Article Tom's Hardware Core i9 9900KS Preview

[Zucker2k]

Power efficiency is incredible for 14nm.
Edit: Consumes 50 watts less than the i9 9900K @ 5GHz.



Link

 

[amrnuke]

It's funny how everybody tries to defend ryzen by talking about gaming at 1440 but then when it comes to power consumptions it's only about running a power virus with every power saving feature disabled...
You get a solid constant 5Ghz while gaming at even 1080 with the advertised 127W TDP.
Unlike not even hitting your single core turbo...like, at all.

In HandBrake and Blender the 9900K system draw is higher than a Threadripper 2950X, if left to its own devices, 250W, while losing to the 2950X by a wide margin. (per TechSpot)

Power under full load for the 9900K is 168W, for the 3900X it's 142W. As a reminder, the 9900K is <10% faster than the 3700X and 3900X in gaming at your critical 1080p level (and at 768p, just for kicks, just as close). (per Anandtech)

Total system draw in The Witcher 3 in gaming, 3900X and 3700X draw more total system power (3900X is 385W, 3700X is 365W, and 9900K is 356W). The Zen2 chips lose in the Witcher 3 by less than 3 FPS somehow. (per TechPowerUp)

Stock vs stock, the 9900KS uses 24W more power than a 3900X (142 vs 118) in y-cruncher with AVX on, while inexplicably LOSING to the 3900X in y-cruncher. (per Toms)

ExtremeTech ran a 3700X on several X470, X570 boards and the X570 chipset uses, it appears, 30-50W more under load than X470.

Conclusion:

The 9900K and 9900KS are the top-tier for gaming, by less than 10%. For multi-thread-heavy workloads they lose out to mid-range Zen2 chips.

For compute-intensive processes the 9900K and KS are generally less efficient than Zen2, and for gaming, they are more efficient. This of course is dependent on many factors including that nearly all benchmarks done on Zen2 also are done on X570 boards, which makes power consumption difficult to disentangle - it's almost certainly not the chips causing power draw to be so high, and I suspect power draw for X470 + 3900X would be much less than Z390 + 9900KS for similar overall performance.

 

[Zucker2k]

For multi-thread-heavy workloads they lose out to mid-range Zen2 chips.

Care to back this up with some sources?

 

[jpiniero]

The 9900K and 9900KS are the top-tier for gaming, by less than 10%.

It's more like 10-20% for the most part, after all 5 is 15% more than 4.3. Of course that depends on not being GPU limited. There's a couple titles for some reason like Far Cry 5 which can be well over 20%. Of course the 9700K continues to be the better deal.

 

[Dudler]

Care to back this up with some sources?

The Intel Core i9-9900KS Review: The 5 GHz Consumer Special

www.anandtech.com

180W of power , and not faster than a 65W 3700X ..

 

[amrnuke]

It's more like 10-20% for the most part, after all 5 is 15% more than 4.3. Of course that depends on not being GPU limited. There's a couple titles for some reason like Far Cry 5 which can be well over 20%. Of course the 9700K continues to be the better deal.

Are you seriously presenting clock speed instead of actual gaming benchmark results? In gaming benchmarks, even at the unrealistic and purely synthetic (for most) 720p, the gap is <10%. At 1080p it's <5%. At 1440p <3%. (As for individual games, there are also some titles where 3700X beats the 9900KS.)

 

[Abwx]

Stock vs stock, the 9900KS uses 24W more power than a 3900X (142 vs 118) in y-cruncher with AVX on, while inexplicably LOSING to the 3900X in y-cruncher. (per Toms)

In y-cruncher with AVX the 9900KS is slower than the 3700X...

Intel Core i9-9900KS mit 5,0 GHz im Test: Benchmarks in Games und Anwendungen sowie Preis/Leistung

Intel Core i9-9900KS im Test: Benchmarks in Games und Anwendungen sowie Preis/Leistung / Benchmarks in Spielen (Full HD und UHD)

www.computerbase.de

So much for THG "review"..

Besides the 9900KS use 50W more than a 3900X at the CPU level in Cinebench.

Max CPU power under Prime 95 is 275W and about 250-260W steadily, dunno what THG used as loads since their Aida test use less power than Cinebench despite allegedly using AVX, wich is not AVX2, isnt it..

Intel Core i9-9900KS mit 5,0 GHz im Test: Leistungsaufnahme, Temperatur und Overclocking

Intel Core i9-9900KS im Test: Leistungsaufnahme, Temperatur und Overclocking / Die Leistungsaufnahme passt zu der TDP und den weiteren CPUs

www.computerbase.de

 

[amrnuke]

Care to back this up with some sources?

TomsHardware review
Anandtech review
3900X has substantial lead overall over the 9900KS in heavily threaded tasks like rendering, encoding, compression, decompression, etc.

I have to imagine once the high-mid range Zen2 chip (3950X) and high-end Zen2 chips (TR3) are released it won't even be close.

As it stands, price for price, Zen2>14nm++++++ in multithreaded applications.

 

[BigDaveX]

TomsHardware review
Anandtech review
3900X has substantial lead overall over the 9900KS in heavily threaded tasks like rendering, encoding, compression, decompression, etc.

I have to imagine once the high-mid range Zen2 chip (3950X) and high-end Zen2 chips (TR3) are released it won't even be close.

As it stands, price for price, Zen2>14nm++++++ in multithreaded applications.

Um, I wouldn't call the 3900X "mid-range" considering it's generally the same price or even higher than the 9900K. When someone says mid-range I usually think of an i5/Ryzen 5, or the cheaper i7/Ryzen 7 chips.

 

[Ajay]

Power virus? Please.

Prime95 AVX tests are like running a power virus, basically. Games, no.

 

[ondma]

The Intel Core i9-9900KS Review: The 5 GHz Consumer Special

www.anandtech.com

180W of power , and not faster than a 65W 3700X ..

Nice, comparing actual power consumption of intel vs rated tdp for the Ryzen. Are you seriously claiming the 3700x will use only 65 watts in a stress test like prime 95?

 

[jpiniero]

Are you seriously presenting clock speed instead of actual gaming benchmark results? In gaming benchmarks, even at the unrealistic and purely synthetic (for most) 720p, the gap is <10%. At 1080p it's <5%. At 1440p <3%. (As for individual games, there are also some titles where 3700X beats the 9900KS.)

That's because some reviewers used stock memory, which hurts Coffee Lake.

Kitguru's review had the gap like this:

Ashes: 10%
Deus Ex MD: 20% / 12%
Far Cry 5: 30% / 26%
GTA V: 20% / 14%
Ghost Recon: 8% / 4%
Hitman 2: 22%
Shadow of the Tomb Raider: 24% / 25%
Divison 2: 3%

If I did the math right, of course. With the caveat that they had their 9900KS at 5.2 Ghz.

 

[amrnuke]

That's because some reviewers used stock memory, which hurts Coffee Lake.

Kitguru's review had the gap like this:

Ashes: 10%
Deus Ex MD: 20% / 12%
Far Cry 5: 30% / 26%
GTA V: 20% / 14%
Ghost Recon: 8% / 4%
Hitman 2: 22%
Shadow of the Tomb Raider: 24% / 25%
Divison 2: 3%

If I did the math right, of course. With the caveat that they had their 9900KS at 5.2 Ghz.

You say that they use stock memory like it's a bad thing. Anandtech are testing the system at maximum manufacturer recommended RAM speeds, which is a very valid, fair way to do it if you are writing an article for the mainstream. With TomsHardware they ran tests on both systems at 3600. Most other articles have them equal at 3200.

I've looked at 6 different reviews of the 9900KS and the average FPS benefit over the 3900X at 1080p is 9.3%. (KitGuru, Anandtech, Hexus, HotHardware, Tweaktown, TomsHardware).

If you trust KitGuru much more than the other sites, then more power to you. I prefer a more broad view of things.

 

[jpiniero]

You say that they use stock memory like it's a bad thing. Anandtech are testing the system at maximum manufacturer recommended RAM speeds, which is a very valid, fair way to do it if you are writing an article for the mainstream.

I don't disagree, but being practical, people buying Coffee Lake K are going to use at least 3200 memory. It's not like it won't work, Intel just limits the ram speed more because of validation (and to ensure people buy Z boards!)

 

[amrnuke]

I don't disagree, but being practical, people buying Coffee Lake K are going to use at least 3200 memory. It's not like it won't work, Intel just limits the ram speed more because of validation (and to ensure people buy Z boards!)

Absolutely agree, but several other review sites did just that and found <10% difference. KitGuru happened to find a 12% difference and 21% more power consumption on the 9900KS. It's all about what you want.

 

[amrnuke]

Um, I wouldn't call the 3900X "mid-range" considering it's generally the same price or even higher than the 9900K. When someone says mid-range I usually think of an i5/Ryzen 5, or the cheaper i7/Ryzen 7 chips.

Fair point. If we revise that, low-end is 3600/3600X, mid-range is 3700X/3800X, and high-end is 3900X/3950X. Maybe a bit skewed since we have a the 3500 and 3950 unreleased right now, so I guess smack in the middle of the current Ryzen lineup is the 3700X.

But comparing 3700X to the 9900K on TechPowerUp just for ease of it, the 3700X beats the 9900K in CB20, Blender, Corona, Keyshot, light baking, close on MS VS compilation, close on machine learning. And the 3700X is within 7% in 1080p gaming, like $150 cheaper too. I think that based on those results it's fair to say that even mid-range Ryzen desktop chips are a better buy than the 9900K for heavily multithreaded tasks.

 

[JoeRambo]

In y-cruncher with AVX the 9900KS is slower than the 3700X...

Yeah, one system is with DDR4 3200, the other is with DDR4 2666 in FP testing that are usually bound by memory BW and latency.
IT IS THE STOCK settings i hear from usual suspect, but let's be honest here, the person who buys this CPU with intention to pair it with 2666 speed mem deserves a stint in hell. The day has come when it is OK to tinker with PL1/PL2, but god forbid to touch memory speeds and latencies.

 

[Zucker2k]

I think that based on those results it's fair to say that even mid-range Ryzen desktop chips are a better buy than the 9900K for heavily multithreaded tasks.

I hope this is not another semantic mishap on your part. You've gone from a performance argument to a value one; and the 9900KS is curiously missing from your argument, all of a sudden.

 

[Zucker2k]

IT IS THE STOCK settings i hear from usual suspect, but let's be honest here, the person who buys this CPU with intention to pair it with 2666 speed mem deserves a stint in hell.

There's definitely a lot of performance left on the table, and it'll show in the real world.

 

[Dudler]

Nice, comparing actual power consumption of intel vs rated tdp for the Ryzen. Are you seriously claiming the 3700x will use only 65 watts in a stress test like prime 95?

I have never mentioned Prime95. 172W is Anandtechs numbers from running Cinebench R20.

https://images.anandtech.com/doci/14979/9900KS255W.png

What Anandtech used for the 3700X review I don't know, I dont find it listed in their review. Max package power for 3700x is 88W.

Techspot lists 164W as max systen consumption, 12W less than 8700k, which uses 86W according to Anandtech(cpu only) So between 75 to 88W for 3700x.

 

[Panino Manino]

This CPU will only be available till the end of the year, for what I understood.
I don't like how Intel is "competing" like this, with products that aren't really widely available, some that exist only for the headlines, and in some cases they still lose.
More symbolical than actual victories.

 

[Kenmitch]

This CPU will only be available till the end of the year, for what I understood.
I don't like how Intel is "competing" like this, with products that aren't really widely available, some that exist only for the headlines, and in some cases they still lose.
More symbolical than actual victories.

They only have high clocked offerings that are competitive. What else can they fuel the forum warriors with?

 

[DrMrLordX]

Prime95 AVX tests are like running a power virus, basically. Games, no.

Tell that to @VirtualLarry who constantly bakes his R5 3600 with a PrimeGrid workload that's actually worse than Prime95 SmallFFTs. You can get similar heat output from "real world" AVX2 workloads, such as video encoding. Prime95 is a bit hotter, but not by enough for the results to be completely inadmissable. The 9900KS is chewing up mad power in stuff like Blender, CBR20, and other AVX2-aware applications. You can't just hand-wave away all AVX2 workloads as being "power viruses".

Yeah, one system is with DDR4 3200, the other is with DDR4 2666 in FP testing that are usually bound by memory BW and latency.

y-cruncher specifically doesn't respond all that well to memory latency/BW, especially not on CPUs with strong cache performance. It's maybe a little more sensitive than CBR20, but not by much. Giving a 9900K DDR4-3200 isn't going to make a huge difference.

 

[Abwx]

Yeah, one system is with DDR4 3200, the other is with DDR4 2666 in FP testing that are usually bound by memory BW and latency.
IT IS THE STOCK settings i hear from usual suspect, but let's be honest here, the person who buys this CPU with intention to pair it with 2666 speed mem deserves a stint in hell. The day has come when it is OK to tinker with PL1/PL2, but god forbid to touch memory speeds and latencies.

FP is not bound by RAM speed, at least not the tests used by AT or Computerbase, the 99000K/KS do indeed quite well in this area.

Now on Integer it could be different like in 7Zip where the 3700X display 30% better perf/clock than CFL at Computerbase.de.

The discrepancy between the 9900K and the KS at AT is either due to the security mitigations if effectively implemented, or more simply to the 9900K being tested at the time with 3200MHz RAM like in their Ryzen 3000 review, but this latter possibility shouldnt had effects on the FP results.

That being said it s up to Intel to validate their IMC at higher frequency, PL1/PL2 tinkering is not technically overclocking since it s still within Intel s official specs and settings, and in this register the 9900KS use 172W in Cinebench to clock at 5GHz, according to the reviewer 159W is not enough to do so.

 

[Ajay]

Tell that to @VirtualLarry who constantly bakes his R5 3600 with a PrimeGrid workload that's actually worse than Prime95 SmallFFTs. You can get similar heat output from "real world" AVX2 workloads, such as video encoding. Prime95 is a bit hotter, but not by enough for the results to be completely inadmissible. The 9900KS is chewing up mad power in stuff like Blender, CBR20, and other AVX2-aware applications. You can't just hand-wave away all AVX2 workloads as being "power viruses".

So, I'd be happy seeing the power output used for: video encoding benchmarks, Blender Benchmarks or Maxon Cinema 4D. Anyway, these sort of benchmarks only matter for content creators. Pro content makers aren't using x570 or z390 boards and associated hardware.

I realize that Intel's i9 9900K(F,S) line eat allot of power - but these usually get blown out of proportion vis-a-vis the vast number of DIY PC users who do not, obviously, run power viruses like Prime. This has gotten freaking ridiculous

 

[Markfw]

So, I'd be happy seeing the power output used for: video encoding benchmarks, Blender Benchmarks or Maxon Cinema 4D. Anyway, these sort of benchmarks only matter for content creators. Pro content makers aren't using x570 or z390 boards and associated hardware.

I realize that Intel's i9 9900K(F,S) line eat allot of power - but these usually get blown out of proportion vis-a-vis the vast number of DIY PC users who do not, obviously, run power viruses like Prime. This has gotten freaking ridiculous

Well, there are those of us that use as many cores as we can get, and do it all day, every day, and I don't mean just me, the entire DC forum, and others like us. Now I don't know the percentage of total users that do this, but its not a niche. For people like me that have a $600 a month electric bill, all due to computers, power usage is king. And again, that applies to the entire DC forum as well as other groups. My son works in the CAD industry, and the numbers of high powered computers in his office is mind boggling. I don't know their power usage curve, but its a lot of high power being used all day.

I am sure there are other groups that have the same power concerns, I know only about these few groups first hand.

And if you wonder about DC, go in there and read a few threads. People talking about when its hot they have to turn them off, or cold, and they fire a bunch up.