer-d_428.html

 

Offcourse one needs to decide what speedsetting TDP would be, but for manufacturers of coolers they should be able to create damn accurate dissipation models. 

 

It should then be able to dissipate what the CPU TDP says is the needed dissipation.

[mr moose]

1 minute ago, GoldenLag said:

Except there is a Pretty standard way to calculate heat transfer

 

https://www.engineeringtoolbox.com/conductive-heat-transfer-d_428.html

 

Offcourse one needs to decide what speedsetting TDP would be, but for manufacturers of coolers they should be able to create damn accurate dissipation models. 

 

It should then be able to dissipate what the CPU TDP says is the needed dissipation.

I've often argued heatsinks should have their thermal characteristics listed (many generic component ones do).  But unfortunately that still doesn't change the need to understand each companies TDP.  Knowing how good your cooler is is one thing, but if you haven't done you homework on the CPU you bought then that is a pointless spec.  

[GoldenLag]

10 minutes ago, mr moose said:

then you need to determine the amount of heat the 9990XE will need to have dissipated to maintain the thermals you require (inside Intel limits) at the load you want

Well if anything is accurate about intel TDP, it should hold its base clock in any non-AVX workloads. They have set the TDP. Cooler designers just need to have a cooler capable of dissipating that ammount lf heat. 

7 minutes ago, mr moose said:

The thing here is you need the specific definition of bequiets TDP rating (it might not be designed to match a CPUs listed TDP)

No its not designed against any CPU TDP. Because its not supposed to. They give the CPU cooler an ammount if heat they can confirm the cooler is capable of dissipating. 

 

8 minutes ago, mr moose said:

This way you don't have to confuse yourself with a spec that is largely misunderstood.

The basic understanding of the TDP spec is that it is the required heat dissipation capacity to run the CPU within spec (base clock or higher

[GoldenLag]

3 minutes ago, mr moose said:

But unfortunately that still doesn't change the need to understand each companies TDP

That is true if the TDP doesnt give info on the heat dissipation needed to cool the product within spec. 

4 minutes ago, mr moose said:

often argued heatsinks should have their thermal characteristics listed (many generic component ones do).

If a company guarentees a certain heat dissipation with a cooler. Then that should have been calculated using uncluded components and a scientific formula that decribes reality, or a model that describes reality.

 

Afaik that is what cooler designers do.

[mr moose]

2 minutes ago, GoldenLag said:

The basic understanding of the TDP spec is that it is the required heat dissipation capacity to run the CPU within spec (base clock or higher

Just base clock,  it is not "or higher",  if the clock goes higher than base then the amount of heat that needs to be dissipated goes higher than the spec.

[mr moose]

12 minutes ago, GoldenLag said:

That is true if the TDP doesnt give info on the heat dissipation needed to cool the product within spec. 

If a company guarentees a certain heat dissipation with a cooler. Then that should have been calculated using uncluded components and a scientific formula that decribes reality, or a model that describes reality.

 

Afaik that is what cooler designers do.

 

Hang on , Are you specifically talking only about coolers, because the conversation was about CPU's. 

 

It is easy for a company making a cooling solution to give it a definitive performance metric.  It appears CPU coolers are one of the few that don't do that.  Every time I look at heat sinks they usually have the thermal resistance spec'd (I.E X^oC/W),  this gives you the ability to choose the appropriate size sink for you project/product.    Should CPU coolers do this.  From an enthusiasts perspective yes, but from a general consumer view point no.  Just like people misunderstand the point of TDP on a CPU spec sheet they would also misunderstand the thermal resistance rating on a cooler and just as likely buy the wrong one. I say this because people looking up those specs are doing themselves rather than relying on a few reviews to pick the best one, and we already know people get confused with technical specs.

 

 

[maartendc]

I have to say the video title by Linus is totally clickbait.

 

And the article, which I read about halfway through, is merely pointing out that any thin and light with an i9 in it would throttle. Which is true. Shame on Apple for offering the option to put an i9 in it, but that is about it.

 

A macbook (pro) with an i5 or i7 (which is what 99% of the market would buy) is NOT slower then a windows thin and light. So: clickbait.

 

I find myself watching Linustechtips less and less these days, simply because the videos are either: clickbait, or over the top stuff that has no pratical use (7 gamers one CPU for example), or for less-tech-savy users (explaining things I already know).

 

I much rather watch Gamers Nexus, Hardware Unboxed and Paul's Hardware nowadays. Actual reviews and topics that I have a use for.

[GoldenLag]

30 minutes ago, mr moose said:

Hang on , Are you specifically talking only about coolers, because the conversation was about CPU's

Yes the ammount of heat a cooler cab dissipate can be calculated accuratly to a high margin.

 

The TDP of CPU is supposed to tell how much cooling capacity is needed. If the TDP isnt indicative of how much cooling is needed, then its a "false" description.

 

32 minutes ago, mr moose said:

I say this because people looking up those specs are doing themselves rather than relying on a few reviews to pick the best one, and we already know people get confused with technical specs

People should allways look at reviews, but that doesnt excuse bad TDP rating

[jagdtigger]

37 minutes ago, GoldenLag said:

If the TDP isnt indicative of how much cooling is needed, then its a "false" description.

Pretty much this, top it off im sure apple is well aware that intel tweaked their TDP definition to get better marketing numbers. But for "some" reason they fail to adapt their cooler design to the real heat output of the cpu.... Apple just keeps on sinking lower and lower.

[GoldenLag]

6 minutes ago, jagdtigger said:

Pretty much this, top it off im sure apple is well aware that intel tweaked their TDP definition to get better marketing numbers. But for "some" reason they fail to adapt their cooler design to the real heat output of the cpu.... Apple just keeps on sinking lower and lower.

Intel TDP is mostly an issue due ti the fact mobo vendors tend to break it out of the box.