Intel i9 9900k to hit 5GHZ - Do you have one Rob?!

[Noel 4,185]

4 hours ago, Rob Ainscough said:

I still maintain why risk uneven distribution of force on the CPU with the D15 and a vertical mounted motherboard? ... no matter the clamping system, gravity will be working against it on a vertical mount over many heat cycles.

Cheers, Rob.

I didn't like that feature of my D14 either and to satisfy my concern I used zip ties to fashion a sling that attaches to the case and supports it well figuring it can't hurt and will help if the case is moved or bumped etc, even though Noctua says their mounting system is not going to harm anything provided the PC isn't transported/shipped w/ the beast installed.

[Noel 4,185]

On 7/25/2018 at 11:24 AM, Rob Ainscough said:

But the CPU performance/price point that is a perfect match for Flight sims is the 9700K as it comes without HT and 8 cores (just 8 real cores no fake ones). 

Cheers, Rob.

EDIT: Best price/performance most likely:

nVidia 1170Ti
Intel 9700K

Either you or SteveW or someone else will hopefully be able to clearly answer a couple of questions.  

I have 3930K and have HT enabled and use an AM to put the main thread on one whole core, core 1 as it were, and have logical processors 5-6-7-8-9-10-11-12 available for terrain texture loading, and the OS on LP 0 and 1.   When flying thru super complex terrain which is keeping the LPs busy I'll often see all of the LP dedicated to terrain texture loading all running 90-100%.  So here is the question...which is better (meaning of course which is doing more required work), or are they in fact really the same, 2 LPs of one physical core each running at 95%, or to have that one physical core in a non-HT enabled environment running at 95%?  One of the reason I have always favored 6+ cores is that these cores can get very busy during complex scenery so it seems more is better!  And it certainly looks impressive to see all 8 of my available LP's busily loading terrain textures.    And I must say, since doing this I've had the clearest textures staying fully updated always of any way of running P3D so far, and perfectly smooth performance.

I obviously don't understand the ramifications of splitting a physical core into two logical cores.  Why is this ever done?  It seems like it must have something to do w/ opening up a pathway to take on a new thread, the type of thread that runs on one core.  Is it that by creating a logical division this makes for example the 2nd 'core' able to take on a new thread whose initiation might be held back on the 1st core as it's yet not done handling the prior task in queue, or something along those lines so that it effectively gets around some barriers to parallelism as it were.   Why does the type of processing involved in terrain texture loading not benefit from hyper-threading?  Does this belief come from anecdote/experience, or is there a technical/theoretical basis for the belief the HT is not worth employing for P3D?

It's all about the solder for me!  If the 9700K is not soldered but 9900K is, it's 99K for me.  If it's 9700K I'd like to know the answers to the questions above as to why HT enabled is not the better option.  I should qualify this as well by mentioning the CPU is going to run hotter and so ultimately be less overclockable w/ HT enabled, but once again, doesn't that imply that CPU is doing more work, and don't we like that?  Unless it's doing 'useless' work, of course!

Edited August 2, 2018 by Noel

[Noel 4,185]

This thing just makes me want to lose the fans and submerge it into the bottom of a 5 gallon bucket of water w/ ice cubes:  Could get a good couple of hours out of that maybe!

 

[SteveW 1,785]

2 hours ago, Noel said:

So here is the question...which is better (meaning of course which is doing more required work), or are they in fact really the same, 2 LPs of one physical core each running at 95%, or to have that one physical core in a non-HT enabled environment running at 95%?

First consider two particular types of process:

One process that can be duplicated across cores or LPs as many as we want. For example a video encoder can work by having each frame to be processed on separate cores multiplying the rate the movie can be encoded.

Another process that is monolithic - one part must complete before the next part until all the parts are completed (Von Neumann), the time taken for the process is the sum of all the times of those parts put together.

Whereas the movie encoding can be accelerated roughly speaking by adding cores (say 8x on an 8 core CPU over one core), but the other process can't be accelerated other than with core speed.

Now consider that P3D and FSX are made from processes that act like both of those, some monolithic, other can be paralleled.

Let's look at the two LP core:

If one LP says it is at 100 % and the other LP says it is at 0% - the core is at 100%.

If both LPs say they are at 100 % the core is at 100% - in other words those LPs are really only at 50%.

So you must aim to place monolithic processes on single LPs of a core, and you can place two of the parallel process per core on each LP.

The monolithic process can be thought of as rendering and user input, the others can be thought of as collecting and arranging data for display, the background process.

As we add cores or LPs for the background process we notice the time taken to load the sim decreases. There's only so much data your PC can collect at once and so there's an optimum number of those cores to use. Allocate more than you need and you don't get extra performance you get less as the cache is stretched more for no extra work done, and also you get extra heat. A big mistake on a P3D system is to ignore this or suffer odd behaviour. P3D will fill up all the cores or LPs you allow it to use so eventually you will saturate the chip and overstretch the shared cache.

 

 

Edited August 3, 2018 by SteveW

[Guest]

21 minutes ago, Noel said:

Could get a good couple of hours out of that maybe!

You'd be better off mounting it external to the Case, attach a bunch of Peltiers around the rad and wire them into a separate power supply.

Peltier Module

or

small portable freezer

stick the rads in there and cut out a wholes for tubing and seal around the wholes ... you'll still need a drain for condensation

Cheers, Rob.

[shivers9 1,173]

1 hour ago, SteveW said:

Allocate more than you need and you don't get extra performance you get less as the cache is stretched more for no extra work done, and also you get extra heat. A big mistake on a P3D system is to ignore this or suffer odd behaviour. P3D will fill up all the cores or LPs you allow it to use so eventually you will saturate the chip and overstretch the shared cache.

OK so let me see if I got this right. With the soon to be released 9700K and 9900K then it sounds like the 9900K may be over kill for a P3D only computer. What would be you guys best guess on future proofing for V5 of P3d. Would we be better off going with the 9900K and turning of HT for now to keep temps lower. Given the very few things we know about LM plans for V5 wiill the 9900K help with the new lighting and some possible changes to the engine.

[brucewtb 229]

On 7/31/2018 at 4:18 PM, zmak said:

Agreed. the z370 bios upgrade is sort of a mild hack in my brain anyway . If something dies down the road there is always the niggling feeling, did the z370 bios upgrade ok??

Well I have taken this advice to heart and have RMA'd the board as well just before the return period from my retailer expired.  I now await with interest the release of the i7 9900K/9700K and supporting MBs.  If this looks like dragging into next year I may use the store credit to get a 1180 GTN which looks like being available in a month or so.

Bruceb

[G-RFRY 1,658]

There will always something coming round the corner Intel latest on CPUs, 10nm CPU Q4 2019.

[Noel 4,185]

13 hours ago, SteveW said:

First consider two particular types of process:

One process that can be duplicated across cores or LPs as many as we want. For example a video encoder can work by having each frame to be processed on separate cores multiplying the rate the movie can be encoded.

Another process that is monolithic - one part must complete before the next part until all the parts are completed (Von Neumann), the time taken for the process is the sum of all the times of those parts put together.

Whereas the movie encoding can be accelerated roughly speaking by adding cores (say 8x on an 8 core CPU over one core), but the other process can't be accelerated other than with core speed.

Now consider that P3D and FSX are made from processes that act like both of those, some monolithic, other can be paralleled.

Let's look at the two LP core:

If one LP says it is at 100 % and the other LP says it is at 0% - the core is at 100%.

If both LPs say they are at 100 % the core is at 100% - in other words those LPs are really only at 50%.

So you must aim to place monolithic processes on single LPs of a core, and you can place two of the parallel process per core on each LP.

The monolithic process can be thought of as rendering and user input, the others can be thought of as collecting and arranging data for display, the background process.

As we add cores or LPs for the background process we notice the time taken to load the sim decreases. There's only so much data your PC can collect at once and so there's an optimum number of those cores to use. Allocate more than you need and you don't get extra performance you get less as the cache is stretched more for no extra work done, and also you get extra heat. A big mistake on a P3D system is to ignore this or suffer odd behaviour. P3D will fill up all the cores or LPs you allow it to use so eventually you will saturate the chip and overstretch the shared cache.

Thanks!

 

OK then it's pretty much as I guessed w/ the big question remaining unanswered:  what then is the optimum number of cores or LPs as it were to use for P3D?   Presumably terrain texture loading is one of those background processes that I can guess is more parallel-oriented, correct?   How can one tell if one is 'oversaturating'?  I have best performance I can imagine on my nearly 6 y/o system, certainly the best setup of any, and you were largely responsible for that and the AM I use!  And I've got 8 LPs dancing happily along.  How can one tell the optimum number of cores/LPs to make available?  If I went w/ a 9900K and HT enabled I would bump that up to 10 LPs.   How can one tell if 10 LPs is better than 5 C's?

Edited August 3, 2018 by Noel

[martin-w 3,113]

18 hours ago, Rob Ainscough said:

 

But I've always agreed with you that it's a good cheap and less hassle alternative if one is willing to sacrifice a few degrees under loads, but where the D15 starts to fail on cooling under loads is when OC is very high.

Cheers, Rob.

 

Yep, absolutely. I 100% agree with you. Always have. That was the first thing I noticed in regard to the Jayz2Cents video when I saw it on his channel. He tested at stock settings. That didn't make sense, he should have tested at a high overclock, for it's under those conditions that custom loops move ahead.

 

Quote

If done correctly they don't leak within a 10 year period

 

Human beings are fallible though, and hardware can be faulty. But yes, if all hardware is sound and the human being building the loop doesn't succumb to human fallibility all will obviously be well.  

 

Quote

Anyway, my point was there is no need to "fear" leaks

 

I agree with you, my point was absolutely nothing to do with fearing leaks, it was a direct response to Noel who requested an AIO that was "impervious to leaks." In fact I'm not sure why this debate even started. I sense you may have jumped into "water cooling defensive mode" unnecessarily. 🙂

 

Quote

it's equivalent to fearing that heavy massive D15 is going to warp a motherboard to the point of failure 🙂.  It has happened and it's pretty rare and not ALL information is required to make it to the Internet. 

 

 I would disagree with you there and point out it is wrong to say "it has happened" you have no definitive evidence of that at all. There is zero evidence that  a motherboard has been "warped to the point of failure" by the D15. On the other hand, AIO's leaking and custom loops leaking are easy to find on the internet. So no, I don't see the two as equivalent. AIO's are better these days though, seem to be more reliable. 

 Actually, I lie, there is one, it was a PC that was won in a competition a few years ago and shipped half way across the US with cooler attached. (Madness) 

 

Quote

But, I still maintain why risk uneven distribution of force on the CPU with the D15 and a vertical mounted motherboard? ... no matter the clamping system,

 

You ask why... because it's not a risk at all.  I've seen none of the uneven distribution you talk of, comparing my CPU with my daughters that's cooled by a Corsiar H100i V2. Same degree of core temp variability. I will tell you what does seem to even out core temp though... delidding I have found. 

 

Quote

gravity will be working against it on a vertical mount over many heat cycles.

 

You forget how long the D14 has been around. Must be 8 years now. And nobody has thrown their arms in the air, yelling in a panic, that the evil D14/D15/D15s has caused such an uneven distribution of force that remedial action is required. Don't think I've ever come across anyone who has claimed this has caused an issue. 

 

Quote

And if one plans to delid and NOT replace the HSI (the most effective use of delid) for direct contact with block, then the D15 will not be a great choice for that scenario (assuming there is a mounting kit specific for the D15 and no HSI).

 

The vast majority of enthusiasts, do not employ direct die contact. If you are going to such extremes you will be a competitive overclocker or at the very least a very very serious individual. In which case you will obviously be installing a custom loop. So not sure why you even mentioned that. And rest assured, delidding is worth while with IHS in place. minus 20 degrees for my daughters CPU and minus 15 for mine. 

 

[martin-w 3,113]

13 hours ago, Noel said:

This thing just makes me want to lose the fans and submerge it into the bottom of a 5 gallon bucket of water w/ ice cubes:  Could get a good couple of hours out of that maybe!

 

 

 

Ha, I love your creative solutions to CPU cooling Noel. 🙂

[Noel 4,185]

6 minutes ago, martin-w said:

 

 

Ha, I love your creative solutions to CPU cooling Noel. 🙂

Creative and very...low tech!

[martin-w 3,113]

19 hours ago, jabloomf1230 said:

But if you are going to worry about leaks you might as well worry about that overweight Noctua warping your Mobo.

 

I would say not equivalent though. You will find zero evidence of the Noctua damaging motherboards, but easily find evidence of AIO's and custom loops leaking. 

But I agree that we shouldn't worry excessively about water cooling in the respect of leaks. I would agree with Rob that a custom loop is under the installers control. You have control over every joint you make and as long as none of the components are faulty and you do the job properly all should be well. And in the respect of AIO's they still leak but it seems less common these days. 

 

 

[SteveW 1,785]

32 minutes ago, Noel said:

OK then it's pretty much as I guessed w/ the big question remaining unanswered:  what then is the optimum number of cores or LPs as it were to use for P3D?   Presumably terrain texture loading is one of those background processes that I can guess is more parallel-oriented, correct?   How can one tell if one is 'oversaturating'?  I have best performance I can imagine on my nearly 6 y/o system, certainly the best setup of any, and you were largely responsible for that and the AM I use!  And I've got 8 LPs dancing happily along.  How can one tell the optimum number of cores/LPs to make available?  If I went w/ a 9900K and HT enabled I would bump that up to 10 LPs.   How can one tell if 10 LPs is better than 5 C's?

"what then is the optimum number of cores or LPs as it were to use for P3D? "

"As we add cores or LPs for the background process we notice the time taken to load the sim decreases. There's only so much data your PC can collect at once and so there's an optimum number of those cores to use. Allocate more than you need and you don't get extra performance you get less as the cache is stretched more for no extra work done, and also you get extra heat. "

In other words - if you are not getting any faster loading then you need no more LPs allocated - which I felt answered your big question.

[SteveW 1,785]

...Noel, this is what you do:

 

You know the first process described by P3D is monolithic, running the user input and rendering. (So with HT enabled) it's obvious you need an AM with one of the LPs unallocated on the first core you allocate the sim. Get a scenario loading a few times and measure how long it takes with a stopwatch. Now add pairs of LPs on subsequent cores and each time measure the time loading the same scenario - let it load once at least before a timed run. Eventually the scenario will not load any more quickly no matter how many cores allocated - this is too many. On a pushed overclock drop one or two LPs. on a cool system leave the LPs at that count.