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

[french toast]

I'm failing to see why anyone with sky lake should even consider these new cpu

Games are held back by gpu and there is no ipc benefits and they run hot and thirsty for zero benefit and the extra cores do nothing.

What exactly am I missing.

Sent from my SM-N960F using Tapatalk

You could drop the frequency down to 4.5ghz and get brilliant efficiency and still great performance, but would be slower in games than a 8700k...hell just buy an 8700k if you want the best gaming performance, just as good really.

 

[DigDog]

See what i mean ?

This isnt incompetence - it's intellectual lazyness.

Intel manager, commercial sales dept:
"Those geeks overclock our processors to 5ghz; and they want soldered tim. Lets solder the tim and overclock the cpus ourselves - IT'S A PLAN THAT CANNOT FAIL".

So without doing any major engineering effort they get to boost "we have 5ghz out of the box" and technically they havent done anything wrong. I can imagine a few people even thinking "finally Intel realized the true potential of their own chips (tnx to the overclocking community, oh joy oh pride)".


I dont know it for a fact i just know it's true
Somewhere in the bowels of Intel there's an engineer named Bob who has the plans to the next miracle architecture but he cant get management to even look at it because they are too damn stupid. Instead they jump at these gimmick solutions because they are 10x easier to do for a company as big and bulky as Intel.

 

[Shmee]

Hey guys, the general AMD vs Intel callouts need to stop. Let's stick to the technical aspects of this CPU. Again, this is not about AMD vs intel in general, not about stocks, not about markets, not about upcoming what ifs. Lets talk about the 9th gen CPUs and their performance. Thanks.

 

[Pandamonia]

This release is basically just a desperate attempt to fend off amd as best it can. No new arch and no new features and I can't even tell if the drop in performance due to the security bugs have been fixed.

What has intel been doing for the last 5 years. Because cpu research doesn't seem to be the answer!

Whilst amd and arm are catching up fast these guys release the same cpu as in 2015 with just more cores. Isn't that just the X series parts?

So nothing new then. Move along.

Sent from my SM-N960F using Tapatalk

 

[krumme]

Its not like Intel don't know what high pl2 does to perf.
Sorry but this reminds me of the 290x testing. Amd would like the performance of the turbo but not the noise and power. AT rightfully gave them some flak for it.
Intel needs to decide if they have a 130 or 200w tdp product here.
And testers need to follow.
It's small margins nowadays between the cpu so either the power consumption is 220w and the performance is held and either all benchmarks is adapted to a lower pl2. 3% performance difference matters.
This test we see now is inconsistent and wrong. Its marketing tricks from the old days.

 

[PotatoWithEarsOnSide]

For me, it'd make more sense for Intel to sell the 9900k as a 4.2GHz base, and stipulate an accurate TDP for it. That way they can say that its base clock is at the maximum you'd get out of a 2700x, and that it has superior single threaded performance.
There'd be less complaints about being misleading, a regular Joe could buy appropriate cooling simply by following the TDP rating, and the 9900k could without doubt justify the increased price tag.
These low base clocks are just a TDP marketing gimmick. I say low, but what I mean is low relative to their listed ACT.

 

[french toast]

9900k if kept at current clocks should be;
-140w normal TDP
-220w peak TDP.

2700x should be.
-105w normal tdp
-140w? peak TDP

 

[DigDog]

Im curious as to what are the average tested clocks with a Intel stock cooler and 95w draw.

 

[jpiniero]

What has intel been doing for the last 5 years. Because cpu research doesn't seem to be the answer!

Trying to fix 10 nm. All their new designs are on that node.

 

[PotatoWithEarsOnSide]

I guess I'm also tired of the ridiculous amount of segmentation on Intel's side.
The AMD approach of auto-overclocking, with base frequencies at only 10% below that, an appropriate cooler supplied for the relatively accurately rated TDP, and a clear statement that you could achieve more with better cooling, is just more honest altogether. It is simple marketing.
Instead, we get listed specs that are actually above the minimum than you could expect if you only ever followed the TDP rating that they supply.
IMO, its not honest to list an ACT that is not achievable except if you go beyond the specs they list for TDP, especially if that ACT isn't just an instantaneous spike that is only momentary. There should be a lower time limit for how long a CPU can be at its rated ACT in order for that ACT to be legally marketable. If a CPU cannot sustain its listed ACT, when combined with a cooler that matches the rated TDP of the CPU, for a given amount of time (insert an appropriate number of your choice here, then they should be forced to lower the ACT that is listed until it does sustain it for that given time period. Sure, you can get still get spikes above the listings, but as a mimimum they should be sustainable for the given time period.

 

[coercitiv]

9900k if kept at current clocks should be;
-140w normal TDP
-220w peak TDP.

TDP by definition is an average, you can't have normal average and peak average.

Intel's power power limitations are pretty clear cut and simple:

• PL1 is set at what we refer as CPU TDP, and is the average power consumption the CPU ends with during continuous heavy loads.
• PL2 is traditionally set at 1.2-1.3 TDP, and is used to capitalize on the thermal capacity of the cooling assembly (metal mass, cooling liquid etc). In other words the CPU can use power in excess for as long as the radiator takes to heat up, after which it must default back to PL1. (either by using timers, temp sensors etc.)

The closer PL2 is to PL1 in relative terms, the more time the CPU can be allowed to stay at PL2 without exceeding TDP. The average power consumption must still be defined by PL1.

You can't have "normal" TDP and "peak" TDP. You can however have stock TDP and... surprise surprise... overclocking!

 

[jpiniero]

Even though it's technically not the default, a large percentage of QC Coffee Lake 15 W U laptops have their PL2 set to 44 W. It's not indefinite however.

 

[french toast]

TDP by definition is an average, you can't have normal average and peak average.

Intel's power power limitations are pretty clear cut and simple:

• PL1 is set at what we refer as CPU TDP, and is the average power consumption the CPU ends with during continuous heavy loads.
• PL2 is traditionally set at 1.2-1.3 TDP, and is used to capitalize on the thermal capacity of the cooling assembly (metal mass, cooling liquid etc). In other words the CPU can use power in excess for as long as the radiator takes to heat up, after which it must default back to PL1. (either by using timers, temp sensors etc.)

The closer PL2 is to PL1 in relative terms, the more time the CPU can be allowed to stay at PL2 without exceeding TDP. The average power consumption must still be defined by PL1.

You can't have "normal" TDP and "peak" TDP. You can however have stock TDP and... surprise surprise... overclocking!

Ok, then I am suggesting ' normal tdp' to be Intel's PL1 of 140w..
State on the box that can consume 220w peak..or include what ever nomenclature required.
Intel has PL2 at 220w does it not?...

 

[The Stilt]

Are the tools shown here available to the public? Specifically DTS2 and LakePM. That's some very useful data to be able to monitor. If not, do you have a recommendation for estimating CPU power draw (e.g. HWinfo CPU package)? Thanks!

"DTS2" (IRI) is a custom tool I've made specifically for ASUS boards (for my own testing purposes). It cannot be released to public because it contains confidential information of three different parties (ASUS, Intel and a third manufacturer).

Based on my experience, the Intel SVID telemetry (which is available on all boards) is generally very accurate but with certain limitations. For the SVID telemetry to be accurate, EVERYTHING must be running at stock settings. Neither the frequencies or the voltages can be altered from the stock, as it will throw off the readings. Also the load-line must be configured to meet Intel specifications. SVID telemetry is available in HWInfo monitoring.

A better solution would be using a motherboard equipped with Infineon (IR) VRM controller. Depending on the VRM configuration, the controller can measure power consumption through either DCR or RdsOn measurements which are very accurate (± 2%).
HWInfo can display the controller telemetry on ASRock, Gigabyte and MSI boards equipped with these controllers.

The CPU PL1, PL2, Tau1 and Tau2 limits are also available in HWInfo, in the main program CPU page (not the sensor view).

 

[epsilon84]

You could drop the frequency down to 4.5ghz and get brilliant efficiency and still great performance, but would be slower in games than a 8700k...hell just buy an 8700k if you want the best gaming performance, just as good really.

Slight nitpick here.. don't you mean slower than an overclocked 8700K? Because a stock a 8700K only turbos to 4.3GHz. Like the 9900K, the 8700K quickly loses efficiency beyond 4.6 - 4.7GHz, so if you want efficient gaming, you'd be wise to stick below that mark as well, even with a 8700K.

 

[Ajay]

Slight nitpick here.. don't you mean slower than an overclocked 8700K? Because a stock a 8700K only turbos to 4.3GHz. Like the 9900K, the 8700K quickly loses efficiency beyond 4.6 - 4.7GHz, so if you want efficient gaming, you'd be wise to stick below that mark as well, even with a 8700K.

3.7/4.7GHz for the 8700k or does the 8700 online turbo to 4.3 in practice?

 

[Zucker2k]

3.7/4.7GHz for the 8700k or does the 8700 online turbo to 4.3 in practice?

ACT is 4.3. Motherboard Multicore Enhancement is 4.7.

 

[alkdeixaddc]

...

 

[Spartak]

And I don't believe that either. I DO believe when Anandtech said an AIO would be at 85c full load normal, and it could not do the job with AVX both at 4.8. That makes me think that the 221 full load (Anandtech) is correct. These power number are all over the board, since they use all sorts of ways to determine that.

Where do you get this 221W Figure you keep repeating? Anandtech gave 168W all-turbo peak, and with PL2 set to 210W, 221W isnt even possible at stock settings.

 

[The Stilt]

Thanks for this informative post. In addition to PL2 being above spec, have you noticed any changes made by motherboard manufacturers to Tau1 (or PL1 and Tau2 for that matter)? This seems to imply that, at least in one case, it's way out of spec:

Furthermore, if the power falls somewhere between PL1 and PL2, do you know how long can this be maintained? It appears to be determined by Tau1 which is an averaging constant for a moving average calculation for PL1.

Edit: Attaching datasheet image to help myself visualize:

In addition to PL2 and Tau1 many motherboard vendors modify IccMax (current limit) as well.
In some cases (e.g. 256-bit workload) the IccMax limit is reached before the (raised) PL2 limit, and that results in clock reduction.

I haven't tested different Tau values too extensively, but it seems that the effective power limit is PL2 until time specified by Tau expires. Once expired the effective power limit becomes PL1 again.
IIRC Tau can be set up to 3758096384 seconds, but the motherboard manufacturers generally configure it to 28 seconds or so (instead the default 1 second value).

PL1 and PL2 can also be disabled completely, as well as most of the other limits.

 

[alkdeixaddc]

...

 

[Mopetar]

Ovens use 1000 watts, which is 6-7 times the power stock loaded 9900K uses. You'd also want to note most high end CPUs are in the "oven" range you are talking about.

I don't think it's a good comparison. You could have an oven with a lower wattage, but it would just take forever to heat. Also, ovens are about the opposite of a computer. They're designed to keep heat it, because that makes them more efficient, where as a computer is designed to vent as much of the heat it produces away from it for similar reasons of efficiency. Oven's will shut off once they hit their designated temperature and if the insulation is really good, they don't need to turn back on very often. Eventually the heat gets released from the oven after you take the food out, but you could easily generate more total heat running a high-end computer for several hours than you might cooking something in an oven.

Bringing up the iGP card? Nobody buys a 9900k or 2700x class cpu to play on a iGPU. The only benefit of the iGPU in the 9900k is it'll be dead silicone for the majority of end users who'll purchase the product....Guess it's were those 720p gaming benchmarks come into play.

No one will use the iGPU for gaming and I don't think anyone using this chip even for just productivity won't pair it with a discrete card. However, I don't think it's worthless since it means that if your GPU dies, you don't need to have a backup on hand just to use the computer. Sure, it's rare, but it's one of those things that's really nice to have when you need it.

 

[Vattila]

There are "pro-AMD" posters all over this thread with outright false claims about this chip and blatant bashing who are not being called out. Ironically, it's the "pro-Intel" poster who gets called out for refuting their claims.

Yeah. Posters expressing excessive disappointment with this chip, when you get a strong feeling the poster never even considered it for purchase — well, it is unnecessary negativism. Let's just celebrate this chip for the last hurrah it is.

Similar to what the AMD engineers did with the FX-9590 (240 W, 5 GHz), the Intel engineers obviously have been given free range to push 14nm Coffee Lake to the limits with i9-9900K. Criticise Intel marketing for high pricing, misleading TDP and competitive claims, but celebrate the engineers for what they were able to accomplish. i9-9900K is about the most concentrated piece of compute on this planet, and probably in the visible universe — engineering on the frontier of human accomplishment. Respect!

 

[coercitiv]

And the presence or absence of clock reduction from hitting or not hitting IccMax can be another source of variation in some power results in reviews.

I'll keep repeating this on the forums until it's no longer necessary: we're no longer reading CPU reviews, we're reading Motherboard & CPU reviews

 

[ozzy702]

Not much. The main part you are missing is that 6700k and 7700k are only quads and if you want more cores you need to upgrade. And what most gaming benchmarks don't show is that >4-cores has significant benefits in certain online multiplayer games notably BF-series. If you are gaming on 64-player maps, a 6-core Ryzen/I7 will beat a 7700k.

But mostly you are right.

^ This. Battlefield runs MUCH higher frame rates and is smoother on my 8700k system than on my 6700k using the same GPU playing on 64 person servers. If you want high FPS then you need at least six real cores moving forward.