Opinions on RAM please

[SteveW 1,792]

 

Greg

If you scroll back a bit I discuss with Rob the fractional relationship of Frequency and CL.

So apparently I am aware then that CL and Frequency are intrinsically linked.

I am aware of how these figures do not necessarily represent the final swiftness of memory transactions. None the less faster is better which is undeniable.

There seems no doubt that I am well aware of these facts.

I clearly already acknowledged these facts to be true and use those same facts in general conversation discussions as you do.

 

OK.

 

Now let’s look at typical examples offered. I use the quoted frequency, remember though that Rob says to half it when talking about latency.

Rather than make up some timings, I can take examples straight from wiki and maths from Tom’s Hardware.

CL / Frequency

17 / 36 = 0.4722

16 / 32 = 0.5

 

Converting to ns and double to show latency as I have not halved the frequency.

0.4722 x 2 = 9.44ns 3600 CL17

0.5 x 2 = 10.0ns 3200 CL16

 

In this example CL17 is faster than CL16. Plenty of examples can show the same problem. I think this is mentioned all over the websites.

In conclusion there are plenty of opportunities for the bloke in the memory shop to sell the unwary customer a slower bit of kit with a faster CL advertised (or a faster Frequency). Frequency and CL are intrinsically linked. If I didn’t make that absolutely clear enough, maybe because it’s been discussed to death already. I did not say to ignore the CL. I said it is more often the frequency (at a particular CL) going higher more often meets the requirement to a shorter time of attainable RAM Transactions. Moving down the table in the wiki shows the gap closes for less opportunity to bluff the purchaser.

Most RAM is produced by the same few manufactures. RAM is slightly different from one model to another to take into account the different types of hardware to be used with. So although there are hundreds of examples, many are similar internally. The frequency and the CL are the times the RAM is accessed. The capability of performance of the RAM must fit into that time. So we often see several CL and Frequency possibilities for the same product.

That can lead to branding of RAM with emphasis on CL when Frequency is more important. We might buy slower RAM at CL16 than CL17. And there are lots of other examples. Mostly, looking at the tables, as we go up in frequency we see the desired step in performance.

So I consider that I am misunderstood by those that don’t fully understand the issues at hand.

 

OK.

 

Now, in response to Greg’s very nice PM enquiry about Ideal Flight: There’s no need to PM me Greg, there is plenty of help and advice on the codelegend forums. I agree, the graphs are indeed ‘Pretty’. I can explain the answer to your major appeals here to save more enquiries, This is a great opportunity to explain IF Pro in more detail – thank you:

The graphs can plot fps along with more than 30 other values from the simulator. With the CPU monitoring Ideal Flight graphs can also include the percent load on each LP at the same time.

It does that by analysing system “performance counters” after each time period. The ratio of count and time is the performance factor used. Consider with HT since 100% on both Logical Processors of a core showing in Task Manager does not equate to 200% core use, that’s 100%, so they must be at 50% each. This is one consideration of the CPU monitor.

Some programs can consume counters with no work done so it important to know that when analysing the performance of the simulator cores, that these are used counters. However there can still be other processes working to consume counters without work. We can avoid using those while testing and assessing performance conditions since they don’t consume performance. Rob brought this up with quotes off the internet, to show that a CPU monitor needs to take care when monitoring CPU performance. That’s just Rob being nice as he knows about IF.

Ideal Flight Pro compares Simulator performance and CPU performance in back to back tests.

Knowing how the sim and PC work, and how to test and measure anything, or the principles used, are not in the instructions and is not something that is easily understood. The homework must be done by the operator.

Standard deviation is used throughout:

 

How to test the simulator and CPU and the System is complicated, misunderstanding is commonplace. There are no instructions other than on my forums.

This part of Ideal Flight is not complete and had been shelved a little over a year ago when development focussed on helping the Blind Pilot community. The requests to improve Ideal Flight’s accessibility was overwhelming so I had to postpone the CPU monitor developments.

There’s no intention to have the monitoring system ratified by some kind of technical council. Basically, IF gathers the data from the system and makes a graph that can show lots of things about performance of the system, and many values from the Simulator like distance travelled and temperature at altitude.

These are all system and simulator values collected, so what it does do very well is enable back to back comparisons which is all we need.

Tests can be compared against each other, meaning there is always tight correlation in tests from one aspect of system setting to the next. Tight comparisons can be measured with an arbitrary length of string. We do not need to know the length of string, and I simply don’t need to have to educate anyone on these basic technicalities in a PM. It is a matter of education for the individual.

The reason I said the CPU monitoring is “experimental” in IF Pro is that it’s not part of the sales remit of the package. It’s a free module which happens to be included with the P3D v4 package due to having a lot of requests for it. So naturally I’ve looked into including it with Ideal Flight. I like it, as do others, because it’s very good for comparisons in testing since It is attached to Ideal Flight which can be set up as a test harness. That goes a very long way to making very solidly repeatable tests. If back to back comparisons are desired.

Graphs logging simulator values in Ideal Flight is a very minor section of the package, but it’s there to have a play with. The product is a massive complete package that fits many styles of flight simulator use. Parts can be used separately or together as a whole complete package.

This complete package attracted the Blind Pilot community as it puts everything in one place, and reports what they do. That’s also very good for any flight sim enthusiast. Ideal Flight is in development all the time and will continue to be developed so long as it is desired.

Ideal Flight can be purchased from Flight1 with a refund available. I urge anyone with IF to have a look at the updates on the Ideal Flight forum New Builds section and in the Accessibility phase 2 discussion in the General section. Please check any notes in the support forum and report via the contact page on the website, by PM at AVSIM, or join the codelegend forum. Please contact me for forum membership.

 

[SteveW 1,792]

15 hours ago, brucets said:

I understand very little of this, but I have had Hyper Threading turned off for several years due to advice from long ago. In this thread Rob and Steve seem to disagree on it's usefulness. So what is it? HT on, or HT off???

My PC specs:

    Cooler Master MasterCase 5 Pro chassis   
    ASUS ROG Strix Z270E Gaming system board
    Intel Core i7-7700K Processor 4.2GHz(unclocked)
    ASUS GeForce GTX 1080 Ti STRIX ROG Overclocked 11GB GDDR5X
    1000W Gold power supply with CableMod's ModFlex sleeving
    32GB G.Skill DDR4-3200 RAM
    Intel 600P Series 256GB M.2 SSD's in RAID 0 (512GB)
    2TB WD storage drive    
    16x Blu-ray DVD Burner Drive
    10/100/1000 Network
    802.11ac Wireless
    Bluetooth 4.1
    Microsoft Windows 10 Pro 64-bit 

Bruce

I just saw Steve's last post(we posted simultaneously). I guess I 'll try his suggested test. Actually my FSXSE is running extremely well and loading times are quite short with HT off.

 

Yes. Good idea is to check if HT increases your loading speed, decreases loading time of the scenario shows the GPU is loaded faster. This will improve again with DX12. If it does not the system is pushed to a limit already by the CPU core count.

We can check that with NO AM HT ON vs HT OFF Loading time only. We are not interested in fps at this point.

[Rocky_53 1,579]

On 2/12/2020 at 4:34 PM, brucets said:

I understand very little of this, but I have had Hyper Threading turned off for several years due to advice from long ago. In this thread Rob and Steve seem to disagree on it's usefulness. So what is it? HT on, or HT off???

My PC specs:

    Cooler Master MasterCase 5 Pro chassis   
    ASUS ROG Strix Z270E Gaming system board
    Intel Core i7-7700K Processor 4.2GHz(unclocked)
    ASUS GeForce GTX 1080 Ti STRIX ROG Overclocked 11GB GDDR5X
    1000W Gold power supply with CableMod's ModFlex sleeving
    32GB G.Skill DDR4-3200 RAM
    Intel 600P Series 256GB M.2 SSD's in RAID 0 (512GB)
    2TB WD storage drive    
    16x Blu-ray DVD Burner Drive
    10/100/1000 Network
    802.11ac Wireless
    Bluetooth 4.1
    Microsoft Windows 10 Pro 64-bit 

Bruce

I just saw Steve's last post(we posted simultaneously). I guess I 'll try his suggested test. Actually my FSXSE is running extremely well and loading times are quite short with HT off.

 

I won't pretend to understand everything that is being discussed in this thread Bruce, I have no inclination to do so either. However, what I will say is that I read on the P3d forum page, that was a post by one of the devs (Beau?) that P3d was made to run with HT off. Following that advice I did more than a little testing and I can quite categorically say that my system, no body elses, but on my system, it ran definitely smoother than with HT turned on. I also found it best to leave any affinity mask setting vacant from the cfg. Take from my experience what you will.  

[lownslo 592]

19 hours ago, SteveW said:

With the CPU monitoring Ideal Flight graphs can also include the percent load on each LP at the same time.

Very nice marketing spiel.  Of course, one wonders how the CPU monitoring tool can be free when the only way to use it is to buy the parent program...

Greg

[kevinfirth 1,396]

7 hours ago, Rockliffe said:

I read on the P3d forum page, that was a post by one of the devs (Beau?) that P3d was made to run with HT off.

I would wager that the root of that comment is the fact that P3D is based on ESP.  ESP was of course designed and created when available CPUs did not utilise HT.  So by definition, P3D in its earlier variants was not made to use it.  However, CPU tech has advanced, and there are demonstrable benefits to HT (as @SteveW repeatedly explains).  This is why LM include the ability to tweak P3D in the cfg to use an AM.  That they do this in the official documentation demonstrates P3D IS now capable and can benefit (in certain circumstances) from HT on. QED.

[SteveW 1,792]

5 hours ago, kevinfirth said:

I would wager that the root of that comment is the fact that P3D is based on ESP.  ESP was of course designed and created when available CPUs did not utilise HT.  So by definition, P3D in its earlier variants was not made to use it.  However, CPU tech has advanced, and there are demonstrable benefits to HT (as @SteveW repeatedly explains).  This is why LM include the ability to tweak P3D in the cfg to use an AM.  That they do this in the official documentation demonstrates P3D IS now capable and can benefit (in certain circumstances) from HT on. QED.

Agreed. The problem with HT is folk continue to ignore how it is set up and then come here and say hay, I just rebooted into HT and didn't apply the AM required.  Also they fail to understand that fillin is as important as fps. They didn't do their homework. Obviously you have Kevin.

[SteveW 1,792]

7 hours ago, lownslo said:

Very nice marketing spiel.  Of course, one wonders how the CPU monitoring tool can be free when the only way to use it is to buy the parent program...

Greg

Free to IF license holders.

I'm hiolding on to your aggressive insinuating and P-Taking PMs so that I can forward them on to AVSIM staff if you continue your abusive behaviour.

[SteveW 1,792]

Bottom line here is that a small clientele like to trash other's quality posts because they don't like the poster. Or that they worry that they posted other advice and are embarrassed.

They don't reciprocate with upvotes to quality posts, they muddy the waters.

I'm not going to put up with insulting and abusive PMs guys, so beat it.

[Luke 782]

9 hours ago, kevinfirth said:

I would wager that the root of that comment is the fact that P3D is based on ESP.  ESP was of course designed and created when available CPUs did not utilise HT.  So by definition, P3D in its earlier variants was not made to use it.

When the first version of FSX that divided work between between cores was released (SP1) the vast majority of "multi-core" systems at the time were Pentium 4s (Northwood core) with Hyperthreading. Cheap multi-core wasn't possible in the P4 days since you needed to do multi-socket.

Cheers!

[SteveW 1,792]

Consequently HT is no joke and is necessary to virtualise more servers per core in the cloud.

It's an optimisation of a single core. Some processors have maybe 4 or more HT LPs per core.

On our desktop processors we have two per core HT enabled.

[SteveW 1,792]

I downloaded a popular tool a few months back that showed latency and memory bandwidth in charts with already recorded example. At this point in time I don't remember what it was. It was try before buy and I deleted it after a few goes. But anyway maybe someone might know of it, I would get it again and have another look, can't remember it just now. In the charts it showed latency and also another chart showed memory bandwidth. There was an example xeon 20 core at the top of the bandwidth chart.

What I found was that the bandwidth seemed to show a different story.  On one PC here a latency showed measured as 40ns but this was actually the fastest PC and showed the highest bandwidth in the table, higher than those showing latency some showing less that 10ns.

So the tool shows a memory latency of 40ns at the top of the table with the highest bandwidth.

Maybe someone knows what's going on there?

 

[SteveW 1,792]

Thanks Rob. I will have to make some time to investigate. Can it be some systems might show double latency, perhaps if quad channel or some other system configuration?

[SteveW 1,792]

AIDA64 Extreme trial version was the one I tried Rob. I got these figures from it. No idea if accurate:

Intel 18 core 2933 CL19 12.96ns
Read 88535 MB/s
Write 79607 MB/s
Copy 80493 MB/s
Measured latency 78.6

Threadripper 2933 CL16 10.91ns
Read 68474 MB/s
Write 73605 MB/s
Copy 67171 MB/s
Measured latency 71.2