Intel Core i9-9900K 9th Generation CPU Review @ [H]

8C/16T without sacrificing the per core performance of the 8600K.

It seems like the holy grail of CPU's if not for one small detail...

The fact that it only has a gimped number of PCIe lanes (16) takes this from a "must buy immediately" product to a "no interest under any circumstance" product.

I just can't bring myself to buy any CPU/motherboard combo with fewer than 40 lanes. I currently have 40 lanes on my i7-3930k and Asus P9X79 WS, and I actually use most of them.


View attachment 113231

Slot 1: GPU (16x)
Slot 2: No slot, and covered by GPU
Slot 3: Empty (if populated, slot 1 would drop to 8x)
Slot 4: 10G Base T NIC (4x)
Slot 5: Creative Titanium HD Sound Card (1x)
Slot 6: 1TB Samsung 970 EVO with PCIe Adapter(4x)
Slot 7: 400GB Intel SSD 750 (because it is has a boot ROM) (4x)

Well, I at least I use 29 of them.

Just out of curiousity, what do you use the 10G NIC for?
 
Based on the temps and the lack of overclocking headroom I'm wondering how long Intel was sitting on these things testing them in the hope that they didn't pull a repeat of the P3 1.13Ghz fiasco. I only say this since it looks like the CPUs are literally at the max out of the box.
 
Most excellent!

BTW, a fun overclock test on the i9900K could be to disable hyper threading and compare it to a 9700K. They should be equal, but maybe the i9900K is of a higher bin quality, giving a higher OC?

Compared side by side like this, the 9900K should still come up on top by a smidge due to the higher per-core frequency mapping. However if you manually locked them both to, say, 5GHz and set them loose they'd be nearly identical. What I'd be curious about is whether disabling HT on the 9900K would lower the monster power draw relative to the 9700K at the same speed.
 
I miss those days, the tech treehouse was a lot smaller then.

Every day it was leapfrog in the CPU market, too. The last year and a half have been great. Almost like the old days. 6+ years of no competition sucked ass.

Intel really pushed the envelope on this one. But didn't push it too far - they learned their lesson last time around. I'm legitimately impressed by both AMD and Intel on this: they are maxing out their respective architectures WITHOUT crossing the line. It's pretty damned awesome - except that it's starting to seem like overclocking by hand is becoming near-obsolete. At least, if you're a normie and not [H]ard, anyway.
 
Every day it was leapfrog in the CPU market, too. The last year and a half have been great. Almost like the old days. 6+ years of no competition sucked ass.

Intel really pushed the envelope on this one. But didn't push it too far - they learned their lesson last time around. I'm legitimately impressed by both AMD and Intel on this: they are maxing out their respective architectures WITHOUT crossing the line. It's pretty damned awesome - except that it's starting to seem like overclocking by hand is becoming near-obsolete. At least, if you're a normie and not [H]ard, anyway.

Well, overclocking was about improving performance for little to no extra cost. These days you can spend a significant percentage of the initial cost and wind up with performance increase in the margin of error or even lower than stock sometimes. I can't help but think that at some point soon, overclocking may be only for the bored.
 
What I'd be curious about is whether disabling HT on the 9900K would lower the monster power draw relative to the 9700K at the same speed.

When you get down to it, power draw comes from the CPU doing work- so if those extra threads find themselves loaded with work, power draw will increase, but so should performance. Same with keeping them unassigned.
 
Hardware Unboxed answers this:



As always, Steve is overly nice about it, but using a 4 phase VRM on an 8 core processor that turbos up to 4.7 ghz just screams garbage tier to me.


So are we saying that the ASUS Maximus XI Hero is actually a 4 phase VRM? Would this also hold true for the ASUS Maximus X Hero VRM?

It is mentioned in the comments section that the 4 phases are "doubled." Does this equate to a legit 8 phase VRM?

It is also mentioned that the ASUS Maximus XI Hero may have a 95W TDP enforced as default. Something regarding "-SVID behavior: "Intel Fail-Safe" / Intel's default" With this disabled it is said that the issue is resolved. Can anyone back this up? Does this mean the VRM setup of the ASUS Maximus XI Hero is not a limiting factor for the 9900k regardless of its physical makeup?
 
So are we saying that the ASUS Maximus XI Hero is actually a 4 phase VRM? Would this also hold true for the ASUS Maximus X Hero VRM?

It is mentioned in the comments section that the 4 phases are "doubled." Does this equate to a legit 8 phase VRM?

It is also mentioned that the ASUS Maximus XI Hero may have a 95W TDP enforced as default. Something regarding "-SVID behavior: "Intel Fail-Safe" / Intel's default" With this disabled it is said that the issue is resolved. Can anyone back this up? Does this mean the VRM setup of the ASUS Maximus XI Hero is not a limiting factor for the 9900k regardless of its physical makeup?

It seems that the 9900k has an automatic Intel-supported MCE of sorts for the Z390 boards. I guess this motherboard does not default to that settings.

Bottom line is that clocks should always be monitored before making temp claims. There is NO WAY an 8 core can run cooler than the same 6 core at the same freq. This is similar to what happened to the 8th gen when some were getting the the opposite effect not knowing that motherboard-supported MCE was active and running above spec.

The easiest way to do this is to just run CB. At 4.7 ghz, this chip should be hitting around 2050. If it is getting far below that, then it is not running at turbo specs.
 
OC3d is laughable for sure. He starts off by saying all cores are running 3.7 ghz stock. Yeah, maybe with an H310. On the website, it shows the 2700x was matching the 9900k stock for stock in heavy workloads like Blender where the 9900k should have close to a 20% lead. No wonder stock temps were so low. He was probably throttling down to 4.3 ghz in those workloads.
 
OC3d is laughable for sure. He starts off by saying all cores are running 3.7 ghz stock. Yeah, maybe with an H310. On the website, it shows the 2700x was matching the 9900k stock for stock in heavy workloads like Blender where the 9900k should have close to a 20% lead. No wonder stock temps were so low. He was probably throttling down to 4.3 ghz in those workloads.
We run "optimized defaults" when we go into these and see how the CPUs act. ASUS, MSI, and GBT all run its optimized defaults out of Intel spec. The way I see it, default settings is the way we are going to test and move forward from there. I don't see any HardOCP readers tuning things to run slower.
 
Was hoping for better overclocking. As expected though, I guess.

Yeah, I think they're running on the ragged edge of what the current process can do. I was hoping for more. My decrepit 5930k is still chugging along at 4.5, no worries.

Given the price and intended market segment I'm surprised this has an iGPU. Intel GPUs are the appendix of gaming rigs. Dunno why they bother.
 
Given the price and intended market segment I'm surprised this has an iGPU. Intel GPUs are the appendix of gaming rigs. Dunno why they bother.

It's a consumer CPU- and iGPUs have many uses, even for enthusiasts. I wish all Ryzen CPUs came with iGPUs...
 
Well, according to Intel these are 95W CPUs, so any set-up using more than that should be defined as "overclocking". (With the stock settings being "automatic OC" if they allow an average power draw >TDP.)
Some motherboards don't support CPU power draw way above TDP.

The i9-9900K is thus an okay overclocker going from all cores 4.3 GHz @ 95W to 5.14 GHz @ 227W. That's a 20% speed increase at "only" 140% more power drawn...
Yep.
 
Didn't see it in the review, how does it handle ram OC. Any change over the 8xxx series?
 
Seems like a kick ass CPU to me. It's the best desktop CPU and they are pricing it that way. I'm actually impressed what they have been able to do on 14mn. For those that wanted high clocks, great IPC, AND 8 cores... here it is.
 
Would be interesting to see if it can do 4266mhz plus without issue. It's a hassle to get that speed with the 8700k without alot of tweaking of the mem settings to where people just run it slower at 3800mhz at higher timings.
Best I can do on a quick try while having about 4 other irons in the fire this morning.

memory.png
 
Not bad but the higher timings and 2t command rate help.

Wish I could lend you my set to test.

19,19,19 timings.

https://m.newegg.com/products/20-232-497?ignorebbr=true

See this thread where they talk about getting 4266mhz and the timings.

https://rog.asus.com/forum/showthread.php?90989-Maximus-IX-Apex-XMP-4266MHz-Issue/page4
You do understand that I literally just threw a set of ram in while I was doing 10 other things to try and get you a number back, right?
 
Yup.. and I'm just telling you and everyone else reading this that the timings you used were relaxed and the 8700k would have no problem using the same. ;)
Thanks, I had no idea.
 
Yup.. and I'm just telling you and everyone else reading this that the timings you used were relaxed and the 8700k would have no problem using the same. ;)

I doubt many here really cared about the crazy ram speeds which have no practicality. Kyle was just doing a quick test as a favor. Quit being so damn nitpickey.
 
Just saw this video from hardware unboxed, looks like Linus and OC3d were using the 370 boards with 4 vrm phase power and possible thermal limits. A little late to the game, but while technically their numbers are accurately reported for their setups, that isnt the full story with the cpu.

 
If you're a streaming and are rocking an RTX card you'll get perfectly excellent results with NVENC.

Not sure the value prop here beyond epeen megahurts. It is the fastest benchmark running CPU currently available. It's also way over priced.

I'll just say this: the benchmarks posted show the CPU in it's best light. But who the hell will run games at "low" on this CPU? Load up a modern title at 1440p high/ultra and let's see the gains.
 
I'll just say this: the benchmarks posted show the CPU in it's best light. But who the hell will run games at "low" on this CPU? Load up a modern title at 1440p high/ultra and let's see the gains.
It is almost like we explained this in our review. That said, since you mention RTX cards, we also did point out in our review where we are seeing CPU limitations in some engines even at 1440p at high IQ settings. So the goalposts are moving somewhat.
 
Just saw this video from hardware unboxed, looks like Linus and OC3d were using the 370 boards with 4 vrm phase power and possible thermal limits. A little late to the game, but while technically their numbers are accurately reported for their setups, that isnt the full story with the cpu.


I hope this means my 10-phase Max Hero X Z370 will fare ok? I’d be pissed if I need to upgrade to a Z390 board...
 
I hope this means my 10-phase Max Hero X Z370 will fare ok? I’d be pissed if I need to upgrade to a Z390 board...

My asus apex board has 10 digital phase and dual 8 pin CPU connector.. world record was reached with this board with a 8700k at over 7ghz (not me) so thinking my board would pass hehe..

Though I'd doubt I'd upgrade my 8700k at 5.2ghz on all cores to the 9900k unless I see it for around $300.. doubt it.
 
When you get down to it, power draw comes from the CPU doing work- so if those extra threads find themselves loaded with work, power draw will increase, but so should performance. Same with keeping them unassigned.

The 9900K has a larger L3 which helps increase the power draw as well. I'd love to see a breakdown of current draw for various areas of the CPU die on one of these things, but it's well beyond the scope of any enthusiast website I suppose.
 
Lots of Watts, LOTS of heat....take a break...wasn't that AMD's strategy to somehow keep pace once they fell behind Intel ?

Now it's vice versa.

Intel, how does it feel to chase instead of lead ?? Haha, serves you right !

I am about to toss my Z370 board, either with the CPU and get me a 2700x or just a new board. Actually I am fed up with Intel's BS chipsets.
 
Now it's vice versa.

Intel, how does it feel to chase instead of lead ?? Haha, serves you right !

It really isn't. Clock the 9900k equal to an R7 2700X, and it'll still be faster- while pulling less wattage and pumping out less heat.

The Dozers were pulling more wattage and pumping out gobs of heat in the process, and they were still woefully behind.
 
It really isn't. Clock the 9900k equal to an R7 2700X, and it'll still be faster- while pulling less wattage and pumping out less heat.

The Dozers were pulling more wattage and pumping out gobs of heat in the process, and they were still woefully behind.

You might be right, I never owned one of those.

I still think Intel smells the "Gosh, we are falling behind". They never were pushed that hard, now they give you Soldered TIM, extra cores, extra Watts, all those things Intel tried to avoid for years while they could afford it.

Now they cannot afford it anymore and hand you a CPU that seems to be "different" to what we are used to from Intel, hardly any oc-headroom, cooking dies despite solder, watts east of eden.


It just smells very very fishy to me, I dunno, for me, Intel lost the lead.

The one who's calling now is AMD, and Intel has to struggle to give answers that match. No, Intel is already on red alert, otherwise you wouldnt see a thing like a 9900k or Z390, not in 2018.
 
I still think Intel smells the "Gosh, we are falling behind".

They were getting close to AMD being 'more or less' parity using a new architecture on a new process, so they pushed their 14nm process to the limit to maintain their lead.

Note lead. And if AMD were actually leading, the product would be more expensive than the 9900K, going by the last time they were leading.
 
They were getting close to AMD being 'more or less' parity using a new architecture on a new process, so they pushed their 14nm process to the limit to maintain their lead.

Note lead. And if AMD were actually leading, the product would be more expensive than the 9900K, going by the last time they were leading.

This is important. AMD caught up in a hurry with Zen and Zen+. Intel couldn't afford to be lazy any more. They had to go balls to the wall to stay ahead. That's what the 9900k represents. I imagine the upcoming 28 core HEDT part will be likewise - balls to the wall.

Intel maintains an architectural lead (it's important to realize this - Zen can only win when price is also factored in to the equation), but it is very small. Small enough that AMD can play pricing games and show good value at various price tiers. Small enough, too, that Zen 2 could potentially reach parity or even obtain the lead. Potentially. Maybe. Intel knows all this, that's why the 9900k exists when, a year and a half ago, all we had was a 7700k ;).
 
This is important. AMD caught up in a hurry with Zen and Zen+

Zen+ is still behind Skylake in terms of both IPC and clockspeeds- AMD has made up a lot of lost Dozer ground, but what they've caught up to (say, Haswell) is still five years behind. Intel getting stuck at 14nm is the main reason they've gotten so close, and good on AMD, but the facts still put a lot of work on their shoulders to 'catch up' ;).
 
Zen+ is still behind Skylake in terms of both IPC and clockspeeds- AMD has made up a lot of lost Dozer ground, but what they've caught up to (say, Haswell) is still five years behind. Intel getting stuck at 14nm is the main reason they've gotten so close, and good on AMD, but the facts still put a lot of work on their shoulders to 'catch up' ;).

Haswell-ish in latency workloads, Broadwell-ish in throughput. But the relatively cheap core scaling (compared to Sky/Kaby/Coffee Lake) CCX/multi-die model makes up for that deficiency a little.

For clockspeed, I think GloFo is holding them back. Switching to TSMC should fix. For IPC... AMD has a ways to go still. Zen 2 needs to address this deficit. That will be key to seeing if AMD can take the lead or reach parity.

And agreed that Intel being stuck at 14+++++++++ nm (how many pluses now?) is the reason why AMD can even pull off as much as they have.
 
From TechPowerUp: i7-9700K "fastest gaming processor"

Across the 720p and 1080p gaming tests, the Core i7-9700K ends up faster than even the Core i9-9900K, for the reason I explained above. The lead is rather slim, around 1-2 percent, but it's there. Even in GPU-limited resolutions such as 1440p and 4K UHD, the i7-9700K ends up a tiny bit faster, but the differences are insignificant. Games such as "Civilization VI" that are heavily multi-threaded, still show the i7-9700K ahead. It is hence safe to conclude that the i7-9700K is the fastest gaming processor - and not the Core i9-9900K.​
 
From TechPowerUp: i7-9700K "fastest gaming processor"

Across the 720p and 1080p gaming tests, the Core i7-9700K ends up faster than even the Core i9-9900K, for the reason I explained above. The lead is rather slim, around 1-2 percent, but it's there. Even in GPU-limited resolutions such as 1440p and 4K UHD, the i7-9700K ends up a tiny bit faster, but the differences are insignificant. Games such as "Civilization VI" that are heavily multi-threaded, still show the i7-9700K ahead. It is hence safe to conclude that the i7-9700K is the fastest gaming processor - and not the Core i9-9900K.​

Hawt!
 
Back
Top