Poor Quad Channel Performance

[Nik81]

First of all good to be here, haven't purchased anything from OCUK in a couple of years but i hope that's not an issue.

Anyways i decided to post here because GB doesn't even reply in their forums so perhaps someone here can give me an idea of what's happening.

I own an 7900X along with an GB X299 Aorus Gaming 9 and for the past 3 months or so i was dumb enough to have my RAM (i have two kits here, Crucial Ballistix Elite 4x16GB and Crucial Ballistix Sport 4x8GB) configured in Dual Channel mode instead of Quad Channel.

So 3 days ago when i found out i configured them in Quad Channel Mode and to my surprise my benchmark results on Passmark, MaxMemm2 and Sandra Platinum Lite were around 20-30% inferior to when i had the RAM configured in dual channel mode.

I also tried Memtest from a bootable CD, all tests passed on both kits.

Ideas? This has been driving me crazy. I know the Aorus Gaming 9 is not the best X299 mainboard out there but Quad Channel not working properly?

Anyways thanks in advance for any input.

 

[Nik81]

Aren't the sticks meant to be identical. For preference a quad channel kit but if not, the same model.

The further you go from a tested kit of 4/8 sticks the more risk they won't work together as I understand.

Wait nvm, thought you were mixing two different types not 4 sticks of each.

Yeah, 4 identical ones of each of the two models i have here.

So am i the only one to ever have this issue?

 

[Nik81]

@GIGA-Man One for you buddy - Time to poke those on the giga forums

@Nik81 - Giga-man is the ocuk gigabyte rep and a top bloke! Also what are you using to test memory performance? I could run some on my quad channel system but being AMD im not sure they would even be remotely comparable.

Passmark Performance Test 9, MaxMemm2 and Sandra Platinum Lite.

I have checked numbers by other people plus like i said in dual channel mode i get 20-30% higher numbers.

 

[Nik81]

Higher numbers on what though? Latency or throughput? I'd almost expect quad channel to have a latency penalty over dual channel. If you could do some screenshots with the difference I would expect many more people to weigh in

For what it's worth you would expect quad channel to have more bandwidth than dual but at higher latency, dual would have lower latency and lower bandwidth at the same frequencies.

That's the issue, i get less numbers in EVERYTHING.

For example PassMark got me an overall score of 2395 with Quad and 3098 in Dual.

MaxMemm2 read was above 11k with Dual now its at 8k with Quad. Likewise write was at over 22k with Dual now its at just over 18k with Quad.

Even worse results in Sandra Platinum Lite, around 48Gb/s with Dual and 16Gb/s with Quad.

This makes no sense. I hope the GB guy knows if there's an option in the Bios i haven't chosen or something like that.

(Don't know why my text got black)​

 

[Nik81]

Question- do the clock speeds/timings step down significantly in quad channel? Those numbers do look way, way off.

Nope, same timings and speed (XMP is used in both cases).

Someone told me that the mainboard may be changing timings on its own and i may not be able to see that but both CPUZ and MaxMemm2 report the same timings.

 

[Nik81]

Have you tried the two sets separately in quad channel? That would give you something to compare the results with all 8 sticks to.

Yeap, more or less same performance levels (even though one kit is at 3200MHz and the other at 3000MHz).
I can try to pair them to use all 8 modules but seeing these are different models i don't know if they will even work.

Really odd some screenshots CPU-z etc from both quad and dual would be helpful I am sure!

Coming up.

 

[Nik81]

As the motherboard has 8 slots it wasn't very clear (to myself at least) whether you were trying to use them all together or not.

No, each kit has 4 modules so i am testing things with those seperately.

Here you can see three CPUZ pictures along with the benchmarks, two are in Dual Mode (this mainboard has two different ways to put modules in dual mode) and one Quad (used the 32GB kit but the same happens with the 64GB kit).

To my Surprise i swapped my 7900X for an 7980XE to see if it was the CPU fault and now Dual and Quad are at the same performance levels (again doesn't make sense however).

 

[Nik81]

Anyone?

 

[Nik81]

I can't help with why this is happening on your intel platform sadly but I can tell you that your memory performance is just pants! I ran some of your benchmarks on my machine with my "safe" esxi clocks so totally stock 2133mhz memory and this is what I got:

I know my numbers are messed up, the thing is i don't know if it's an BIOS setting i need to use.

20 years working with computers, first time i had such an issue.

 

[Nik81]

Yea I was trying to work out if it was just one benchmark that was all over the place but that maxxmem does seem to scale properly with system changes, I just changed my memory to 3200 cas 16 1t and gave it another go:



Hopefully somebody from gigabyte will look and tell us what is normal.

Not much faith in GB, posted the same question in their USA support forums 3 days ago, still no reply.

Makes me wish i had gotten an ASUS or EVGA mainboard

 

[Nik81]

@8 Pack Any chance you ever had the same issue? If this is the first you see it also then well then my luck sucks

 

[Nik81]

Well i even took the CPU out to check for bent pins, nada.



So it seems i am the only one who has ever encountered this problem, nice

 

[Nik81]

Yea I was trying to work out if it was just one benchmark that was all over the place but that maxxmem does seem to scale properly with system changes, I just changed my memory to 3200 cas 16 1t and gave it another go:



Hopefully somebody from gigabyte will look and tell us what is normal.

Any chance you can run your memory at 3200MHz but with the CPU at stock? I haven't overclocked my system to get a more "accurate" reading on the memory modules so my last hope is that these numbers are also tied to CPU OC.

 

[Nik81]

There isn't really a "stock" to speak of with TR as it just seems to sit wherever your cooling wants it to, the base clock is 3.4 with a boost of 4.1 mine seems to hover around 3.7 and boost to 4.15/4.25 (with load line and oc mode turned on) on very light loads, under a heavy load something like an avx it bottoms out at 3.4. Mine for that run was effectively stock with precision boost on. The problem is with my memory and cpu, to get the cpu to accept 3200 (rated speed on gen1 TR parts is 2133 with all dimms populated) I need to poke a few more mv to the soc and also set up load line on the memory, these two changes in turn make the cpu boost harder so with my chip at least I have to fiddle. Fwiw I was running all 8 sticks in those tests.

So it's basically like it has Turbo Boost like the Intel CPU. I have that enabled too but i thought you had manually OCed the CPU.

Well it's official, i am lost. No bent pins, no Bios settings i haven't tried. I also reset the bios twice, maximum memory performance on MaxMemm2 was at 14.4Gbyte/s which is 10 less than what you are getting so again something is wrong.

Defective mainboard perhaps?

 

[Nik81]

No, no manual OC just using precision boost and a 100mv bump to the SOC, memory load line at level 2 and away it goes. The CPU is basically stock. The best you could really expect anyway is a all core of 4ghz and that would have 0 effect on the memory and you would lose the extra 100/200mhz on the boost cores so yea Threadripper is not really worth overclocking, just get the memory at like 3000/3200 and let it do it's thing.

How much is another board to test? Must be somebody out there on an intel platform who could just run this. Cmon lads come out of the woodwork. How many ram slots do you have an how many are populated? I wonder what happens if I take half my memory out leaving me with 32gb quad channel but with only half my dimms populated.

Tried every configuration i could think of, nothing. Like i said i even checked for bent pins just in case nothing.

no one else i know has an X299 mainboard but for this to be defective it's very strange. It passed all Memtest86+ tests and a few burnin tests, so how can it be defective? I don't know really, even asked 8pack for input but nada

The X299 Aorus Gaming 9 has 8 RAM slots, i am using 4 (1 64GB and 1 32GB kits) in quad/dual mode with the same results as you have seen above..

 

[Nik81]

Are you sure you have the memory in the correct slots? Looking at the manual it needs to go:

16GB | 8GB | 16GB | 8GB -------- 8GB | 16GB | 8GB | 16GB

Actually i am not going for something as complicated, i am testing both kits seperately so its MODULE - EMPTY - MODULE - EMPTY - EMPTY - MODULE - EMPTY - MODULE.

I don't care whether i have 64GB or 32GB, my problem is the performance i am getting which doesn't make sense.

 

[Nik81]

Indeed, just wanted to make sure you had them in the right slots as that can often cause low memory speeds if you aren't using the primary slots first, or have your memory module sizes mixed. But it looks like you have them in the right place, so that's that ruled out.

The funny thing is that the system passes all load tests and the memories passed memtest86+ from a boot drive.

I was hoping someone else might have had the same issue but the problem is that unless you run some benchmarks you can't see it and i imagine many people have no need to run benchmarks (at least not RAM ones).

 

[Nik81]

Ok decided to reinstall Windows and now i am getting 3196 points in Passmark....

Still MaxMemm2 i still can't surpass 14.4Gbytes/s