Intel 9th Gen CPUs revealed

[BP]

I've been looking at the liquid cooling too, it's a lot easier than it was the last time I built my PC. I was looking at the Corsair H100i, £100+ seems to be the middle of the road price.

Yea I saw that one as being highly recommended as well. My big fear is that I may leave my computer on overnight to download large files and it springs a leak (or it springs a leak at anytime really).

Well, funny enough I managed to straighten the bent fins on the Phantek cooler (everything else seems okay) so went the 8700K gets here early next week I’ll use that and see what the temps are like.

 

[Turk]

Yea I saw that one as being highly recommended as well. My big fear is that I may leave my computer on overnight to download large files and it springs a leak (or it springs a leak at anytime really).

I'm assuming they come pre filled with coolant and then sealed at the factory, I'd have more confidence in a factory filled system, they probably put in the correct amount of coolant and test that it's correct, so I wouldn't be too worried about leaking.

The only issue I can see with a pre filled system is I wouldn't be able to mount it outside the case, my case does have two holes for a water cooling system.

 

[F1Aussie]

I have been using liquid cooling since 2013 and have left my PC on many times overnight, have not had a single drop leak out, it is nothing to worry about. Custom loop systems may be a different story though.

 

[Kek700]

Been a firm beliver in water cooling, AIO type.
Pushing 5M pixels at max field of AI.
Mounting 240 mm on the outside of my case has created a CPU @ 49 Deg C at max use in AC.
Plus water cooling my Gtx 980 Ti at max use give a GPU @ 44 Deg C.

( I admit i am deaf as a post, so i can run my fans at near full pelt. ( silent my system is not ))

Found this article very interesting..

My 6600k is running at 4.7 Ghz, i can find no real reason from this aticle, as yet, for updating.
I think 2019 may have something to change my mind.
I’m probably 30% down in CPU ram, how much this actually affects AC performance i cannot really
quantify. I just have a suspicion that i am in real terms not that far behind.
It would take about £600, for not much gain. ( mother board + 8700k )

 

[BP]

Well, forgot to update: I used the Phanteks cooler with the 8700K and I almost immediately start running benchmarks and stress tests, and then started feeling down because it was hitting up to 88 C...but then I actually put all that aside and started gaming...stays in the mid 60's to low 70's, even on properly multi-threaded games, and this is OC'd to 5.0 Ghz on air and not de-lidded, running most sims on high/max settings in triple monitor 1080p. I have the de-lidding tools and thermal compounds (conductonaut & kryonaut), but I keep wondering if it's worth the effort/risk. I swapped out the damaged Phanteks for a Cryorig R1 Ultimate; looks better and cools just as well.

I do plan on doing music production work with soon and possibly some light video editing, so maybe my opinion will change...but stressing over trying to maintain below 60 C CPU temps in benchmarks can be a meaningless pursuit if gaming is the primary focus (gaming + streaming is another matter), especially since the CPU will throttle down when getting close to the thermal limit anyways.

 

[Case_]

One really important thing is that I also ran process explorer with all games I had installed. Not ONE single game really utilizes more than 4 threads. rf2 for example only uses 1-2, AC 1-2, Raceroom 1-2. World of Warcraft only 1 although a MT patch is on its way.
ACC I had hopes for.. nope, still only 2-4 threads really pushing!

How do you figure out the number of threads? If I run Process explorer and look at the number of threads for the the game executable process, there's a lot more shown than the number you've mentioned, even if I count just the ones for the game executable itself...?

 

[RasmusP]

How do you figure out the number of threads? If I run Process explorer and look at the number of threads for the the game executable process, there's a lot more shown than the number you've mentioned, even if I count just the ones for the game executable itself...?

You are right, it's not about the actual overall amount of threads, which is really a lot!
It's about how many threads show some substantial CPU loads. Looks like this when you go into the "properties -> threads" of a running application:

You see there are a lot of threads but there are only 3 with more than 1% of overall CPU load.
For my processor (8 threads in Taskmanager) the limit is 100/8= 12.5%.
Once one of the these threads comes close to 12.5%, my CPU will be limiting. Above you see Raceroom with maximum amounts of AI at the Nordschleife at 800x600 to make sure it's really hitting the CPU limit.

The tricky part is that you can't look at this and see if the CPU is actually really at its limit. Rocket League for example shows 11% all the time on one thread but it doesn't matter if you limit the FPS at 50 or at 400. More or less stays the same. But that's easy: when the GPU isn't at 99%, it's the CPU limiting.
If you then check with process explorer how the threads are using the CPU you'll either see one thread close to 12.5% (or 25% for 4 cores, no HT) or the overall CPU load will be at 99%.

But it explains roughly why 12 cores at 3.5 GHz are not really quicker than 4 cores at 5 GHz for games like Raceroom. Windows manages to split the load more or less across all cores but you simply can't load 12 threads to 100% with only 3 threads!
Where my knowledge ends is how Windows manages to get Raceroom to use more than 37.5% overall as 3x 12.5% wouldn't allow for more. Due to this "shuffling" how I call it, it's somehow possible and that's also why I'm getting higher fps numbers with hyperthreading active vs disabled. Overall load gets lowered but still more fps! Probably has something to do with Cache stream etc.

Hope that clarifies things? If not feel free to poke me again and again


This is Assassin's Creed: Origins btw. It manages to peak up to 90% overall CPU usage. Still one thread too big to hit 100% but for this game, the AMD CPUs beat the crap out of my old i7 2600k!

 

[Case_]

So you only consider threads with certain CPU usage as relevant. OK, got it. Cheers.

 

[Terry Rock]

Thanks DucFreak, I think you have done my research for me. I currently have an i5 4670k oc'd to 4.4 stable on air. I think this Ryzen 5 2600x and 32gb DDR4 ram with 470 mb will be a good upgrade. I already have a 1080 gpu. Do you think that will be a good upgrade performance wise?

What you are proposing is the exact combination I'm running right now...coming from an O/Ced I7-3820@3.9 GHz.
It may not be as fast as the top-tier Intel chips (framerate-wise)... but I can tell you this with a degree of certainty, "It is extremely satisfying for every sim currently on the market...including ACC".
What I generally look for, is the deviation between minimum to maximum framerate...not the outright framerate number itself.
Everything I've run since building on that CPU, is 'glass' smooth.
Do not manually overclock the Ryzen5 2600X. It does it on it's own better than we ever could, based on conditions.
Set the RAM speed and voltage to what ever is recommended and go run your sims.
Links are not allowed...but go look at my Youtube channel under Saabjock.
(Bear in mind that uploading to the channel will induce some small stutters here and there that you won't actually experience locally on your PC.)
Even with that aside....look at the Rfactor2 runs with every setting maxed out...at night in that Oreca 07 at Sebring. Those are the only videos I've uploaded so far with the new build...everything else is the I7-3820.
Pay particular attention to the 'built-up' areas (pit-stalls, lights, wall, etc..) especially in the start/finish area. Look at the smoothness as I go pass objects at the side of the track.
Tell me if that won't be enjoyable for you, while you have that extra $540 (processor only savings) in your pocket.

 

[Kek700]

@Martin Fiala @RasmusP

Forgive my total ignorance on this subject. It has just caught my attention reading the above,
due to thinking about a future CPU change.
100/8 = 12.5% per thread of your 2600k
25% per core i assume that represnts 100% of the whole 4 core CPU.
Why is it that 1 thread cannot be 25% and represent 1 whole core being used.
( the other thread not being switched ).
I realise ( i think ) that there is a thread of software program and a thread attached to a process
core. ( usually, i assume, 2 threads allocated in the cae of hyper-threading )

There is something basically wrong with how i am seeing this, but i unable to fathom it.
( Usually means i do not have enough knowledge.)

If it’s too complicated and i am obviously a lost cause, pleae ignore.

 

[RasmusP]

If it’s too complicated and i am obviously a lost cause, pleae ignore.

Not a chance with you being such a nice person to know Ernie

I'm not sure if I got your question right but I do think your's asking:
"When with 1 application thread, the overall limit is still the one of 1 CPU thread, then why isn't just 1 CPU thread loaded to 100% but instead it's evened out over the whole CPU?"

Or in pictures:
Why is THIS:

Looking like THIS:

And not like THIS:

Is this correct?
Then the simple answer is: Because it's faster!
The more complex answer is: When split across all cores, the overall CPU load is higher than the maximum possible load on 1 core would be possible.
Or in other words: The sum is more than the single parts

From my technical knowledge it has something to do with RAM+Cache. Task Manager is only showing an "average". If you would look into an infinite short moment of time you would see that actually 1 CPU thread is loaded to a 100% and right after that the next CPU thread gets loaded to a 100%.
And while 7 CPU threads are doing 'nothing' most of the time, they already prepare their next "100% load moment" and fill their cache and stuff.
But I'm no software developer nor am I a PC part engineer so this is just guessing!

A simple analogy for this is my beloved "hammer and the nail" story. I'll copy paste it from another thread:
Imagine that each thread of a game is one big massive nail. Like a little smaller than a flock+sledgehammer so you can still do it with one hand on the hammer but it takes some time to hit again with it.
Now:
- Real core = one person with its right hand hitting the nail.
- More real cores = one person hits the nail with full force, moves aside, next person hits. (swapping the person doesn't cost time though as long as there aren't enough nails for everyone).
- Hyperthreading: each person hits a second time but with the left hand! lol..
- Higher per core performance: Each person has more strength
- Higher GHz: Speeding up the whole process

- More Game threads: the big nail is split into multiple smaller ones so each person can hit his own nail. A lot quicker as you can imagine than to swap persons (although the swapping time is really short, they still can't hit at the exact same time in parallel!)

And finally: more game threads than real cores (hyperthreading area):
Each person gets 2 small nails instead of one so the person can hit the second nail with the left hand.
Yes, still quicker but not exactly as quick as it would be with 2 right hands

From my excel sheet: This is how different CPUs perform over different amounts of application threads. You clearly see how the curves go flat when the limit is reached!
And also when the hyperthreading area begins. Most noticeable for the 8700k!

Just for the overview:
i5 3570k: 4 cores, 4 threads
i7 2600k: 4 cores, 8 threads
i7 8700k: 6 cores, 12 threads
R. 2700x: 8 cores, 16 threads

 

[Kek700]

I'm working my way through it...

 

[RobertR1]

Thought I'd share my findings a bit more. i turned off HT on my 9900k because none of the sim racing games take advantage of it (or VR) and it's just there to add more heat and require more voltage to maintain clock speeds.

With HT off, the best combination I've been able to run is 5.1ghz on all cores at 1.28v That's gameplay and stress test stable. The jump to 5.2ghz requires 1.35v which spikes up the temps a fair amount so not worth the trade off for an extra 100mhz while on an air cooler. If I had an AIO/CLC then I might run that.

A few years down the road when the 9900k starts getting taxed by newer games/sims/vr, I'll turn HT back on and extend the lifespan of the chip without needing an upgrade.

 

[Kek700]

Ok, i “think” i get it.
Forgive my simplistic outlook....

I ran AC with full field and always assumed only one or two cores were utilised. Apparently all 4 cores are chugging away at 80-90 odd %.
I also noticed the “PHY late” was going quite high per lap when the FPS was at 117 fps. ( I’m assuming this means physic’s late ).
Dropped the FPS to 60 and immediately noticed the PHY late was reasonably static. Only moving when something unusual was happening in the race.

My point is, ( this is the only way i can rationalise this )

i have only 4 whole cores to number crunch all the requirements for graphics card and all the physic’s requirements and constant calculations required within the game.

“i don’t have at my disposal all these little men with their little hammers to do all the small tasks in parallel.”

I “ think “ i “ have to use a whole processor core to do these tasks in series.
The other cores doing other work thats required, so if the priority is higher that process ( core ) is reassigned.

( Maybe there is no grand plan but a sequential plod through whats required, if its late, its late. The “PY-late” seems to suggest this.
It’s a pity i cannot just easily swap my 6600k for a 8700k and see the difference. )

So my one core has stopped, cache is loaded and it Is serially going through the calculations for the higher priority requirement.

A “device” with more cores and hyper threads can keep the original core still working, pick another core or cores with two threads or more, do the sub calculations in parallel via threads. Thus speeding up the whole process.
( if this is not the case then it would be a bit difficult for me to give preference to a processor with more cores and hyper-threading ) ( my i5 is running a highest 4.7Ghz )

I obviously do not know how Kunos designs this into the software, as far as coping with 6 core 12 threads, i5 4 core or 2 core processor. From my point of view, if mine is using all 4 cores then i must assume it will use 6 cores of a processor .

“Or maybe it’s a windows 10 function.”
Maybe 4 cores are working on supporting the gaming process!!

This is really, “all guessing”. Am i miles off, do i need to rethink it again. Is any of this legible.

 

[Bjarne Hansen]

Running i7 7700k @ 5.0 Ghz (AVX @ 4.8 Ghz) Voltage 1.32. Cooling :212 EVO air cooler.
Made a tool for delidding but has not done it yet.
Kept HT on mainly due to I feel it runs more smooth racing against AI in Rf2.
That is a feel have not tested on way or another.

Find it a shame that even the the newer sims are still basically single core applications.
Would not mind that a new sim came out that stated minimum 8 cores and used them!
While fun to OC to the max for your CPU this chasing of 0.1- 0.2 Ghz more OC are not making a huge difference on any of the current games.

 

[RasmusP]

Yep, you got it right apart from one point:
More cores or hyperthreading barely give you more fps. I'm not sure about this "phys late" thing but in general it's like this:
Overall performance boost from your 6600k to a 8700k:
80% of it due to increased single thread performance
10% due to more real cores, faster swapping, more cache flow or whatever
10% hypterthreading, doing the same like real cores just different

For the following images it's important to see that the values are averages for all "# of samples". It might be interesting for you to download the PassMark benchmark and run the CPU tests for your specific overclock

And to the Ryzen 2600:

Single thread difference between your's and the 8700 is about 26%, multi thread performance about 98%.
Now let's check a real test between them:

You see that more cores/hyperthreading are not the crucial point there. It's all about the single thread performance!
Whether or not it makes a difference for "phys late", I don't know... I think not since just because you have more headroom with your CPU, the phys late thing is still within the two threads of Assetto Corsa and once they reach their limit, it will cause "being late". More cores or HT will give you a slightly better performance overall as you can see when you compare the 6600k to the Ryzen 2600 but it's tiny compared to a better raw single thread performance!

I bet if you'd disable 2 cores on the 8600k it would still put out 120 fps in Hitman!

All images taken from this guy: LINK
And from the PassMark homepage

 

[Kek700]

@RasmusP , many thanks, a million miles from an expert; But a better understanding now.

 

[Kek700]

just out of interest, did passmark for my i5 6600k 4.7Ghz = 10255

 

[RasmusP]

just out of interest, did passmark for my i5 6600k 4.7Ghz = 10255

And single thread score?
Maybe you could just upload the whole bench-score via snipping tool, copy paste?
This is mine for comparison (i7 2600k @ 4.4 GHz)

 

[Deleted member 197115]

Thought I'd share my findings a bit more. i turned off HT on my 9900k because none of the sim racing games take advantage of it (or VR) and it's just there to add more heat and require more voltage to maintain clock speeds.

With HT off, the best combination I've been able to run is 5.1ghz on all cores at 1.28v That's gameplay and stress test stable. The jump to 5.2ghz requires 1.35v which spikes up the temps a fair amount so not worth the trade off for an extra 100mhz while on an air cooler. If I had an AIO/CLC then I might run that.

A few years down the road when the 9900k starts getting taxed by newer games/sims/vr, I'll turn HT back on and extend the lifespan of the chip without needing an upgrade.

So you essentially turned it into 9700K. Did you try going higher than 5.1GHz, I have settled for 5.0 on all cores with HT on, I think it's at around 1.325v (adaptive voltage), CPU cache at 47. No throttling or heat issues so far with AIO Corsair.

This helped with OCing process