Review Intel Core i9-10900K Review: Ten Cores, 5.3 GHz, and Excessive Power Draw

[lga1156_ftw]

"Little overclocking headroom", well that should be added to every single AMD cpu ever sold aswell if its a negative

Looking at ryzen overclocking videos it seems like a joke since most people can get is 1-5 percent more, as for intel its more like 10-40%

Otherwise its well written review.

 

[st379]

Meanwhile....back on planet Don't bother me with facts.
https://hothardware.com/reviews/intel-core-i9-10900k-and-i5-10600k-benchmarks?page=4
The 10900k draws less power than even the 3700x on idle and on one core at 5.3Ghz

https://www.phoronix.com/scan.php?page=article&item=intel-10500k-10900k&num=10
And the 10900k is faster than the 3900x,a CPU with 20% more cores.
While the average power draw of the 10900 is actually exactly at 124W.
Yes it will peak much higher but that's the whole point of velocity boost and turbo,to boost as high as possible if overall power draw allows.
That's why you can disable all that crap, if you want to stay below a certain TDP you can.

Idle watt? lol are you running your pc on AA batteries?
I love how you ignore the fact of the "small" 60 watt different between the 2 cpu (228 vs 286 watt).
According to the data that you supply there is 3% difeerent in favor of the i9 for extra 100 dollar for the cpu, lets say a good air cooler like noctua for extra 100-80$ (depends on your country, in my country it is more), you really don't want to cheap out on the motherboard so at least 100$ difference there too (if not more) and in total every 1%=100$ of additional cost.
That's nice, not to mention the cpu is out of stock and ryzen 4000 coming in september-october.
It is look at the moment like the 9900ks, a paper launch that Intel want for the ryzen 4000 to be compared.
And according to data you supply the peak is 380 watt.
Like you said "back on planet Don't bother me with facts".

 

[RodroX]

Zen 3 IPC increase, which was about 15%~17% according to AMD "official" numbers, seems to be in fact much higher, man 2020 is going to be a very intersting year CPU launching wise.

 

[st379]

Zen 3 IPC increase, which was about 15%~17% according to AMD "official" numbers, seems to be in fact much higher, man 2020 is going to be a very intersting year CPU launching wise.

Just the unified ccx should put it on top in gaming if you look at the 3300x review. No clock and IPC imrovement is needed but of course it is welcomed.

 

[InvalidError]

Just the unified ccx should put it on top in gaming if you look at the 3300x review. No clock and IPC imrovement is needed but of course it is welcomed.

In all likelihood, Zen 3's monolithic CCD gains are likely built into AMD's ~17% IPC gain projection.

 

[st379]

In all likelihood, Zen 3's monolithic CCD gains are likely built into AMD's ~17% IPC gain projection.

The IPC is not for single core? I think you can't really measure it in multi core scenario because of different architecture, like Intel do in skylake with lower latency although zen 2 ipc is higher.

 

[jeremyj_83]

Meanwhile....back on planet Don't bother me with facts.
https://hothardware.com/reviews/intel-core-i9-10900k-and-i5-10600k-benchmarks?page=4
The 10900k draws less power than even the 3700x on idle and on one core at 5.3Ghz

That is showing Total System Power not CPU Cores or CPU Package power. Since we have been talking about CPU power draw that graph is irrelevant.
Actual Power Draw

CPU​	Ryzen 3700X​	Ryzen 3900X​	i9-9900k​
Power (Cores), Idle	0.33W	0.3W	2.61W
Power (Cores), 1T	13.16W	16.75W	30.53W
Power (Cores), 4T	31.94W	34.97W	59.31W
Power (Cores), 8T	57.2W	62.6W	87.57W
Power (Cores), Full Load	74.59W	134.61W	164.67W
Power (Package), Idle	16.68W	14.02W	7.79W
Power (Package), 1T	34.57W	39.72W	35.52W
Power (Package), 4T	52.35W	53.07W	63.08W
Power (Package), 8T	74.34W	77.95W	91.89W
Power (Package), Full Load	90.26W	141.36W	168.48W

https://www.anandtech.com/bench/product/2520?vs=2263
https://www.anandtech.com/bench/product/2519
From that we can see that the cores on the Ryzen CPUs draw FAR less power than the i9-9900k. While the 10th Gen isn't available in Ananadtech's Bench, the differences will be minimal due to same process tech and CPU core. In actuality they will probably be higher with 10th Gen due to the higher clock speeds. When looking at the full package, we see that the Ryzen starts to draw more power because of the IF. However, it doesn't take long for the 9900k to shoot past it. Even the 3900X with its 105W TDP is more efficient. Where it is using less power for both package and cores at 24t than the 9900k uses with 16t.

 

[InvalidError]

The IPC is not for single core?

When AMD and Intel speak of IPC improvements in % figures, they are clock-for-clock comparisons with other CPUs in some selection of workloads such as Cinema4D and will reflect the sum of all performance improvements such as improved multi-threading, increased number of execution ports, increased number of ports that support each given instruction, larger cache, faster uncore/fabric clock, etc.

 

[JarredWaltonGPU]

In all likelihood, Zen 3's monolithic CCD gains are likely built into AMD's ~17% IPC gain projection.

Is ~17% what AMD is actually saying now? Because last I remember seeing, AMD said it was a range of ~5% to ~17% with an average increase of ~12%. That was the official AMD slides from the financial numbers IIRC. AMD could have revised that upward, but 17% overall would be a huge jump. But yeah, IPC is just a theoretical value calculated from a variety of tests and reflects overall performance uplift from all areas, including cache.

 

[InvalidError]

Is ~17% what AMD is actually saying now? Because last I remember seeing, AMD said it was a range of ~5% to ~17% with an average increase of ~12%. That was the official AMD slides from the financial numbers IIRC.

Well, if you look at how the 3300X is ~15% faster clock-for-clock than the 3100 in games, it wouldn't be surprising that Zen 3's monolithic CCD would do something similar for 6-8 cores SKUs in moderately to heavily threaded latency-sensitive scenarios too.

By the same token, it would make sense that non-threaded and lightly-threaded stuff would see a much smaller benefit and be where the 5% comes from.

 

[jeremyj_83]

Well, if you look at how the 3300X is ~15% faster clock-for-clock than the 3100 in games, it wouldn't be surprising that Zen 3's monolithic CCD would do something similar for 6-8 cores SKUs in moderately to heavily threaded latency-sensitive scenarios too.

By the same token, it would make sense that non-threaded and lightly-threaded stuff would see a much smaller benefit and be where the 5% comes from.

Even an average of 12% is still a large increase. In theory that would mean that a Zen 3 CPU could be 30+% faster clock for clock over a Zen 1 CPU from 2017 in 3.5 years. We haven't seen that level of increase in such a short time in almost a decade. The last time Intel did something like that was Conroe > Nehalem > Sandy Bridge and even that was over a 5-6 year period.

 

[InvalidError]

Even an average of 12% is still a large increase. In theory that would mean that a Zen 3 CPU could be 30+% faster clock for clock over a Zen 1 CPU from 2017 in 3.5 years. We haven't seen that level of increase in such a short time in almost a decade.

Better than the last 10 years, sure. though it'd still take a whole lot more than that for me to really get excited. Back when I first got into PCs, 50% more performance per dollar YoY was considered a slow year and 70% merely typical.

 

[Gurg]

Now that the rush is over will TH be reviewing the 10700K and running its full gaming suite performance analysis so that we can compare the 10900k, the 10700k and 10600k to the full range of AMD and 9000 series CPUs in the other comparison reviews TH has published in the past couple of months?

 

[Landsharkk]

The bottleneck is wherever you want it to be. If you want it to be the CPU, reduce graphics details until it is. If you want it to be the GPU, increase details until it is. I bet I could still bottleneck a 3080Ti/S on my i5-3470: set graphics to ultra with full RTX on a 4k display and FSAA at 16X so the GPU has to render in 16k.

The CPU-GPU bottlenecking range is extremely wide thanks to how malleable graphics details are.

I'm not sure I agree with that statement.

I have a i7-3930k and a gtx 1080. Are you saying I can make the GPU the bottleneck by maxing out graphics? But then what you say suggests that even if I put in a faster CPU, I shouldn't really see any more performance gains, because the GTX 1080 is already the 'bottleneck'.


I think in reality, if you take a i9-10900k and put a GTX 2080, max graphics, you'll find a bottleneck.
Then if you take the same i9-10900k, but put in a 'yet to be released' GTX 3080, you should see a performance increase.

That's the number I'm curious about. If you compare the 10900k vs the ryzen 3950 but use the next gen GPU, which one is going to see the biggest gains there?

I'm not ready to make my final decision on a CPU when 1) it's not even released yet and 2) it has been tested with the next gen GPU's which will likely release around the same time as the 10900k.

 

[InvalidError]

I have a i7-3930k and a gtx 1080. Are you saying I can make the GPU the bottleneck by maxing out graphics? But then what you say suggests that even if I put in a faster CPU, I shouldn't really see any more performance gains, because the GTX 1080 is already the 'bottleneck'.

When you upgrade your CPU, you can turn details back down to get however many extra FPS as you want up to whatever your new CPU is capable of or the GPU can put out at whatever the lowest graphics details you can bear.

As I wrote earlier, graphics are vastly malleable, you are only limited by the minimum frame rate and minimum details you can bear.

 

[Kevinoli2005k]

Intel's Core i9-10900K cements itself as the fastest gaming processor on the planet, but that performance comes at a cost.

Intel Core i9-10900K Review: Ten Cores, 5.3 GHz, and Excessive Power Draw : Read more

Wrong timing to release a CPU that costs that much in times where the economy is crashing down on us> :/ Idk

 

[Gurg]

Wrong timing to release a CPU that costs that much in times where the economy is crashing down on us> :/ Idk

Exactly the right time to release new products that the consumer will want to purchase to help get the economy going again. As soon as Nvidia has inventory of its new 3000 series GPUs it should also start releasing them and selling them hot off the production line. Intel and Nvidia will sell everyone of their new CPUs and GPUs they can produce at least through the first quarter of next year. From what I can tell online Intel doesn't seem to have a large over hang of 9000 series CPUs that need to be sold at discounted prices and neither does Nvidia with its top line 2000 series GPUs.

In contrast despite all the gushing AMD reviews, retailers appear to be flush with AMD products that even new lower prices aren't clearing out. There are also reports coming out that AMD will be changing its CPU configuration after its upcoming release so that just like Intel, new motherboards will be required after this next release of CPUs.

Yes some consumers are hurting, but others have been working and collecting checks, with many working from home on old PCs and laptops. Lots of those want to upgrade their old computers as well as others that just want the latest and fastest gaming systems.

 

[Dwardu1055]

Bros if someone could run this chip on X-plane 11, that would be great, 5.3 ghz single core might actually be amazing for that sim.

 

[TerryLaze]

Idle watt? lol are you running your pc on AA batteries?
I love how you ignore the fact of the "small" 60 watt different between the 2 cpu (228 vs 286 watt).

I said:"The 10900k draws less power than even the 3700x on idle and on one core at 5.3Ghz"
AND :
"Yes it will peak much higher but that's the whole point of velocity boost and turbo,to boost as high as possible if overall power draw allows. "
AND
I posted another graph showing how often the power will be at ~300 and how often it will be much lower.

Maybe you just got bored reading half way through?!

According to the data that you supply there is 3% difeerent in favor of the i9 for extra 100 dollar for the cpu, lets say a good air cooler like noctua for extra 100-80$ (depends on your country, in my country it is more), you really don't want to cheap out on the motherboard so at least 100$ difference there too (if not more) and in total every 1%=100$ of additional cost.
That's nice, not to mention the cpu is out of stock and ryzen 4000 coming in september-october.
It is look at the moment like the 9900ks, a paper launch that Intel want for the ryzen 4000 to be compared.
And according to data you supply the peak is 380 watt.
Like you said "back on planet Don't bother me with facts".

Yes sure,if you need 20% more cores to do the same job you better be cheaper.

And according to data you supply the peak is 380 watt.

That's like having a gaming benchmark that only shows you the highest recorded frame,even if that was in a menu and like 400FPS.

That is showing Total System Power not CPU Cores or CPU Package power. Since we have been talking about CPU power draw that graph is irrelevant.

Sure that's a valid point if you don't care about actual power draw of your system and look at your CPU as something abstract.
(Do they have any link on how and what exactly they measured there? )

When AMD and Intel speak of IPC improvements in % figures, they are clock-for-clock comparisons with other CPUs in some selection of workloads

I very much doubt that,at least for intel they have used IPC with stock settings,so including the clock differences.
What you describe is what reviews and benchmarks do.

 

[InvalidError]

I very much doubt that,at least for intel they have used IPC with stock settings,so including the clock differences.

The clock frequency does not really matter since "Instructions per CLOCK" normalizes performance per GHz. If a CPU scores 20% higher in a given benchmark at 10% higher clock, IPC is 1.2X the performance / 1.1X the clock frequency = 9% extra performance from IPC gains.

The only thing running CPUs at the exact same frequency does is artificially decrease the faster CPU's throughput by increasing its latency on external dependencies, which makes whatever IPC figure you get out of the exercise even less representative of how the CPU would normally behave than it already was.

 

[jeremyj_83]

Sure that's a valid point if you don't care about actual power draw of your system and look at your CPU as something abstract.
(Do they have any link on how and what exactly they measured there? )

We are talking about power draw of a CPU. Total system power is NOT relevant in this case and only goes to provide erroneous data. I labeled the table with what was being measured, cores and total package. Cores is the power draw of the CPU cores only. Total package is that of the CPU and un-core. I also linked where I got the data. If you need more info on what is being measured. https://www.anandtech.com/show/14605/the-and-ryzen-3700x-3900x-review-raising-the-bar/19 Here is the 10th Gen power info. https://www.anandtech.com/show/15785/the-intel-comet-lake-review-skylake-we-go-again/5

 

[cryoburner]

In contrast despite all the gushing AMD reviews, retailers appear to be flush with AMD products that even new lower prices aren't clearing out.

Nice try, but 8 of the top 10 bestselling CPUs on Amazon are currently Ryzen processors. : P

https://www.amazon.com/Best-Sellers-Computers-Accessories-Computer-CPU-Processors/zgbs/pc/229189

Yes sure,if you need 20% more cores to do the same job you better be cheaper.

Except the 3900X is generally shown to be faster at most multithreaded workloads, meaning the 10900K isn't quite doing the same job, despite its massively higher power consumption and heat output when presented with those tasks. And while multithreaded workloads certainly aren't everything, unless someone is specifically needing more performance for heavily-multithreaded tasks, there's little point in spending extra for a processor with more than 8 cores at this point. The 3900X fulfills its niche of providing significantly more multithreaded performance than the competition at a given price point, while the 10900K costs significantly more while generally providing slightly less performance in the workloads that you would want a higher core-count processor for.

And that "geometric mean of all test results" chart you posted needs some context. That review site performs their testing on Linux, and in general I would say desktop software optimization tends to be a bit hit-or-miss on that platform, especially for newer hardware. Perhaps for the subset of people using that specific software under Linux, some of those results might be relevant, but desktop Linux isn't exactly mainstream. What's more, some of those tests where Intel shows strong performance is in Intel's own software. Oh look, the 10900K slightly outperforms the 3900X in the Intel Embree path-tracer. I'm sure that's been optimized for AMD's processors. It's nice that a site is doing hardware reviews in Linux, but those results should not be posted elsewhere as if they are representative of the performance that most people will be seeing, especially a "geometric mean" chart that's heavily influenced by software that relatively few people will be using.

 

[PCWarrior]

Except the 3900X is generally shown to be faster at most multithreaded workloads, meaning the 10900K isn't quite doing the same job, despite its massively higher power consumption and heat output when presented with those tasks. And while multithreaded workloads certainly aren't everything, unless someone is specifically needing more performance for heavily-multithreaded tasks, there's little point in spending extra for a processor with more than 8 cores at this point. The 3900X fulfills its niche of providing significantly more multithreaded performance than the competition at a given price point, while the 10900K costs significantly more while generally providing slightly less performance in the workloads that you would want a higher core-count processor for.

And that "geometric mean of all test results" chart you posted needs some context. That review site performs their testing on Linux, and in general I would say desktop software optimization tends to be a bit hit-or-miss on that platform, especially for newer hardware. Perhaps for the subset of people using that specific software under Linux, some of those results might be relevant, but desktop Linux isn't exactly mainstream. What's more, some of those tests where Intel shows strong performance is in Intel's own software. Oh look, the 10900K slightly outperforms the 3900X in the Intel Embree path-tracer. I'm sure that's been optimized for AMD's processors. It's nice that a site is doing hardware reviews in Linux, but those results should not be posted elsewhere as if they are representative of the performance that most people will be seeing, especially a "geometric mean" chart that's heavily influenced by software that relatively few people will be using.

What you actually mean is that the 3900X is faster in some very specific workloads that tech reviewers tend to overrepresent in their test suites giving the false impression that Ryzen or 3900X in particular is faster in productivity tasks in general. But if you do extensive testing like Phoronix did for Linux (and Linux has superior multithreaded support than Windows anyway) you can see that the 10900K is overall faster.

And the above tests were performed on Ubuntu Linux. If you use Clear Linux (which is made by Intel but which runs faster than pretty much any version of Linux on almost every workload for BOTH Intel and AMD) Intel’s lead is widened as Clear Linux can extract more performance out of Intel cpus than any other generic Linux version.

And if you think the above is Linux-only, you are badly mistaken. Intel’s superiority is actually even more pronounced on Windows and paid/professional software. Just most lazy reviewers, for their productivity tasks comparisons, tend to use free software/benchmarks, software that is often built with generic (and suboptimised) compilers instead with Intel's superior compiler. And even in those cases the outcome is dependent on the workload.

Anyway I compiled a list of various benchmarks.

Let’s start with Blender that most will say that AMD wins. Well put a different workload and viola stock 10900K scores a win over a PBO 3900X:

Same applies for 7zip:

Want more open source software results on Windows? Here are the results for LAME

Want even more open source? How about Google's Tensorflow (the most popular machine-learning software packages)?

How about a bit of Engineering simulations? Here s the Euler3D benchmark test - a fully parallelised workload

And here is comparison for DataBase - MySQL

And some paid software:
Agisoft Metashape

WinRar

Topaz Video Enhance AIT

Microsoft Office

Word

Excel

Power Point

Adobe Premiere Pro

Adobe After Effects

 

[Phaaze88]

Bloody hell...

 

[cryoburner]

Just most lazy reviewers, for their productivity tasks comparisons, tend to use free software/benchmarks, software that is often built with generic (and suboptimised) compilers instead with Intel's superior compiler.

You mean Intel's compiler that's often been shown to not incorporate hardware features and optimizations for processors other than their own?

Anyway I compiled a list of various benchmarks.

Maybe you should start a review site aggregating only the handful of outlier results supporting your argument while ignoring the bulk of other results. : D

No doubt there is plenty of software and workloads where the 10900K will perform a little better, whether due to compiler optimizations, limited threading, or just architectural differences favoring certain tasks, but for heavily-threaded software, it is often at a core disadvantage resulting in lower performance than the competition in many applications. Intel is at the very least offering more processor for the money than they were with the 9900K though, and their slashing prices for a given core count makes this generation a better value than their 9th-gen offerings, and a lot more competitive with Ryzen. The revised pricing makes this a relatively decent lineup, in my opinion, aside from the cooling requirements for these higher-end parts. But ultimately, the 3900X is a year-old processor that is currently priced similar to the 8-core i7-10700K, not the i9-10900K, and that's even before getting into cooling.