I9 9900k Official Reviews from Anandtech, Tomshardware. Add your own links to others !

[LTC8K6]

The Core i9-9900K is about 10 to 15 percent slower on laptops
https://www.notebookcheck.net/The-C...to-15-percent-slower-on-laptops.355161.0.html

I expected a lot worse...

 

[rbk123]

In perusing the Anand review (good grief is it long-winded), the Handbrake performance difference of the 9900k over the closest AMD processor is pretty large. Any idea why? Is Handbrake highly Intel optimized?

 

[TheGiant]

In perusing the Anand review (good grief is it long-winded), the Handbrake performance difference of the 9900k over the closest AMD processor is pretty large. Any idea why? Is Handbrake highly Intel optimized?

AVX performance

 

[The Stilt]

In perusing the Anand review (good grief is it long-winded), the Handbrake performance difference of the 9900k over the closest AMD processor is pretty large. Any idea why? Is Handbrake highly Intel optimized?

The most recent X265 encoder gains 43.5% of performance from its AVX2 implementation.
Zen lacks the resources required to execute the code at the same rate as Haswell and newer Intel designs.
Zen 2 will address that, for the AVX2 part. X265 also has some AVX512 optimizations, which improve the performance by few more percents on Skylake-X.

 

[rbk123]

The most recent X265 encoder gains 43.5% of performance from its AVX2 implementation.
Zen lacks the resources required to execute the code at the same rate as Haswell and newer Intel designs.
Zen 2 will address that, for the AVX2 part. X265 also has some AVX512 optimizations, which improve the performance by few more percents on Skylake-X.

Interesting, thanks. Looks like no AVX512 for Zen 2.

Big dropoff from the 9900 to the 9700 and then another big dropoff to the 8700, though.

 

[epsilon84]

Interesting, thanks. Looks like no AVX512 for Zen 2.

Big dropoff from the 9900 to the 9700 and then another big dropoff to the 8700, though.

,

Seems reasonable considering encoding takes advantage of all available cores/threads.

In terms of MT throughput, 9900K is a lot better than 9700K due to HT.

9700K is faster than 8700K partly due to clockspeed differences of 4.6GHz vs 4.3GHz all core turbo. 8C/8T should also slightly outperform 6C/12T, even at identical clockspeeds, so there is that as well.

 

[epsilon84]

https://www.gamersnexus.net/guides/3389-intel-tdp-investigation-9900k-violating-turbo-duration-z390

A look at why we potentially see all the fluctuating power consumption numbers across a range of different boards and reviews.

Basically most mobo manufacturers ignore Intel's PL1/PL2 spec guidance and allow unlimited 4.7GHz turbo boost when in reality a '95W' TDP would allow approx 30 sec of 4.7GHz all core turbo before settling at 4.2GHz for sustained ~95W power draw. You do lose approx 10% performance as a result of this though, render times go from 3.8 minutes to 4.2 minutes. Still, not a bad tradeoff if performance/watt is your thing.

Left unchecked and not bound by TDP limits, a 9900K @ sustained 4.7GHz can pull up to 150W in Blender, and 190W at 5.0GHz 'MCE' all core turbo.

This makes me think that a 9900 'non K' that adheres to the TDP spec (in a properly configured motherboard) and with slightly lower boost clockspeeds would be a great addition to the 9th gen lineup, particularly for those that don't intend to overclock or use lower end mobos with 95W power limits like the B360 mobos, for example.

 

[PotatoWithEarsOnSide]

I guess what bugs me about that though is that if the sustainable ACT is 4.2GHz, the fact of it being sustainable within the 95W TDP would have me questioning why they didn't set 4.2GHz as the BaseClock. You could segment for lower TDP ratings with the same chip design, using locked lowered speed variants.
At present, ACT represents a speed that can be sustained for 28 seconds on all cores.
It gripes me that CPUs get described as very overclockable when in actual fact what is happening is simply that the base clock is intentionally set lower than it seemingly needs to be.

 

[epsilon84]

I guess what bugs me about that though is that if the sustainable ACT is 4.2GHz, the fact of it being sustainable within the 95W TDP would have me questioning why they didn't set 4.2GHz as the BaseClock. You could segment for lower TDP ratings with the same chip design, using locked lowered speed variants.
At present, ACT represents a speed that can be sustained for 28 seconds on all cores.
It gripes me that CPUs get described as very overclockable when in actual fact what is happening is simply that the base clock is intentionally set lower than it seemingly needs to be.

Yeah that is true, GN actually shows performance and power figures with 'turbo off' which is basically 3.6GHz base clock and the power draw is 65W, which makes me think the base clock is practically useless. Maybe an artificially high workload like small FFTs in Prime95 can force 95W even at such low clocks, but in reality 4.2GHz should indeed be the 'base clock', not 3.6GHz.

CPUs these days are practically pre overclocked out of the box though, just unofficially. They call it 'turbo boosts' or whatever fancy marketing term but in essence that is automatic overclocking over the 'base clock'.

 

[PotatoWithEarsOnSide]

Many of the clear positives of the 9900k got drowned out at release precisely because of the reasons given in that GN article. To avoid the negative feedback in the future they really need to be stricter about power limits are enforced.
I guess the take back from it all is that AMD's XFR2 is a clear design win, and something that Intel should be looking at themselves.
"at spec, this is what you'll get, but if you have a better cooling solution then you'll get more, and the technology will automatically calculate that for you."
I suppose you could argue that it takes the fun out of overclocking, but if it offers more out of the box performance to more people, then it is better for the branding.

 

[jpiniero]

Yeah that is true, GN actually shows performance and power figures with 'turbo off' which is basically 3.6GHz base clock and the power draw is 65W, which makes me think the base clock is practically useless. Maybe an artificially high workload like small FFTs in Prime95 can force 95W even at such low clocks, but in reality 4.2GHz should indeed be the 'base clock', not 3.6GHz.

The IGP I imagine draws a bit if enabled. You have to also account for some process variation as well.

 

[LTC8K6]

Intel's Thermal Velocity Boost is probably part of all the Coffee Lake chips, but just not enabled on the desktop chips.
I expect it works pretty well and could be expanded.

 

[dlerious]

Hardware Unboxed re-tested the 9900K with a 95 watt TDP.
https://www.youtube.com/watch?v=kmAWqyHdebI

 

[Panino Manino]

This 95W controversy made me think.
If at 95W the i9-9900K is unable to hit the public published clocks, why this is not a 125W or whatever W necessary? Why Intel insisted on 95W in this case?

 

[dlerious]

This 95W controversy made me think.
If at 95W the i9-9900K is unable to hit the public published clocks, why this is not a 125W or whatever W necessary? Why Intel insisted on 95W in this case?

I don't know that it doesn't. Intel says 3.6 Base and max turbo of 5GHz. I believe I've seen some reviews where it does hit 5GHz on 1 (and possibly 2) cores while using less than the 95W. Found another video from him where he looks at total package power using Intels multiplier table. It stays under the 95W through 4-cores. Starts around 6:15 in.
https://www.youtube.com/watch?v=XfGz22ZjeGk

Edit to add, I've always thought Intels' TDP rating was for base clocks.

 

[UsandThem]

This 95W controversy made me think.
If at 95W the i9-9900K is unable to hit the public published clocks, why this is not a 125W or whatever W necessary? Why Intel insisted on 95W in this case?

Ian just did an article on this question yesterday. Let's say the reason why is complicated:

https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

 

[jpiniero]

Edit to add, I've always thought Intels' TDP rating was for base clocks.

It is. Violating the TDP is how Turbo Boost 2.0 works and has worked since it was introduced.

 

[Sable]

if motherboard makers run Intel CPUs over spec there's literally NOTHING Intel can do about it.

What control do they have over it? There's literally nothing they can do to prevent motherboard makers bypassing the 95W TDP number for their own CPU because TDP isn't a real number.

 

[jpiniero]

What control do they have over it? There's literally nothing they can do to prevent motherboard makers bypassing the 95W TDP number for their own CPU because TDP isn't a real number.

Intel I'm sure is encouraging it.

 

[Panino Manino]

I don't know that it doesn't. Intel says 3.6 Base and max turbo of 5GHz. I believe I've seen some reviews where it does hit 5GHz on 1 (and possibly 2) cores while using less than the 95W. Found another video from him where he looks at total package power using Intels multiplier table. It stays under the 95W through 4-cores. Starts around 6:15 in.
https://www.youtube.com/watch?v=XfGz22ZjeGk

Edit to add, I've always thought Intels' TDP rating was for base clocks.

Ian just did an article on this question yesterday. Let's say the reason why is complicated:

https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

So Intel only guarantees 3.6Ghz?
Even thought they sell the ability to hit 4.7GHz they say NOTHING about how much power the processor will need to hit those clocks? For what I saw on GN to stay inside the 95W this CPU needs to lower the clocks to 4.0GHz on all cores, a bit disappointing.
As a customer I wish Intel would tell how much their CPUs can consume and dissipate to operate on Turbo.

 

[epsilon84]

So Intel only guarantees 3.6Ghz?
Even thought they sell the ability to hit 4.7GHz they say NOTHING about how much power the processor will need to hit those clocks? For what I saw on GN to stay inside the 95W this CPU needs to lower the clocks to 4.0GHz on all cores, a bit disappointing.
As a customer I wish Intel would tell how much their CPUs can consume and dissipate to operate on Turbo.

In practice a heavy AVX load under a 95W TDP adherence will yield sustained clocks of 4.2GHz, which isn't that bad if you consider a 8700K can only do 4.3GHz under the same power envelope and has 33% fewer cores.

At default TDP the 9900K is actually a very power efficient CPU (better than 2700X performance for 10% lower power draw) but the problem is that Intel wants a bigger performance lead due to the much higher price, plus mobo manufacturers want to look 'as good' as their competitors that also pegs the 9900K at full 4.7GHz boost, so you end up with the current situation...

 

[TheGiant]

Hardware Unboxed re-tested the 9900K with a 95 watt TDP.
https://www.youtube.com/watch?v=kmAWqyHdebI

that review scope is under the usual quality of techspot/HU. https://www.techspot.com/review/1744-core-i9-9900k-round-two/ IMO because of choosing only extremes with 9900K that is not giving the full info like their other articles do
when I look at the performance, with 95W it is the 2700X level and that is a nice achievement from AMD
but the second look- at the power consumption of 2600X not 2700X, so AMD follows their 2600X TDP(95W)
but 2700X should consume a 10W more ideal (105W TDP) and with some board losses lets say 12W more and it consumes 30W more
they should retest it IMO with 2700X limited to 105W
and for 9900K there is a need to test something in the middle, not just 95W which cuts the CPU considerably but lets say 125 and 150W how they perform

 

[PotatoWithEarsOnSide]

TDP is a measure of the amount of power that a cooling solution needs to dissipate in order to maintain a set temperature.
The GN article showed that a 9900k could be rated at 65W TDP if it was locked at its base clock of 3.6GHz, and no turbo.
The same article also shows that it'll run 24/7 at 4.2GHz within a 95W TDP limit.
It's ACT is listed as 4.7GHz, which means that it can maintain that speed for up to a set amount of time. The Intel specs say that this can be set at any number up to 100 seconds, though the GN article saw it capping out at 24 seconds. IIRC, The Stilt mentioned maybe a month ago that 28 seconds was the typical upper limit for an ACT.
What happens during that time is that the cooling solution has enough spare capacity to dissipate more heat, so additional power is drawn until any further ACT would heat the CPU beyond the point that the cooling solution could handle. At this point the CPU drops back to 4.2GHz, which it can maintain indefinitely.
This is why in shorter tasks the 95W limited 9900k still performed close to the unlimited 9900k, also why it falls well behind it for longer tasks; the unlimited 9900k will run at 4.7GHz, and heat your room, whilst the 95W limited 9900k will run at 4.2GHz.
The 11% difference in clock speed is pretty much in line with the performance hits you get by sticking to the 95W limit.
As you can imagine, a 9900k running at 4.2GHz 24/7 is not going to be massively ahead of a 2700x that is also automatically running very close to the same clockspeed. In some cases it falls behind even, though for gaming it still beats the 2700x.
If you want to run your 9900k within its proper TDP, you are effectively limited to just the Asus Maximus XI Hero, with XMP set to On, and the other enhancement Off. All other motherboards, in all of their configurations, will perform outside, and above spec. If you know and understand that, and are prepared to research your build properly, then the 9900k can be a good buy at its price point. However, for maybe 50% of 9900k users it will be doing stuff that they might not expect, and these guys would likely be better off with a 2700x, if only because the 9900k with Asus Maximus XI Hero spec is their expected behaviour, and it's performance at that level cannot justify the price premium.
Caveat emptor.

 

[coercitiv]

that review scope is under the usual quality of techspot/HU.

but 2700X should consume a 10W more ideal (105W TDP) and with some board losses lets say 12W more and it consumes 30W more
they should retest it IMO with 2700X limited to 105W

If this is the only complaint you have against that review, you may want to know this is AMDs fault, not TechSpot's.

AMD has a different way of defining TDP which I kinda despise as it relates to heatsink performance and temp delta to define a thermal dissipation rating that completely ignores electrical power usage. It may be a correct engineering solution, but this spec is one advertised to consumers and should be better anchored with this requirement. Intel's definition is far more easy to understand, consistent across their entire product stack and across previous generations (when properly implemented and obeyed).

To keep things short, an AMD CPU can have different TDP rating depending on the cooler spec used to calculate that TDP. Since both 2600X and 2700X come with boxed cooler, and since the Wraith Prism that comes with 2700X is clearly superior to Wraith Stealth.... I think you see where this is going.

 

[PotatoWithEarsOnSide]

The whole point is that you are guaranteed to get certain performance based upon the rating of your cooling, and that using better cooling may give better performance.
As a consumer that is much easier to understand than saying it'll use X amount of power, but what cooler you'll need to buy is your own business.
For the most part, TDP does closely match power draw, because 24/7 sustainable is more reflective of actual electricity cost calculations than thousands of short bursts interspersed with idle time; if I run my PC for 4 hours a day, I can expect to use 0.38kwh of electricity with a 95W rated CPU, maximum. Obviously there's other stuff drawing power inside my PC too, but since we're only talking about the CPU here then that's all I'll discuss.

The issue here is regarding whose responsibility it is to ensure that your CPU runs at spec. If it is Intel's then they really ought to list the 9900k as a 4.2GHz base clock 95W TDP CPU. If it is the motherboard manufacturers' then out of the box performance should equal Intel's recommended spec, which is 3.6GHz base, PL1 95W, PL2 119W, Tau up to 100 seconds, Turbo up to the listed specs.