Is there any benefit getting more than 32GB of RAM for DCS?

[BitMaster]

The CPU behaves as advertised by AMD. If you use 2 modules per channel max supported speed is 3600MT/sec. 1 module per channel is 5200MT/s.

Anything faster is a "may or may not work". The faster, the more modules, the lower the latency the lower the chances of success.

Coupled with a board that by definition ( trace layout for 2nd slot per channel ) doesn't like 4 modules at any speeds higher than 3600 MT/s and you are stuck.

 

What you experience is the guaranteed performance, not a failure or shortcoming. What you ask for is not guaranteed and may or may not work depending on your particular choice of

hardware.

Check your boards QVL if it supports 4 modules at 6000 MT/s, I doubt it does.

[dugite57]

It's all explained in this youtube, jump to the 3 min part if time poor.

 

[kksnowbear]

Which is why the person who designed the system should've been aware - and should stand behind it, as I described.

Doesn't change anything: Like I said, if I sold it, I'd make it work as I represented or give a refund.  You *can* run 4 modules, and it doesn't have to be at 3600:  Below is a picture of a build I did earlier this year, running MemTest 8.  64G, all four slots populated (second pic).  Running 4800** (not 3600) and if I'd played with it more I am fairly certain I'd get more out of it, I just didn't have time; the client was happy to get better than what AMD says is officially max and was glad to take the system without needing to wait for me to fiddle with it more.

** Yes, I know it's 6000 RAM, but that ain't the point  lol...AMD says not more than 3600 with 4 modules - which clearly isn't always true.  It simply depends on a number of factors.

 

Edited September 23, 2023 by kksnowbear

[dugite57]

So I'll rephrase my conclusin, With AMD Zen 4, 2 sticks good, 4 sticks bad.  Yes, 4 sticks will work, but the probability is they'll run at or significantly closer to the guaranteed 3600hz.  Wjen running at 3600 my experience is DCS stutters, even with 64gig. I'm currently running 2 sticks at EXPO 6000hz with a dramatic decrease, in stutters.  Should be noted that I'm using 7800X3D Nvidia 4900 and only "fly" in VR with quest 2.

There's a better youtube where a guy from ASUS explains how Zen4 handles Ram and the reasons why they have 4 slot mobo's, however I can't locate it.  It explains how these CPU's are 2 channel and not 4 etc etc...

[kksnowbear]

Yes, I am aware of all that...yet the results I've shown above are conclusive proof that the official specs are not absolute.

And 4800 is a damn sight beyond the 'guaranteed 3600'.  A full 33% more, to be exact, which is definitely significant (and in fact just as close to 6000 as it is far from the 'guaranteed 3600', so hardly "significantly closer').

In case you hadn't noticed, that's a Zen4 CPU in the picture I posted.  Looks like "AMD Zen4, 4 sticks = outperforming factory specs" to me.

There's also no proof at all that any system that runs 3600 will have issues with stuttering.  I'm sure, if I were to try, I could find someone who runs DDR5 at 3600 and has no issues with stutters.  What makes me so sure?  Well, among other things, at least according to AMD, that's the max speed you can get with 4 modules - but I'd bet there are at least a few people running DCS with four modules on an AM5 board without having stutters.

If running 4 modules means no more than 3600, and 3600 means stutters...then logically, everyone with 4 modules would have stutters.  Somehow I don't really think that's the case.

Edited September 23, 2023 by kksnowbear

[Eugel]

5 hours ago, kksnowbear said:

If running 4 modules means no more than 3600, and 3600 means stutters...then logically, everyone with 4 modules would have stutters.  Somehow I don't really think that's the case.

 

Indeed.
And it´s debatable if the performance difference between DDR4 and DDR5 is really noticable in gaming where Ram is rarely the bottleneck.
And even if that Ram is not running at it´s full capacity but "only" 3600 MT/s, having 64 GB over 32 Gb is still better and should not result in stuttering.

@dugite57 did you buy a 4x16 GB kit or did you just buy 4 seperate 16 GB Ram sticks ?
It rather sounds to me like you possibly got Rams with different clock speeds/timings or even a broken stick or broken slot on your mainboard.

[kksnowbear]

8 hours ago, Eugel said:

And it´s debatable if the performance difference between DDR4 and DDR5 is really noticable in gaming where Ram is rarely the bottleneck.
And even if that Ram is not running at it´s full capacity but "only" 3600 MT/s, having 64 GB over 32 Gb is still better and should not result in stuttering.

Absolutely.  (It's worth mentioning here that there are 64G kits in two modules - so it is attainable, even within the AMD-specified limits)

8 hours ago, Eugel said:

@dugite57 did you buy a 4x16 GB kit or did you just buy 4 seperate 16 GB Ram sticks ?
It rather sounds to me like you possibly got Rams with different clock speeds/timings or even a broken stick or broken slot on your mainboard.

I'm not sure it's bad RAM, or even dissimilar modules (although I believe it is accurate to say you can't actually get an AMD EXPO kit with 4 modules due to the discussed limitations)...

I actually wonder if it doesn't have more to do with the quality of the motherboard, and/or that of the RAM itself, and/or the combination of RAM/motherboard. 

That being the case, it is most likely as BitMaster said above, not something you could work around - with that hardware.  Which is what I was saying: Whoever designed/spec'd the build is responsible, and if it was a shop then they should be willing to work out a solution.

I'm trying very hard to be diplomatic and avoid a pissing contest if I say anything too specific about the hardware itself - but again, if I sold it from my shop, I'd stand behind what I sold.  This support is the single most important reason it matters who does the build/who you work with, especially if you don't have experience.

My original comment (in this part of the thread) was "There are actual technical reasons it can be problematic, but just as well, there are real technical solutions. "

The technical reasons include hardware selection, and the solution - albeit in retrospect - might well involve different hardware.

Still, for me, comes down to who's designing the system, particularly if it was a "shop" (though it's not clear in this case if that means just a retail parts store, or a place that actually builds machines).  Retail parts salespeople are usually morons, but a business that builds machines should know better.

Edited September 23, 2023 by kksnowbear

[BitMaster]

Having a 2nd slot for a channel is mildly said a "complicated" idea first off, there is no denying in that. There are basically TWO ways of solving the problem of how to connect a 2nd slot aside the first one, it's not a true channel, it's a 2nd slot to one channel. 

There is T-Topology and Daisy-Chain Topology, both have advantages and disadvantages but none is beneficial for higher clocks or lower latency, it's the other way around. You have to let loose and talk slower.

 

Same context in other words:

IF you plan to max out your RAM capacity for a given platform and are forced to use the 2nd slots per channel, stay away from the top tier modules, it's wasted money.

IF you can achieve your desired amount of RAM with 1 Dimm per channel only, do that, you can go all the way into MHz and low latency.

If you are somewhere in the middle, like I am with my current config from machine in sig, you have to choose a motherboard that can cope with 4 modules ( watch reviews )

and find the desired, matching RAM. Fir example, I could have gone 2 modules á 32GB but those have a bad latency, I wanted B-Die but B-Die doesn't come in 32GB modules

as the only make 8Gbit dies, so one of my 4 16GB Dimms is already dual ranked with 16 dies. B-Die loves oc, my mobo allows 4 modules to be oc'ed up to the limit where the AMD Infinity

Fabric has it's limit, 1800MHz. I didn't know before if it works out, I was lucky, XMP worked OOTB and giving 1.48v let's them run at 3600 with same CL14 settings...just lucky !

My other PC, same B-die, 4 modules with 8 8Gbit dies each, 4x8GB, will not run at 3600, only 2 modules allow XMP and even go well beyond 4000. Just with 4 modules, anything over

3000 makes it go BSOD in a matter of weeks. Tried many many nights to fix that, it's the poor Board RAM layout to blame, the RAM works superb, just not in that board with 4 modules.

 

With DDR5 you can easily do 64GB in 2x32GB and stay away from the headache. For 128GB I would also suggest 2x64GB module once available for mainstream. The DDR5 limit on

Desktop is 256GB with 4 modules, once those modules/dies are available to the market.

[dugite57]

1 - The memory is: Two of G.Skill 32gb kit {2x16) DDR5 Trident Z5 Neo AMD EXPO RGB C36 6000mhz.  Serial numbers of both kits were pretty close for whatever that's worth.  My understanding is you can't by 4 stick kits of EXPO at this stage.

2 - Yes the PC shop I got it from sold me a pup with regards to RAM.  My dealings and experience with the joint were probably the worst consumer experience I've had.

3 - Set ram to 3600 and I get DCS stutters. Set ram to 6000 and I don't.  This is in VR. It's probably fine in 2d.

4 - So, for those who don't want to spend untold time tweeking and going up blind alleys, who simply want to hit EXPO and have it work, who are not expert memory tweekers etc, I strongly recomend 2 sticks.  I do intend going back to 64 gig but will be doing it with 2 x 32gig kit. Yes, I know some can get 4 sticks running sweet. My recomendation is for those who don't want to have the potential hassels associated with setting up 4 sticks.

[kksnowbear]

On 9/25/2023 at 3:14 AM, dugite57 said:

1 - The memory is: Two of G.Skill 32gb kit {2x16) DDR5 Trident Z5 Neo AMD EXPO RGB C36 6000mhz.  Serial numbers of both kits were pretty close for whatever that's worth.  My understanding is you can't by 4 stick kits of EXPO at this stage.

2 - Yes the PC shop I got it from sold me a pup with regards to RAM.  My dealings and experience with the joint were probably the worst consumer experience I've had.

3 - Set ram to 3600 and I get DCS stutters. Set ram to 6000 and I don't.  This is in VR. It's probably fine in 2d.

4 - So, for those who don't want to spend untold time tweeking and going up blind alleys, who simply want to hit EXPO and have it work, who are not expert memory tweekers etc, I strongly recomend 2 sticks.  I do intend going back to 64 gig but will be doing it with 2 x 32gig kit. Yes, I know some can get 4 sticks running sweet. My recomendation is for those who don't want to have the potential hassels associated with setting up 4 sticks.

Well, I am truly sorry for your woes, but we must caution ourselves against inaccurate advice in public forums, as it misleads people who (often) tend to look at it as gospel.  I understand what you're trying to convey, and I appreciate your making the distinction.

Interestingly the memory you cited above is very similar to what I used and got four sticks to run just fine (see pic below).  In fact, I think one of the few differences in yours and mine is that mine is not even EXPO, yet it worked OK in my case. I was trying to avoid being too specific, so as not to step on toes - but I honestly believe it's the ASRock motherboard that's the biggest (if not only) problem here.  Just my opinion.

My point about the shop was that you shouldn't be without recourse.  If they talked you into something, they should make it right - up to and including a full refund on *everything* they sold you.  (If you had some parts already, got them elsewhere, etc...well, can't fairly blame that on the shop.  Again, I ran memory like that at higher than 3600 no problem).

I think your advice is perhaps the best: Use two modules, avoid the potential for problems.  Sorry if it seems I'm arguing with that, because I'm not.  I'm just saying that there's a lot more to it (and, as a professional system builder myself), that it matters who you work with.

FWIW I still disagree that 3600 will automatically cause stutters, regardless of VR or not.  Something else is going on - in my opinion.  As I explained previously, if that were true then logically everyone with 3600 (even two sticks) would have stuttering - and I just really don't think that's true.

Good luck.

The RAM in the build I showed earlier, which successfully ran 4 passes (6 hours) in MemTest8 at 4800 is shown below.  That's in addition to FireStrike/TimeSpy stress tests I ran, countless benchmarks, and (perhaps most important) zero call-backs from the client.  I"m actually fairly sure it's running at rated speed/6000 ever since, but I can't readily prove that ATM. I don't have another AM5 setup right now, so it would mean asking that client to bring his back...ain't gonna happen

Edited September 26, 2023 by kksnowbear

[BitMaster]

From my understanding you can't compare 3600-DDR4 to 3600-DDR5 as I am very confident that the DDR5 kit would run at a far higher true latency and that may cause the stutter.

We would need the timings for 3600 with DDR5 and DDR4 to calculate the true latency in nanoseconds and compare those, Apples to Apples.

 

Buying RAM remains rolling the dice to some extend

[Eugel]

8 hours ago, BitMaster said:

From my understanding you can't compare 3600-DDR4 to 3600-DDR5 as I am very confident that the DDR5 kit would run at a far higher true latency and that may cause the stutter.

We would need the timings for 3600 with DDR5 and DDR4 to calculate the true latency in nanoseconds and compare those, Apples to Apples.

That could be it, timings for DDR5 modules are about twice as high as for DDR4 modules (standard DDR4 is 16-18, DDR5 is 36 and up). Usually, that is offset by much higher clock speeds, but if the mainboard clocks those down to 3600 MT/s, that´s the same speed as DDR4, just with much slower timings.
I´m beginning to understand why many magazines wrote that DDR5 wouldn´t have any benefit for gaming... but nowadays, putting RGB lights on the sticks is more important ...

Still, that would mean that most people with 4 modules should experience stutters, so I wouldn´t rule out the mainboard yet.

[kksnowbear]

DDR4-3600 CL18 is 10nsec absolute or 'first word' latency.

He's got CL36 DDR5 now; running at 3600 with default timing, that's *double* the absolute latency (20ns)

A lot of people went to DDR4-3600 CL16 as a 'sweet spot', which is 8.88ns, faster still than the CL18 stuff.

I believe this is among the reasons early DDR5 reviews were not very flattering.  DDR4 with its inherently lower CAS levels will often outperform DDR5 with CAS Levels up around 36(+).  Even DDR5-8000 at CL36 still has higher latency (9ns) than 4-3600/CL16.  It's not a new thing, but basically the chip manufacturers can really only increase the speed of the 'devices' on the memory sticks with a proportional increase in CAS level.

8 hours ago, Eugel said:

That could be it, timings for DDR5 modules are about twice as high as for DDR4 modules (standard DDR4 is 16-18, DDR5 is 36 and up). Usually, that is offset by much higher clock speeds, but if the mainboard clocks those down to 3600 MT/s, that´s the same speed as DDR4, just with much slower timings.
I´m beginning to understand why many magazines wrote that DDR5 wouldn´t have any benefit for gaming... but nowadays, putting RGB lights on the sticks is more important ...

Still, that would mean that most people with 4 modules should experience stutters, so I wouldn´t rule out the mainboard yet.

About the stuttering with 4 modules: Precisely.

About the RGB lighting...yeah, it's unfortunate but I know exactly what you mean.

Still not clear if the shop only sold the memory or the motherboard too...if the latter, I'd be for getting a refund myself, and go with a different board.  It might not make things better - but it isn't likely to be any worse!

I had a totally different experience with an Asus board and memory that was very similar.

And I think all the hype about the burning CPUs was exactly that - a couple of instances where loudmouth 'reviewers' sensationalized a very (very) isolated fraction of the whole, and made a public spectacle out of it - mostly I'm sure to further their own notoriety as opposed to any other goal.  These 'influencers' have gotten to the point where they're not much better than the news media, which can always be trusted to sensationalize everything.

Edited September 26, 2023 by kksnowbear

[blkspade]

The funny or rather annoying thing about RGB modules is that with at least G.Skill (Ripjawz vs Trident)I think that put better subtimings in the XMP profiles of the RGB kits. I usually steer clear of the RGB kits, but I had the opportunity to compare my old non-RGB DDR4 3600 CL18, vs someone else's RGB kit with the same advertised latencies and likely even the same ICs, but there was a repeatable 10ns-12ns difference in AiDA 64 latency test. I used to assume you're only paying for the RGB when the modules should be otherwise the same, but that might not be the case.

[BitMaster]

It might just have been bad luck in the silicon lottery or maybe another batch of the same spec IC that just runs better due to unknown reasons for us mortals.

I wouldn't nail it on the RGB premium. You would need a whole batch of each to verify this.

[kksnowbear]

On 9/30/2023 at 3:01 PM, BitMaster said:

It might just have been bad luck in the silicon lottery or maybe another batch of the same spec IC that just runs better due to unknown reasons for us mortals.

I wouldn't nail it on the RGB premium. You would need a whole batch of each to verify this.

+1

[dburne]

Just expanded mine from 32 GB (2x16GB) Gskill 3200 MHz CL14 to 64 GB Corsair Vengeance (2x32GB) 3200 MHz CL16 .

Will be curious to see if any difference at all in my DCS SP flying. 

[dugite57]

So after my 4 sticks debacle and in order to feed the insatiable DCS beast, I've decided to go from 32gb 2 x 16 to 64gb 2 x 32.  MB is an Asus X6760E-Plus Wifi. CPU 7800X3D. I've looked at the QVL and before I take the leap I'd like some advice on whats the fastest from the selections below. It's the timmings that I'm vauge on.  

ADATA    AX5U6000C3032G-DCLARWH    2x 32GB    XMP/EXPO    6000    6000    DS    Hynix A    30-40-40-76    1.35
G.SKILL    F5-6000J3238G32GX2-TZ5N    2x 32GB    EXPO    6000    6000    DS    SK Hynix    32-38-38-96    1.4
KINGSTON    KF560C36BBEAK2-64    2x 32GB    XMP/EXPO    6000    6000    DS    HYNIX A    36-38-38-80    1.35

If anyone knows of something else out there that is better and works on this mobo I'd be happy to consider it.

Thanks in advance

 

[kksnowbear]

Of the three sets you listed, the ADATA would be 'fastest'.  The numbers to pay attention to are the 'speed' (in your case, all three are 6000) and the first number of the nn-nn-nn-nn combination.  That first number is called CAS Latency, and it's a factor in performance because it is proportional to the time between data being requested from memory to the memory actually beginning to respond (this is sometimes called 'first word latency').  CAS stands for Column Address Strobe, but that's essentially academic detail; if you're really interested you could look into how PC memory works.

Anyhow, the speed and CAS latency are the two numbers that affect overall performance most.  Speed is usually obvious in all the numbers; Latency is almost always the first of those four number in the group of four with dashes.

Ideally, you want the lowest number in CAS Latency, and the highest speed - though in practice it is always a compromise between the two.  Often you'll see that as speed increases, so does CAS Latency, so it's not automatically faster just because the speed is higher.

In your example, the three sets are all 6000; the CAS numbers are best on the ADATA modules (30) followed by the G.Skill (32) and highest is the Kingston (36).  So the ADATA set has the lowest latency.

The other numbers are also factors, but if BIOS is set to read the values programmed into the memory, they'll be set according to their specs, and you don't really need to worry much about the other numbers.  There are also yet further numbers - collectively, the 'secondary' and 'tertiary' timing values, but again, this is all programmed into the modules and the BIOS will use these values if it's set to do so (i.e., Intel XMP or AMD EXPO)

It's probably worth mentioning that the differences we're discussing here are incredibly small - and it is genuinely unlikely you'd ever notice or be able to tell one of these sets from the others if you're using the machine.  There is a measurable difference, yes - but the likelihood it's noticeable is pretty much zero.  There would have to be a much (much) bigger difference in the speed and CAS Latency before you'd actually notice/see/'feel' the difference - and I think even then it's arguable.

What usually winds up being the case is that, changing from a system that runs say DDR3 to a much newer system that runs say DDR5, the whole system is going to seem much faster though memory is a small part of that difference.  By comparison, the performance of a system that is otherwise identical isn't going to vary a great deal just because of memory timings.

HTH Best of luck

Edited November 6, 2023 by kksnowbear

[dugite57]

Great, thanks very much for the advice. 

[BitMaster]

There is a GSkill 6000-Cl30 out there, it’s on my list for a similar 7800x3d built I am ordering for a friend.

i personally would not take Adata, they have been cheating in the past with Ssd IC‘s…. 

[kksnowbear]

4 hours ago, BitMaster said:

There is a GSkill 6000-Cl30 out there, it’s on my list for a similar 7800x3d built I am ordering for a friend.

i personally would not take Adata, they have been cheating in the past with Ssd IC‘s…. 

I think it's reasonable to say that pretty much every manufacturer out there has been caught doing something questionable at one point or another...changing internal components without changing external part/model numbers, stuff like that (and other stuff).

Nothing new, really - and from the manufacturer's spec, nothing wrong with it at all.  They 'reserve the right' to make changes, typically to meet production demand.  And I believe that, at the end of the day, it is their prerogative - as much as consumers may not care for it.

Is it questionable? Yes.  Is it 'good business'?  Maybe, maybe not.  I had a boss once tell me that "Good business is whatever keeps us IN business".  "Good" means different things depending on your perspective

I personally have used ADATA's XPG RAM on perhaps a dozen builds by now (probably more by now).  Never had any issues.  (Interestingly, I've actually had to replace a couple of Corsair modules that failed over the same time period - take that for whatever you will).

It's also important to realize that, of the choices given, it does have the lowest CAS latency - and it appears all the modules specified use Hynix devices (actual chips) anyway.

I would agree that if everything were equal, including price and all features, I'd probably pick G.Skill over ADATA.  But that wasn't specified as part of the criteria in this case.  Without actually looking, I would venture the G.skill kit with CL30 is going to cost more than the ADATA (I could be wrong).

[BitMaster]

Sure, just that the specific device underperformed compared to the one sent to reviewers and to initial customers. Nothing wrong with it as long as performance is ~95ish% but that wasn’t the case. 
Anyway, it’s an option among many others and one needs to decide wisely.

[kksnowbear]

19 minutes ago, BitMaster said:

Sure, just that the specific device underperformed compared to the one sent to reviewers and to initial customers. Nothing wrong with it as long as performance is ~95ish% but that wasn’t the case. 
Anyway, it’s an option among many others and one needs to decide wisely.

Yup, they've all been guilty of 'cherry picking' review samples at one point or another.  In fact, it got to be such a problem years ago, that reviewers who could actually afford it went out and bought *retail* examples for their tests, just to make sure they were getting same stock as anyone else.

Problem is, lots of 'reviewers' don't actually have budget for buying all the stuff they review, and if they don't take samples sent by manufacturers, they can't afford to keep doing reviews.  OTOH, If they do take manufacturer's samples, they are at risk of cherry picked examples.

Again, none of this is new.  Yes, choose wisely But even then you have to acknowledge, this is their game - a typical DIY consumer buys memory perhaps, what? 5 times in a lifetime? 10?  Meanwhile, the manufacturers build/sell thousands upon thousands every day...they know what they're doing to stay profitable, else they'd be out of business.

...doubtful you're going to 'catch' them every time they do something questionable.  I would readily assert they've gotten away with (far) more than they've been caught at.

Edited November 7, 2023 by kksnowbear

[motoadve]

On 9/22/2023 at 7:39 PM, dugite57 said:

So I'll rephrase my conclusin, With AMD Zen 4, 2 sticks good, 4 sticks bad.  Yes, 4 sticks will work, but the probability is they'll run at or significantly closer to the guaranteed 3600hz.  Wjen running at 3600 my experience is DCS stutters, even with 64gig. I'm currently running 2 sticks at EXPO 6000hz with a dramatic decrease, in stutters.  Should be noted that I'm using 7800X3D Nvidia 4900 and only "fly" in VR with quest 2.

There's a better youtube where a guy from ASUS explains how Zen4 handles Ram and the reasons why they have 4 slot mobo's, however I can't locate it.  It explains how these CPU's are 2 channel and not 4 etc etc...

Same system as yours, bought 2 more sticks of RAM, so a total of 4, no improvement at all.

Ordered 2 sticks of 32 Gig to replace them, hope to see an improvement, this are G.SKILL Trident Z5 Neo RGB Series 64GB (2 x 32GB) 288-Pin PC RAM DDR5 6000