Intel Core i9-11900K

[ent]

Well those CPUs are disappointing, but I guess most people expected this. Not much to add here.

I have a question about test setup/testing procedure tho. I was looking at SPI 32M results in this review and one thing caught my eye. If you compare those 2:

11900K review

5800X review

You can see that all Zen2 based CPUs are about 2,5 minute slower in the graph from 11900K review compared to 5800X review. I assume those tests were re-run as test setups differ between both reviews (different AGESA, different Windows 10 version). It seems to me that Zen2 SuperPi 32M results in 5800X review are the results from the time reviews for Zen2 were published and were not rerun for Zen3 launch. Of course SuperPi 32M is quite outdated benchmark and it can behave strangely on modern SW and HW, but those results are still surprising to me. Especially that Zen3 results are not affected by whatever breaks Zen2 scores, if you look at SuperPi 32M results from 11900K review it seems that 5800X needs HALF the time 3700X needs to calculate 32M. I know IPC gains between Zen2 and Zen3 were significant, but those should not be that huge.

As a 3900X owner I actually see the same issue (my SuperPi 32M just got approx. 2:30 slower) on my system and I am quite puzzled what causes this (especially seeing that Zen 3 is not affected). I tried to dig into system settings to fix this somehow with no luck, but I did not look into downgrading AGESA/Windows versions yet.

Does anyone have any hints? Otherwise maybe it is time to retire SuperPi 32M from benchmark list, as it seems to be very unreliable?

 

[Lomskij]

Not to join Intel bashing here, but in the UK 11900k is priced at £560 vs £420 for 5800x. I'd need a very compelling reason to buy Intel at this moment..

 

[Jism]

Completely misleading TDP. "125W" and yet consuming over 200W easily on multithreaded. Why cant they stick with the model of AMD where TDP is pretty much as advertised.

 

[Deleted member 205776]

Completely misleading TDP. "125W" and yet consuming over 200W easily on multithreaded. Why cant they stick with the model of AMD where TDP is pretty much as advertised.

Because advertising their TDP from boost clock would make their CPUs less compelling than they already are.

 

[Why_Me]

Not to join Intel bashing here, but in the UK 11900k is priced at £560 vs £420 for 5800x. I'd need a very compelling reason to buy Intel at this moment..

Bashing Intel is perfectly acceptable. Sticking up for Intel is considered trolling.

 

[Deleted member 205776]

Sticking up for Intel is considered trolling.

Few people in this thread could use that information.

 

[laszlo]

Intel today launched its 11th Generation Core "Failed Lake"

i really don't understand why they've wasted money to make this on 14nm ... but seems they have a lot of $ for now... scalpers won't buy these for sure ..

 

[W1zzard]

You can see that all Zen2 based CPUs are about 2,5 minute slower in the graph from 11900K review compared to 5800X review

Could be new AGESA. Maybe boosting behavior changed again, or core placement. since it's a 1-core load, it might end up on a slower core.

Or my change in memory settings. 3800CL16 with IF at 1700 is probably slower than 3200CL14 with IF at 1600 1:1. But running one platform at 3200CL14 vs the other at 3800 CL16 will cause tons of drama, especially with people who don't understand all the details.

Maybe I should have just stuck with 3200CL14, for everything, but there were SO MANY complaints

 

[HenrySomeone]

Maybe I should have just stuck with 3200CL14, for everything, but there were SO MANY complaints

Would have probably been better, yes, just like RandallFlagg mentioned and we also know who was complaining...(namely a particular, errr...overzealous group to say the least)

 

[Showman]

ASRock B450 PRO4 - 30 dollárért, 5600X- ért - 15 dollárral kevesebb, közel azonos CPU-teljesítmény mellett, miközben 2 magja kevesebb, és körülbelül FÉL annyi energiát fogyaszt, mint az 11700. %.

You forget it’s at a starting price of 11,700, compared to the 5600X on the market for half a year.
And that also qualifies you for looking at the price of the 5600X msrp, even though you can’t get it anywhere for that much.
Then only the OC that is important to the few remains the object of your struggle? And consumption?

 

[1d10t]

And as of today the AMD 3600 is out of stock on both Newegg and Amazon and not expected to be back in stock until the third week of April. lol

Try refresh your browser

 

[HenrySomeone]

And you should try refreshing your reading comprehension! He talked about the 3600, not the 5600x (which you can now occasionally (if you are very diligent in refreshing your pages indeed), almost 6 months after release finally get at the MSRP, hurrah!). And he was right in his point too - AMD has currently pretty much zero (sensible) options for a budget-mid range build!

 

[ent]

Could be new AGESA. Maybe boosting behavior changed again, or core placement. since it's a 1-core load, it might end up on a slower core.

Or my change in memory settings. 3800CL16 with IF at 1700 is probably slower than 3200CL14 with IF at 1600 1:1. But running one platform at 3200CL14 vs the other at 3800 CL16 will cause tons of drama, especially with people who don't understand all the details.

Maybe I should have just stuck with 3200CL14, for everything, but there were SO MANY complaints

Thanks for your reply.

It could be AGESA version indeed. If that is the case it would be great if AMD could say something about this, as it seems to affect only Zen2 and not Zen3, so I guess it is not "working as intended". All Zen3 SuperPi 32M runs I saw are in the 6:00-6:30 range, while for Zen2 in the old days SuperPi 32M was able to finish in about 8:00-8:30, now all runs are taking longer than 11:00. Keep in mind 2700x was calculating 32M of SuperPi in about 9:30 (https://www.techpowerup.com/review/amd-ryzen-7-2700x/5.html), so 8:00-8:30 seems about right for Zen2.

I actually run into this post: https://www.techpowerup.com/forums/...st-windows-10-20h2.275152/page-2#post-4401000 which suggests that Windows 10 20H2 update is the culprit, but I did not verify this yet.

On my Zen2 system I tested quite extensively different CPU affinities for SuperPi as well as different power plans and different process priorities and those seem to have small impact (at least compared to 2:30 gap we are seeing). I am also running same mem speed and IF speed as previously (1867 1:1) and the gap is still there. So based on my own personal tests I would lean to say it is neither memory nor problem with assigning fastest core, of course I could be wrong.....

Sorry for off-topic, but this issue is bothering me for quite a while and now I found a confirmation that there is indeed something wrong with SuperPi 32M on Zen2 thanks to this review. I guess I need to find some time to test different builds of Windows 10 and possibly try to downgrade BIOS on my mobo to see if any of those could solve the issue. It just seems very strange to me that those times degraded just like that.

 

[W1zzard]

So based on my own personal tests I would lean to say it is neither memory nor problem with assigning fastest core, of course I could be wrong.....

There goes my theory

Sorry for off-topic, but this issue is bothering me for quite a while and now I found a confirmation that there is indeed something wrong with SuperPi 32M on Zen2 thanks to this review. I guess I need to find some time to test different builds of Windows 10 and possibly try to downgrade BIOS on my mobo to see if any of those could solve the issue. It just seems very strange to me that those times degraded just like that.

Keep me updated please, email, PM or Skype is fine, too.

Have you tried downgrading the BIOS?

 

[Forza.Milan]

not bad

 

[ent]

Have you tried downgrading the BIOS?

Nope, did not try it yet. Actually not sure how far back someone can go safely on X570, I think there were some limitations with downgrading BIOS.

I will let you know if I figure out something.

 

[RandallFlagg]

Could be new AGESA. Maybe boosting behavior changed again, or core placement. since it's a 1-core load, it might end up on a slower core.

Or my change in memory settings. 3800CL16 with IF at 1700 is probably slower than 3200CL14 with IF at 1600 1:1. But running one platform at 3200CL14 vs the other at 3800 CL16 will cause tons of drama, especially with people who don't understand all the details.

Maybe I should have just stuck with 3200CL14, for everything, but there were SO MANY complaints

In a sense it doesn't matter. Different sites use different memory setups, different motherboards, both of which can affect performance 1-5% and can compound. That is enough to knock almost any of the top CPUs from middle of the pack to the top, or from the top to the middle.

In my experience most do not even look at the test setup. People who haven't looked at that really just don't know what they are looking at, charts are nice but they really are only half the story. There's nothing wrong with the test setup really, it is what it is, it's people not looking at it to see what the charts are really telling them.

Example-

TPU uses Asus Maximus XIII Hero Z590.

Here's SOTR from the MB review. This is -5% from the best performing motherboard :

This board has good memory bandwidth which positively affects many productivity apps.

However it is one of the worst for latency, which negatively affects games, This combined with gear 2 settings explains the better than normal app performance on TPU vs worse than normal gaming performance.

So basically the TPU test system is built to perform well on bandwidth hungry apps, but does poorly on latency sensitive games, from what I can tell.

 

[W1zzard]

So basically the TPU test system is built to perform well on bandwidth hungry apps, but does poorly on latency sensitive games, from what I can tell.

Interesting hypothesis. Nate is using a different board/memory/CPU though, he's in the States, I'm in Germany.

Intel provided the ASUS motherboard to me. ASRock provided a Z590 Taichi, but I can't use it because it has no option to turn set power limit to default. You can only type in numbers, but for that you have to know the default PL values first

 

[Makaveli]

Well those CPUs are disappointing, but I guess most people expected this. Not much to add here.

I have a question about test setup/testing procedure tho. I was looking at SPI 32M results in this review and one thing caught my eye. If you compare those 2:

11900K review

5800X review

You can see that all Zen2 based CPUs are about 2,5 minute slower in the graph from 11900K review compared to 5800X review. I assume those tests were re-run as test setups differ between both reviews (different AGESA, different Windows 10 version). It seems to me that Zen2 SuperPi 32M results in 5800X review are the results from the time reviews for Zen2 were published and were not rerun for Zen3 launch. Of course SuperPi 32M is quite outdated benchmark and it can behave strangely on modern SW and HW, but those results are still surprising to me. Especially that Zen3 results are not affected by whatever breaks Zen2 scores, if you look at SuperPi 32M results from 11900K review it seems that 5800X needs HALF the time 3700X needs to calculate 32M. I know IPC gains between Zen2 and Zen3 were significant, but those should not be that huge.

As a 3900X owner I actually see the same issue (my SuperPi 32M just got approx. 2:30 slower) on my system and I am quite puzzled what causes this (especially seeing that Zen 3 is not affected). I tried to dig into system settings to fix this somehow with no luck, but I did not look into downgrading AGESA/Windows versions yet.

Does anyone have any hints? Otherwise maybe it is time to retire SuperPi 32M from benchmark list, as it seems to be very unreliable?

So I just ran Super PI 32m on my 5800X system and here is my result

 

[RandallFlagg]

Interesting hypothesis. Nate is using a different board/memory/CPU though, he's in the States, I'm in Germany.

Intel provided the ASUS motherboard to me. ASRock provided a Z590 Taichi, but I can't use it because it has no option to turn set power limit to default. You can only type in numbers, but for that you have to know the default PL values first

I'm just saying, a CPU review can never be just a CPU review. It's always a review of the chosen configuration.

 

[W1zzard]

I'm just saying, a CPU review can never be just a CPU review. It's always a review of the chosen configuration.

Yup, and if even Intel themselves fail to provide a proper config, their loss

 

[cadaveca]

Yup, and if even Intel themselves fail to provide a proper config, their loss

Meh. You put CPU in board, changed settings, ran tests, same as anyone else might do with any CPU. That's how we use 'em, right?

Intel shouldn't give configurations at any point except if they are presenting the results themselves.

 

[Tom Sunday]

I love pancakes and all the way back to IHOP! I would however say that opinions or 'words of wisdom' between Intel and AMD will never go away. The majority of Intel followers over the years will always stick with Intel no matter what and just like the AMD crowd doing their continuing thing. Pricing will also not matter as once someone with deep pockets and who actually has an easy $600 to shell-out for a CPU, for him $100 here or there does not really matter. With more smooth marketing and future commentaries by the "influencers" to come, will then further do its bidding and as many new so-called comparative test will surely then show conflicting results to confuse us even more.

The "influencers have arrived" and as I mentioned yesterday in my previous post. Conflictiing testing commentaries, opinions and reports prevail now that the Intel CPU saga broke loose. Suggest to having a look at WCCFTECH and its "Executive Review: How Does Intel’s Core i9 11900K Compare To AMD’s Ryzen 5800X." Some key comments there noted that Rocket Lake marks Intel as being very competitive with AMD again (up to 8 cores at any rate) and depending on supply, could see that Intel very well will defend its position with also (finally) a PCIe 4.0 platform in its ranks!"

What we need to remember is that certain 'review sites' are tainted and as manufacturers are placing income producing ads with them and thus reviewers are told to be more cooperative. And money in ones pocket always wins and as the site-owners want those advertisers to come back and not getting ruffled with questionable or even bad reviews. I am quite sure that many here will not appreciate me saying this, but it's a stark reality. I also see this happening in many tech and gaming magazines as well where certain products or manufacturers corporate statements get preferred treatment and positions. Therefore I always read at least a dozen or so important reviews of the many competing sites available to form my own oipnion. Besides its my own money at stake.

To be sure I am not a fanboy either way but rather a daytime investor. Made some quick and real money with AMD in a short 8-months time last year and now my AMD windfall in 2021 riding on Intel waiting for their 4th quarter to arrive. So there you have it with me having a motive as well and it's all with having money in my pocket as well.

 

[RandallFlagg]

Yup, and if even Intel themselves fail to provide a proper config, their loss

Intel did not force you to use DDR4-3800 gear 2. Not guaranteed to work on either system. You decided to do that due to pressure from the AMD community.

This is what happens when you use 2x 16GB DDR4-3200 CL16 gear 1 on Z590 Rog Maximus Hero XII and 4 x 8GB DDR4-3200 CL 16 with X570 Aorus Master. Both systems are running memory synchronously.

It's actually far, far more likely that a typical DIY type will use this config.

 

[R0H1T]

And you sound like a typical Intel fan/apologist ~ are you now gonna blame system instability, observed by many reviewers, on "AMD community" as well