25 minutes ago, LRW said:

Of course, having said all of that, I have the TridentZ 4400 CL18 to put in my Maximus XI Code whenever I receive my unicorn..sorry.. i9-9900K (I get those confused as they have the  same availability).  The total latency of the 4400 is lower than that of the TridentZ's with lower frequencies.  I hope it boots when I turn on XMP.

 

If the Maximus XI Code is anything like my Maximus X Code you will love it. Good choice. 

 

Don't forget to give Ai Overclock a bash, even der8auer himself says it's very good. 

Edited November 1, 20187 yr by martin-w

1 hour ago, vc10man said:

Waiting to pick Martin's knowledge base for a suitable RAM+MoBo. CPU air cooler is fixed.as per Martin's.

He’ll probably recommend G.Skill but as to what speed who knows after this morning’s discussion. 😁

Mobo? Everyone seems to go for Asus.

That Crucial white paper has been around a long time, and I found that its primary fault is that it throws a bunch of facts up on the wall, and then draws a conclusion that isn't supported by those facts...no real logical linkage from data to conclusion.  They are trying to advance the mantra that faster memory clock speeds are better, which universally holds true only if you hold the CAS latency constant.  Why?  because then they can charge premium prices for average memory by clocking it up in conjunction with higher CAS latencies.  It's analagous to the question of whether it's better to take the fast train or the slow train...faster is certainly better if the number of stops is the same, but if not, you have to do the math, because you may well find that the faster train makes more stops than the slower one, and in fact doesn't get you there any faster (or indeed maybe slower) as a result.

The memory read process involves putting the address up on the address lines, issuing the read request, then waiting for the data to be ready to be accessed on the data bus.  The process is synchronized by a sort of a metronome...the clock, and the speed at which it ticks is the memory's "speed" in MHz, or millions of ticks per second.  There is a waiting period--latency--where the memory electronics have to be given time to reliably change states and put the data on the bus to be read.  The memory controller observes this latency by counting the clock ticks from the request--CAS wait states--and then after the configured number of ticks have occurred, it reads the data from the bus.  That latency is the real measure of how fast your memory is...just as in the train example, if your clock ticks faster, but you have to wait more ticks for the data to be ready, it isn't necessarily available on the bus any faster than with a slower clock and less wait cycles.

As far as the question of whether faster memory makes a difference in P3D...I think that it can, especially with today's fastest 6-10 core CPUs.  Not in terms of frame rate, but a noticeable reduction in stutters when the CPU and GPU loads are not maxxed out.  If the sim stutters when the CPU is hitting 100%, especially on the core running the primary thread (usually core 0), then it's expected that stutters may occur due to the CPU not being able to keep up with the demand for steady frame production.  If it happens with the GPU(s) at/near 100%, same thing...the GPU is bottlenecked.  But I see stutters, especially in autogen-intensive areas like ORBX SoCal, even with CPU and GPU loads well below 100%.  I suspect those are due to the CPU being "starved" of data, perhaps when it's trying to gulp a large block of autogen objects because the memory isn't able to keep up.

I'm seeing a nice reduction in those small stutters...still get an occasional one, but much less.  Most of the conventional wisdom--that memory performance makes no difference--seems to me just parroting of experience from previous generations of hardware that were running much slower and feeding much less voracious graphics subsystems.  It stands to reason that as CPU throughput increases, memory has to keep up, or the CPU will spend a lot of time idling and waiting for data.  Memory performance hasn't been increasing nearly at par with CPU throughput, especially as more cores are brought into the mix--all of which feed from a common memory subsystem.  So I take "memory speed doesn't matter" as cliche and somewhat suspect now.  Does that mean we should break the bank on exotic high-speed DRAM?  No, but OTOH building a power system with average consumer-grade memory may prove underwhelming, too.  A balanced system, as has been said many times here, is the key.

Regards

 

54 minutes ago, Ray Proudfoot said:

He’ll probably recommend G.Skill but as to what speed who knows after this morning’s discussion. 😁

Mobo? Everyone seems to go for Asus.

Me too, Ray, that is why I am following your thread keenly. The only one. in this Forum.

46 minutes ago, vc10man said:

Me too, Ray, that is why I am following your thread keenly. The only one. in this Forum.

I’m honoured Rick. 😁 I do enjoy researching the components for a new system.

As an aside I asked Chillblast about BIOS updates and whether I should install them and they said not to. They balance the system for that BIOS version and any updates could throw it out of kilter.

1 hour ago, w6kd said:

As far as the question of whether faster memory makes a difference in P3D...I think that it can, especially with today's fastest 6-10 core CPUs.  Not in terms of frame rate, but a noticeable reduction in stutters when the CPU and GPU loads are not maxxed out.  If the sim stutters when the CPU is hitting 100%, especially on the core running the primary thread (usually core 0), then it's expected that stutters may occur due to the CPU not being able to keep up with the demand for steady frame production.  If it happens with the GPU(s) at/near 100%, same thing...the GPU is bottlenecked.  But I see stutters, especially in autogen-intensive areas like ORBX SoCal, even with CPU and GPU loads well below 100%.  I suspect those are due to the CPU being "starved" of data, perhaps when it's trying to gulp a large block of autogen objects because the memory isn't able to keep up.

QFT  And I'll repeat: Regarding memory speed, it's not about FPS... it's about the rate the system can move data.  Smooth flowing data is the goal! 😁

I assembled my 8086K system this past weekend and have been testing it since.  I chose a XI Hero Z390, 16 GB GSkill 3600 CL 15 Samsung B die, and added a Samsung 970 Pro NVMe SSD.  I've been testing this week (have not yet installed P3Dv4) and been quite happy.  Have not overclocked the memory but the CPU has no problem with Cinebench, AIDA64 and crunching the numbers on all my CAM programs running at 5.3GHz with temps in the 80's.  I could certainly put more power to it but it only requires 1.355v at 5.3 now.  I'll soon delid the CPU and begin testing again, then of course I'll go to work on the memory as well.  I'm hoping to get the memory to 4000 and CL 17... 16 if I got lucky. 🤞

Overall and so far this is a nice combo.  The 8086K is just a binned 8700K... and the latter will never be a slouch!  I paid $375US for the 8086K, which was a deal I couldn't pass on.  I had pre-ordered a 9900K (at $530) but cancelled right after the first reviews came in and delivery dates started evaporating.

Greg

Haven't we gone over this before Ray?  I wouldn't trust a memory manufacturer telling you what is better for you … sorta like a chicken asking a fox where is the best place for to eat me? 🙂  Memory manufacturers will obfuscate and confuse as much as possible to "sell" more product.

What is "frequency" … it's a signal sent at a specific interval of time … Hz cycles per second (measure of time) … MHz is millions of cycles per second.  4400Mhz means the memory clock operates at 4400 Million times per second (it's "speed").  Bare with me as this is over simplified (EE out there will probably cringe) in order to make it more (hopefully) understandable for those interested … the memory can't read or write every time a clock signal is generated (would be great if it could, but current technology can't operate that fast). 

The memory module takes time to flush/update or access the state of it's transistors (representing on/off or 1/0 and the 1/0's represent your application data) … the time it needs to do read/write operations is what CL/CAS (memory timings) is basically about … they need a specific number of cycles to operate a read or write request.  So if you increase the memory frequency/clock but don't adjust the timing/delay then read/write operations will not complete and BSOD. 

It's all about Latency (less you make the CPU wait for data the faster things gets done) … here is an older but very good article on what the various memory timing really mean, it's not "entirely" about CL, tRAS and CMD are very important also.  Memory Timings Article (good read)

Cheers, Rob.

7 minutes ago, Rob Ainscough said:

Haven't we gone over this before Ray?  I wouldn't trust a memory manufacturer telling you what is better for you … sorta like a chicken asking a fox where is the best place for to eat me? 🙂  Memory manufacturers will obfuscate and confuse as much as possible to "sell" more product.

Cheers, Rob.

Rob, I'm not sure why you're asking me about this? I followed the advice given by Bob linked on the previous page. I'm not backing Crucial's video. Not sure why you think I am.

I took advice here and then discussed it with Chillblast who advised that the 3200 memory was not worth the extra £130 over the 3000.

13 hours ago, joemiller said:

Well.. Who would've thought. After so many years I can finally (finally) say that I have  found a CPU that has  tamed P3D 4.3

I just purchased the Intel 8086K and managed to Overclock this guy to 5.3Ghz, along with 16Gb of RAM at  4266 Mhz. This along with an Nvidia 2070 card (will be getting an 11Gb soon).

I will tell you, now I can fly my sim with any settings I choose and frames remain constant at 29-30 ... smooth, and clean! Sounds too good to be true but it actually is. I managed this with Hyperthreading  Off. I actually took it all the way to 

5.4Ghz but will not hold for more than 5 minutes and BSOD. No worries, I am indeed a very happy simmer. 

What addons are you using with P3D?

3 hours ago, Ray Proudfoot said:

He’ll probably recommend G.Skill but as to what speed who knows after this morning’s discussion. 😁

Mobo? Everyone seems to go for Asus.

 

I recommend 100,000 MHz! And CAS one. Or better still RAM from Area 51. Alien tech. 😉

Edited November 1, 20187 yr by martin-w

2 hours ago, w6kd said:

 They are trying to advance the mantra that faster memory clock speeds are better, which universally holds true only if you hold the CAS latency constant. 

Regards

 

 

That's what they said though. True latency hasn't improved much so therefore faster memory speeds are better. 

Any chance of more discussions about overclockability between the 8086k and the 9700k? 🙂   Rays got his own entire thread (one of the most constructive I've seen here on AVSIM for a while) here

 

C'mon Ray - how many threads on RAM does one man need!??! 🙂 (jk)

 

 

21 minutes ago, martin-w said:

That's what they said though. True latency hasn't improved much so therefore faster memory speeds are better. 

Faster speeds are only better if they result in lower true latency.  That's really the bottom line...anyone that buys based on speed alone is the rightful prey of marketers hoping for an uninformed target audience.

Currently, true latency varies from near 8.0 ns at the high end, to over 12ns on the skunky side.  It's correct to say that the cutting edge of true memory latency has not advanced much, but the range of possibilities in retail channels is wide enough that it's still important to know where the stuff the memory manufacturers are trying to sell you fits.  There are only three manufacturers of the actual chips on the planet right now, so the differences between the best GSkill, or Corsair, or Mushkin DIMMs will be in how they bin and select the cherries from batches of the same memory chips, the quality of the board and supporting electronics, and (huge) how they warranty and stand behind their products.  A high-volume operation with a good binning selection process and a good handle on the statistical characteristics of their binned sets of chips make it possible to push the limits of the silicon and still not lose their shorts replacing DIMMs that don't perform as advertised. 

The fastest production DRAM chips are produced and warranted by Samsung/Hynix/Micron as 2666 MHz devices...a DIMM sold as anything above that is purely the result of an OEM overclocking drill...that's what we pay for when buying fast memory.

Regards

8 minutes ago, KL Oo said:

Any chance of more discussions about overclockability between the 8086k and the 9700k? 🙂   Rays got his own entire thread (one of the most constructive I've seen here on AVSIM for a while) here

C'mon Ray - how many threads on RAM does one man need!??! 🙂 (jk)

It was interesting and informative but as I’ve now ordered my system there’s no point me discussing it further. Someone might tell me I’ve made a terrible mistake when something even faster comes out next week. 😁

Anyway, I’m away from tomorrow until Sunday evening. Going to Lelystad, Netherlands for their annual FS show. Going with Pete and John Dowson so I imagine L-M will be releasing a point release of v4. 😄

1 hour ago, w6kd said:

to over 12ns on the skunky side

I like that saying, "skunky" … have to steal that from you.

1 hour ago, Ray Proudfoot said:

Someone might tell me I’ve made a terrible mistake when something even faster comes out next week.

You've made a terrible mistake, something faster did come out last week … but was sold out before it was actually available online. 😉  You worry too much Ray … say hello to Pete from me and enjoy.

Cheers, Rob.