14nm 6th Time Over: Intel Readies 10-core "Comet Lake" Die to Preempt "Zen 2" AM4

[jpiniero]

So they require a 130w cooler for a 90w CPU?

The actual text on Ark: "Intel Reference Heat Sink specification for proper operation of this SKU"

 

[Topweasel]

I meant that to be sarcastic. But I looked at what you were refering to. Its not even a true spec (though ARK pretends it is) It's a statement of what OEM cooler Intel would use for the CPU if they came with coolers. Not an actual statement of recommended cooler capability.

They state the TDP as a complex workload benchmark for coolers at 91w. But then says refer to datasheet without any link to it never pointing to the Intel OEM cooler part number.

It's a poop show.

 

[PotatoWithEarsOnSide]

We shouldn't confuse the power dissipated by the heatsink with thermal throttling.
Where you hit the TDP limit of power dissipation, the CPU will revert to p1 state, and operate at 4.2GHz AC within the 95w limit. This is intended behaviour, though on the Maximus XI Hero conforms to it.
Thermal throttling occurs where the temperature on die exceeds a predetermined limit. This can, and should, happen when the power draw is well above the heatsink's ability to dissipate power (heat) for a prolonged period of time. It is designed to prevent damage to the CPU in the event of an inadequate cooling solution. It will override any power state until the thermals allow normal operation. For those that overclock, they tend to ensure that their cooling meets their needs. However, all of the mobos (other than the one mentioned earlier) will default to operate at a higher power draw than spec. This choice does not mean that the CPU doesn't perform as advertised, because quite frankly it does. It simply means that the mobo manufacturers are willfully ignoring spec in order promote their own product's competitiveness. Afterall, who would care which one they bought if they were all the same due to being at spec? The problem is that none of them make it easy for you to set your CPU to perform at spec.
The reality is that the mobo manufacturers have decided that they know best as to how we want to set up our systems, and for me that is simply the wrong mindset. Most K-series owners do not even overclock, and these are the same folk likely to get blindsided by inadequate cooling, thus thermal throttling...potentially.
Intel clearly isn't an innocent bystander, since they should insist that default BIOS settings should be such that their CPUs operate at spec. IMO, that is the omission that they are most guilty of here. However, that doesn't mean that it is primarily their fault.

---

For anyone that has read my past posts, it's clear that I favour AMD, and have been heavily critical of Intel rather often.

 

[Zucker2k]

That really is the rub there. Intel actively sells this as 95w CPU, but they also sell it as a up to 5GHz CPU per their page. Sure they also give you base performance. But they don't say "hey anything above base performance uses more power than our rating (which hasn't been the norm till now)". But in fact to get this CPU to running ad advertised (which is not base speed) you need twice the cooling power. This is fine print crap we should be past and well if it has to be there, you know, actually put it in the fine print.

https://ark.intel.com/products/186605/Intel-Core-i9-9900K-Processor-16M-Cache-up-to-5-00-GHz-

Without a single note on the site of max power draw. Unless I was digging through the forum, how am I supposed to figure out what size cooler to get. 95w+ like the TDP implies. Nope that's out. How about a more run of the mill enthusiast 120w cooler? Nope no good either. Well lots of LGA2011 CPU's ran at 140w would one of those work? Nope. Now we are getting into dual chip TR or TR2 territory with ~160w cooler required, but not sure, and probably sometimes more because not all silicon is going be the same. This isn't a big issue if you know ahead of time, but now that's on us to make sure people know because Intel sure doesn't. Then the final nail in this power coffin is that this 90w CPU no matter what level you are looking at it requires pretty much a cooler rated at twice its rated power usage to not be running at it's thermal limit. It's not great science to say this but whether its a CPU or Powersupply or Video card I don't like them sitting at their limit for extended periods of time and treat it as running your car at the rev limit for a long time. Now we are talking about any load whether you got a 90w or 160w cooler the CPU is running at it's cutoff temp. For me if I got this CPU I wouldn't feel comfortable without something like a 250w cooler to keep the CPU far enough off it's limit.

There is just so much bad about the power rating on this CPU. The problem is they wanted to keep it accessible to OEM's but the fact is if you say your CPU is 160w CPU it isn't the end of the world. I know I wouldn't be nearly as bothered by buying a known 160w CPU then finding out I am not seeing nearly the upper end clocks because I mistakenly thought a 120w cooler would be more than enough headroom for a 95w CPU.

A lot of what you posted is not true. PL2 is turbo, and has inherent checks. You don't have to get twice the cooling, only if you want to run at 5GHZ all day. That's your choice, as with any other chip. The 9900k is running great in a notebook. How's that even possible for this 210w monstrosity? I hope Intel doesn't follow rants like yours cos we'll be back to locked enthusiast chips.

 

[PotatoWithEarsOnSide]

If the 9900k was a locked, no Turbo Boost, 8c/16t CPU, it would be clocked at 4.2GHz and would be operating within 95w.

 

[Topweasel]

A lot of what you posted is not true. PL2 is turbo, and has inherent checks. You don't have to get twice the cooling, only if you want to run at 5GHZ all day. That's your choice, as with any other chip. The 9900k is running great in a notebook. How's that even possible for this 210w monstrosity? I hope Intel doesn't follow rants like yours cos we'll be back to locked enthusiast chips.

Sorry I missed the launch of a 8c16t Laptop CPU from Intel. Can you point me to the model number? It's Clockspeed? It's max turbo? Its power usage with that turbo?

PL2 is A turbo not the turbo but I get your point. What is the average time this is running on non-OEM boxes?

As for choosing to run this at 5GHz all day you missed my point and outside PL2 boxes where the OEMs build the system around a specific set of BIOS settings and over all cooling solution. 90% of the motherboards for this guy have no time limit on PL2 because of a few features set by default I forgot the specific setting. Something experience. It started showing up around 7th gen but got a lot of coverage when reviewers were going over the 8th gen option.

So I as an average enthusiast user gets a 9900k. I plop it into a 390z board. I am not a big overclocker. That's why I got the 5GHz chip from Intel. I decide to fire up handbrake. So what does the CPU do during a few different cooling solutions? Well if I get a 90-95w cooler, the CPU applies the turbo's, the CPU quickly hit thermal limit. I am very quickly running with a very small turbo, cpu stays pinned against its thermal limit. Okay so not a good solution. So I thought ahead got a 120w cooler or the intel "spec" of 130w cooler. This is 30% more in cooling then the TDP of the CPU. So what's the result, I get a modest turbo, but the CPU is pinned against it's thermal limit. So lets say I have Anadtechs CPU and I was smart I went and got a 160w cooler, now on Anandtech's CPU I finally am getting max turbo's with a cooler ~66% more cooling capability than the 95w TDP rating implies. But guess what I am still pretty near pinned against the thermal limit of the CPU. Now this is Anandtech's chip. I doubt anyone would end up with a 9900k that needed 210w for max turbo (would be binned for something else if it was that bad) but Intel certainly left room for more power usage to meet advertised performance, review performance, everything performance related mentioned everwhere but the "base clocks". So what if it's 180w and 180w cooler? Pinned against thermal limits. So if I get a base 90w cooler all the way up to 160w-180w cooler. I could the whole time I am running anything semi intensive be at the CPU's thermal limit. Something else that has been creeping up. Not too long ago it was just above 70c, now operational limit is at 100c.

So the point being that unless I go and turn features off or enter in my own non-default P states. The CPU until you go way overboard will always be running at it's temp limit. It's crazy to think of the cooling effort someone would have to go to, to run a "95w TDP" CPU at lets say 70c under load.

 

[PotatoWithEarsOnSide]

It's true; you are not an overclocker. However, the mbos are default overclocked, so the choice has been taken away from you.
In the event that you want a CPU that runs to spec, a) buy the only supporting mobo, b) understand how and what Turbo Boost is, and c) don't expect miraculous performance that doesn't fit with the known parameters of previous generations of essentially the same CPU.
A non-overclocker should only ever buy the Maximus XI Hero with XMP On, and further enhancements Off. That way you get a 9900k that adheres strictly to spec, which is 2c turbo to 5GHz, and ACT to 4.7GYz for a short period of time, typically less than 30s. With that you get an effective 24/7 frequency that is 4.2GHz all core, which is within 95w, and above the advertised base frequency.
For all other motherboards, pray.

 

[coercitiv]

Intel has actually required a 130W cooler for K for some time.

The actual text on Ark: "Intel Reference Heat Sink specification for proper operation of this SKU"

They changed the reference cooling solution for K processors back during the original Skylake launch, went from a "95W" TDP solution to "130W" TDP solution.

However, back then the change was clearly meant to improve temperatures rather than actually increasing operating TDP. Intel clearly stated the Tcase for 6700K as being 64C. The 130W cooler Intel mentions for both 9900K and 6700K is PCG 2015D, and we can get a peek at it's thermal profile:

You can also check the thermal dissipation values in this datasheet, but the TDP of this cooler for a Tcase of 64C is 92W.

In the case of 9900K Intel conveniently mentions only Tjunction, meaning we have no idea of what TDP the PCG 2015D would have while cooling this CPU.

 

[coercitiv]

And here we go, Ian Cutress joins the true 95W TDP review club.

 

[PotatoWithEarsOnSide]

"it acts like a 9900k in single thread mode, and like a 9700kin multithread mode"

 

[LTC8K6]

One of the interesting things in these benchmarks is that when in 95W mode, especially in shorter tests, the 9900K actually performs better than the full grunt settings. This could be because the system doesn't have to consider current limits of the power delivery, as 95W is the guaranteed limit no matter the loading.

I think that should have been investigated to see what was going on.

 

[LTC8K6]

I would like to see it clamped to Intel's specs, rather than just 95W or unlimited.

Unlimited is not Intel's spec.

 

[LTC8K6]

So while Intel defines a value for PL1, PL2, and Tau, almost zero consumer motherboard manufacturers actually follow it.

So what happens when you follow Intel's definitions?

 

[coercitiv]

So what happens when you follow Intel's definitions?

Straight from a stock CPU confession:

• CPU is idling around.
• Sudden tsunami workload hits, on all 16 fronts: Blender render alert.
• All cores wake up from idle.
• Frequency ramps up towards max turbo bin.
• Uncore sees the PL1 sign flashing before it's eyes, workload is so demanding it barely remembers to write down the moment it went past 95W power usage.
• Soon enough it smashes into the PL2 blockade, going over 120W is not allowed for more than a few milliseconds. This is it! The POWER!
• Power usage stays at 120W for the next 100 seconds. Good thing this is a desktop, CPU thinks, those notebook champs can only get 30 seconds of Super Sayan transformation.
• Time passes quickly when you're having fun: after 1m40s the CPU is forced to slow down, returns to PL1 flag.
• It's all a 95W cruise from here until load goes away, the highway is so beautiful and peaceful under this light fan breeze.
• Reached the final mark, workload complete!
• CPU ramps down under base clocks, then parks most cores in C7.
• Hot air smell is great in the evening. So many fans stop spinning under the moonlight.

 

[jpiniero]

And here we go, Ian Cutress joins the true 95W TDP review club.

Yeah, that was what I was kind of getting at. You/the board could certainly configure to do PL2=95W but because of the core count and stock max turbo clocks it can draw much higher than that if allowed to without it technically being overclocking.

 

[TheGiant]

And here we go, Ian Cutress joins the true 95W TDP review club.

the review lacks the information of all other cpus and better MB+ram+cpu components with a certain level of power draw and better MB recommendation (so the MB can handle the power)
like 95W, 120W,150W, whatever it takes, max reasonable overclock power, max stable overclock power with custom loop

that review is again comparing the 95W 9900K media debacle with all others, but without the same setup

either we focus on power or not, but when we do, IMO the power should be the CPU, MB, RAM combo
if I care about power, then the CPU is only one of the parts inside the system

 

[Topweasel]

It's true; you are not an overclocker. However, the mbos are default overclocked, so the choice has been taken away from you.
In the event that you want a CPU that runs to spec, a) buy the only supporting mobo, b) understand how and what Turbo Boost is, and c) don't expect miraculous performance that doesn't fit with the known parameters of previous generations of essentially the same CPU.
A non-overclocker should only ever buy the Maximus XI Hero with XMP On, and further enhancements Off. That way you get a 9900k that adheres strictly to spec, which is 2c turbo to 5GHz, and ACT to 4.7GYz for a short period of time, typically less than 30s. With that you get an effective 24/7 frequency that is 4.2GHz all core, which is within 95w, and above the advertised base frequency.
For all other motherboards, pray.

Where is it defined that 4.2 GHz is the limit within the 95w TDP?
Is there any chance of of not getting that 4.2GHz sustained with a 95w cooler?

What you don't understand is that while I admit to not being an avid overclocker back when it really matters. The settings behind controlling the upper limits of the CPU are not beyond my grasp. I am just noting that in every situation that some buys this CPU they will end up with 2 situations one that they will be capped below the advertised speed (and don't try to ignore that, Intel sells this as a 5GHz CPU, not a 4.2, not a 3.5GHz CPU, a 5GHz CPU). Or they wise up and have to get a cooler 2.5 times the rated TDP to prevent the CPU from running against it's thermal limit under load.

Turbo did't used to mean Overclocked. It meant running higher momentarily till the CPU judged itself to have exceeded the cooling limit of its rated TDP and stepped back down. Both AMD and Intel have basically turned it into an art of selling a CPU overclocked by default. But at least with AMD you know what the cooler limited clock speed is in it's TDP rating and specifies XFR as a bonus possibly limited boost that can be sustained if you have spare cooling. This guy throws that out of the water so they can make sure they can advertise a 8c CPU at 5GHz.

I want to note that I am in no way calling this a bad CPU. It's a bit of a bad value imho but that's besides the point. I am only saying the TDP rating, power usuage, and lack of useful information from intel (and not forums like Anandtech) make this a huge issue in my mind and that fall down the slippery slope people saw coming back when Intel stopped measuring TDP as max power usage with the Pentium 4.

 

[jpiniero]

I am just noting that in every situation that some buys this CPU they will end up with 2 situations one that they will be capped below the advertised speed (and don't try to ignore that, Intel sells this as a 5GHz CPU, not a 4.2, not a 3.5GHz CPU, a 5GHz CPU).

The advertised base clock speed is 3.6.

 

[coercitiv]

that review is again comparing the 95W 9900K media debacle with all others, but without the same setup

Yup, as I said: this comes back and bites Intel where only infrared shines.

They should have handled this TDP conflict the same way they handled XMP memory: make it a switch in BIOS and advertise the higher performance numbers for a win-win scenario. I'm actually surprised AMD marketing people knew to present this better in the form of XFR, which if you think about it, is exactly what Intel wanted: TDP violation cheered by consumer.

The consumers want this: more performance for more power is a trade-off many are willing to make - as long as it's done transparently. The problem was never the power consumption per se, but rather lying about power consumption, and leaving the press to explain it to consumers, which they did poorly as they were blindsided and left in the dark. Not the brightest PR strategy imho.

Personally I'll patiently wait for the Zen 3000 and the approval of the community for whatever XFR brings to the table when pushing the CPU beyond stock. I bit my tongue this year and openly agreed that power consumption is not relevant for enthusiasts (at least less relevant than performance), I hope next year doesn't surprise us with a 180 degree spin on CPU power usage and the importance of stock TDP ratings in reviewing competing products on equal terms.

 

[LTC8K6]

Straight from a stock CPU confession:

• CPU is idling around.
• Sudden tsunami workload hits, on all 16 fronts: Blender render alert.
• All cores wake up from idle.
• Frequency ramps up towards max turbo bin.
• Uncore sees the PL1 sign flashing before it's eyes, workload is so demanding it barely remembers to write down the moment is went past 95W power usage.
• Soon enough it smashes into the PL2 blockade, going over 120W is not allowed for more than a few milliseconds. This is it! The POWER!
• Power usage stays at 120W for the next 100 seconds. Good thing this is a desktop, CPU thinks, those notebook champs can only get 30 seconds of Super Sayan transformation.
• Time passes quickly when you're having fun: after 1m40s the CPU is forced to slow down, returns to PL1 flag.
• It's all a 95W cruise from here until load goes away, the highway is so beautiful and peaceful under this light fan breeze.
• Reached the final mark, workload complete!
• CPU ramps down under base clocks, then parks most cores in C7.
• Hot air smell is great in the evening. So many fans stop spinning under the moonlight.

this serious moonlight.

 

[PotatoWithEarsOnSide]

I agree with you that it is a marketing gimmick, though it is one that is technically correct with the manner in which Turbo Boost works. They set a base clock that allows thermal headroom for short bursts at increased clocks, which when reached revert the clocks to its 24/7 full load state at 95w, which is 4.2GHz as per Gamers Nexus (and others; I credit them because they were the first to highlight the functionality properly).
The issue that you are encountering, which is effectively ignoring Turbo Boost in favour of being automatically overclocked above TDP, is not a function of Intel's marketing. It is down to mobo makers ignoring Intel spec.

 

[jpiniero]

They should have handled this TDP conflict the same way they handled XMP memory: make it a switch in BIOS

It is a switch. Just that on most Z boards 210W/unlimited is enabled by default.

 

[coercitiv]

this serious moonlight.

I wasn't joking around as much as it may seem: the order of events, power figures, timings, are for real. The C7 sleep state is non stock behavior though, mobo makers usually disable it on Auto to earn whatever performance they can. (which is a pity, watching a 6-8C high performance CPU idling at ~2W is a thing of beauty).

 

[coercitiv]

It is a switch. Just that on most Z boards 210W/unlimited is enabled by default.

My comparison was XMP, a switch that needs to be enabled, otherwise you're stuck with JEDEC timings.

 

[jpiniero]

My comparison was XMP, a switch that needs to be enabled, otherwise you're stuck with JEDEC timings.

The difference is that XMP is overclocking, while increasing PL2 is technically not.