Question Poor RAM performance even on XMP and overclock

[dnm_13]

I own Corsair Vengeance RGB RT 16Gb x 2, with XMP profile 3600MHz @ 16CL
Tested with memtest86 (on XMP), no error were found so it's all good
But on benchmark test (userbenchmark), it performed poorly and the site said mine had high latency (71.3ns)

Even on overclock setting, 4000MHz 1.480V @16CL (48'C), no error on memtest86, the site also said it performed poorly and the latency is still high (65.7ns)
The result were worse without XMP nor overclock, I'm not sure what I did wrong here, or it's just another silicon lottery and I lose the bet?

Also are userbenchmark a good site to compare benchmarks with?

 

[Roland Of Gilead]

Hey there,

Download AIda64 and run the mem/latency benches. It will give you a better idea of where your ram is at.

Whilst userbench is useful for a comparison, I wouldn't think twice about it's results and what it means for my hardware. It hardly taxes your components in anyway. It's only an indication of how your system compares to others with similar hardware. However, there are so many variables that can skew those results, the outcome of the test should be taken with a pinch of salt.

 

[rgd1101]

post the link to userbenchmark result

 

[dnm_13]

Hey there,

Download AIda64 and run the mem/latency benches. It will give you a better idea of where your ram is at.

Whilst userbench is useful for a comparison, I wouldn't think twice about it's results and what it means for my hardware. It hardly taxes your components in anyway. It's only an indication of how your system compares to others with similar hardware. However, there are so many variables that can skew those results, the outcome of the test should be taken with a pinch of salt.

Hi there!

Thanks for your reply
Here's the result from Aida64 memory latency benchmark

post the link to userbenchmark result

Sure thing
Here's the link to the benchmark result on userbenchmark

 

[Roland Of Gilead]

On your userbench, your SSD doesn't show results. I think this is normally down to running an overlay whilst testing. If you MSI afterburner or anything like that running, disable it for the userbench run, just to make sure it's not anomaly.

I've a 5600x/B550 combo, so this is just for reference. I'm running Vengeance 3600 CL18, I get 68ns in both the latency test and cache and mem bench (under tools).

Is your ram running 1:1 infinity fabric/memclock?

 

[dnm_13]

On your userbench, your SSD doesn't show results. I think this is normally down to running an overlay whilst testing. If you MSI afterburner or anything like that running, disable it for the userbench run, just to make sure it's not anomaly.

I've a 5600x/B550 combo, so this is just for reference. I'm running Vengeance 3600 CL18, I get 68ns in both the latency test and cache and mem bench (under tools).

Is your ram running 1:1 infinity fabric/memclock?

Before the benchmark (userbenchmark), I made sure to kill any unnecessary process, including MSI Afterburner, in order to get clean result and 0% CPU usage
The SSD (System Drive) didn't show any result because the benchmark was incomplete due to insufficient free space on it

Hmm, I'm not sure about the infinity fabric thing, I've heard about it but never mess with it, nor I do know how. How to check on it?

 

[Roland Of Gilead]

Before the benchmark (userbenchmark), I made sure to kill any unnecessary process, including MSI Afterburner, in order to get clean result and 0% CPU usage
The SSD (System Drive) didn't show any result because the benchmark was incomplete due to insufficient free space on it

Hmm, I'm not sure about the infinity fabric thing, I've heard about it but never mess with it, nor I do know how. How to check on it?

Do it through the bios. Set Your ram to XMP/DOCP and set the infinity fabric manually to run at 1800mhz. It may show in CPU-z as an odd ratio of 53:4 or something like that. You can use HWInfo to see if the clock is set at 1800 and in synchronous mode. It will get better performance, and possibly decrease latency.

 

[dnm_13]

Do it through the bios. Set Your ram to XMP/DOCP and set the infinity fabric manually to run at 1800mhz. It may show in CPU-z as an odd ratio of 53:4 or something like that. You can use HWInfo to see if the clock is set at 1800 and in synchronous mode. It will get better performance, and possibly decrease latency.

Looked it up on google, it's FCLK setting on MSI bios, set it to 1800MHz (since my RAM speed is 3600MHz on XMP)
Both userbenchmark and Aida64 result were roughly the same, so I don't think that's the issue I'm facing here

Edit: IDK why on userbenchmark it said the RAM was clocked to 4000MHz, I checked on BIOS before restarting and I didn't set it on overclock setting, only XMP @3600MHz
Edit2: CPU-Z agrees with me
Edit3: WRONG IMAGE, updated with real image, looks like I set the wrong XMP profile (3200MHz), gotta restart and retest now

 

[dnm_13]

There's a noticable progress after I correctly set the XMP profile to 3600MHz with infinity fabric set to 1800MHz
It's much more better result than when I did it with overclock (4000MHz) setting
There's progress, but it's still a bad result (?)

What now...? looks like the problem doesn't end with infinity fabric? Although this is a step forward for sure

 

[rgd1101]

do you actually have issue or just userbenchmark?
because they hate AMD system

 

[dnm_13]

do you actually have issue or just userbenchmark?
because they hate AMD system

LOL looks like I'm not the only one that noticed userbenchmark reviews were quite biased against AMD

I'm not sure what's the standard, nor what's good score for latency
And the score is quite the same (margin error included) with AIDA64, so I don't think it's limited to userbenchmark

 

[Zerk2012]

LOL looks like I'm not the only one that noticed userbenchmark reviews were quite biased against AMD

I'm not sure what's the standard, nor what's good score for latency
And the score is quite the same (margin error included) with AIDA64, so I don't think it's limited to userbenchmark

Are you sure the memory is in the correct slots for dual channel.

 

[Unolocogringo]

Userbenchmarks rating system is crap.
If you look at your last memory results you were 96.1% of the tests run with the same memory.
If you have not overclocked or tweeked memory settings ,that is a good score.
It is a useful benchmark in the system information it gives but their ratings and comparisons are way off.

 

[dnm_13]

Are you sure the memory is in the correct slots for dual channel.

yes, it's in A2 B2
if it's in A1 or B1 my BIOS would prompt me

Userbenchmarks rating system is crap.
If you look at your last memory results you were 96.1% of the tests run with the same memory.
If you have not overclocked or tweeked memory settings ,that is a good score.
It is a useful benchmark in the system information it gives but their ratings and comparisons are way off.

while I do want to agree with you, AIDA64 gave out somewhat similar result with userbenchmark

You have any recommendation for benchmark?

 

[Roland Of Gilead]

yes, it's in A2 B2
if it's in A1 or B1 my BIOS would prompt me


while I do want to agree with you, AIDA64 gave out somewhat similar result with userbenchmark

You have any recommendation for benchmark?

Hey,

To me your ram is running as it should, and the latency is about where it should be.

Your latency is right for CL16 v CL18 (my own) which has slightly higher latency.

IMO you've nothing to worry about.

There's a noticable progress after I correctly set the XMP profile to 3600MHz with infinity fabric set to 1800MHz
It's much more better result than when I did it with overclock (4000MHz) setting
There's progress, but it's still a bad result (?)

What now...? looks like the problem doesn't end with infinity fabric? Although this is a step forward for sure

If you're thinking the latency highlighted above (for your CPU), should be better than the systems above you in the chart, that's the wrong way to think about it.

IMO your ram is running as it should be. Best keep at 3600/1800 mclk/fclk. That's the lowest latency you can have, unless you can run the DIMMS at 4000 with a 2000mhz fclk. Your IMC may not be up to it. You will only know if you try.

 

[Bob.B]

yes, it's in A2 B2
if it's in A1 or B1 my BIOS would prompt me


while I do want to agree with you, AIDA64 gave out somewhat similar result with userbenchmark

You have any recommendation for benchmark?

Just for yuks.
Download UBM.
Reboot and wait a few mins.
Run UBM with the browser closed.
Post a link to the results.

 

[bfollett]

Enabling Xmp only sets a few primary memory timings in the motherboard's bios. The rest is up to the bios to decide and those setting are usually pretty conservative. I have 3600 C16 ram and with a combo of using Dram Calculator and this guide: MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper · GitHub
I tightened the rest of the timings and I'm running way above the average for my ram. If you don't want to go through setting everything yourself, Download/run ZenTimings and send us a screenshot and we can give suggestions of a few timings that might quickly improve your benchmarks. My Userbenchmark results below.
Asrock B450M Pro4 Performance Results - UserBenchmark

 

[dnm_13]

Hey,

To me your ram is running as it should, and the latency is about where it should be.

Your latency is right for CL16 v CL18 (my own) which has slightly higher latency.

IMO you've nothing to worry about.


If you're thinking the latency highlighted above (for your CPU), should be better than the systems above you in the chart, that's the wrong way to think about it.

IMO your ram is running as it should be. Best keep at 3600/1800 mclk/fclk. That's the lowest latency you can have, unless you can run the DIMMS at 4000 with a 2000mhz fclk. Your IMC may not be up to it. You will only know if you try.

do you think so?
what bugs me out is that userbenchmark said mine was at 23rd percentile, meaning there's about 77% other user with same RAM that performs better than mine
I know that most of the time RAM performance has insignificant impact on gaming or daily task, but it just bugs me out you know

Just for yuks.
Download UBM.
Reboot and wait a few mins.
Run UBM with the browser closed.
Post a link to the results.

I did exactly this on every test

Enabling Xmp only sets a few primary memory timings in the motherboard's bios. The rest is up to the bios to decide and those setting are usually pretty conservative. I have 3600 C16 ram and with a combo of using Dram Calculator and this guide: MemTestHelper/DDR4 OC Guide.md at oc-guide · integralfx/MemTestHelper · GitHub
I tightened the rest of the timings and I'm running way above the average for my ram. If you don't want to go through setting everything yourself, Download/run ZenTimings and send us a screenshot and we can give suggestions of a few timings that might quickly improve your benchmarks. My Userbenchmark results below.
Asrock B450M Pro4 Performance Results - UserBenchmark

Tried overclocking these sticks, the only thing I was able to move was the clock, CAS latency can't be less than 16 (even on 15 it can't boot immediately)
I'll take a look at the guide you gave and search ZenTimings though, never heard any of it

 

[Roland Of Gilead]

do you think so?
what bugs me out is that userbenchmark said mine was at 23rd percentile, meaning there's about 77% other user with same RAM that performs better than mine
I know that most of the time RAM performance has insignificant impact on gaming or daily task, but it just bugs me out you know

Did you run the cache and mem bench on Aida? It's in the tools tab. It will show you the bandwidth and latency. You can compare it directly to the bandwidth of other DIMMs. You can also see the benefits or not of changes in primary and secondary sub timings.

Whilst increasing performance is a good thing, chasing synthetic benchmark scores is can be futile.

 

[SkyNetRising]

what bugs me out is that userbenchmark said mine was at 23rd percentile, meaning there's about 77% other user with same RAM that performs better than mine
I know that most of the time RAM performance has insignificant impact on gaming or daily task, but it just bugs me out you know

Can you show screenshots from CPU-Z - memory and spd sections?

 

[dnm_13]

Did you run the cache and mem bench on Aida? It's in the tools tab. It will show you the bandwidth and latency. You can compare it directly to the bandwidth of other DIMMs. You can also see the benefits or not of changes in primary and secondary sub timings.

Whilst increasing performance is a good thing, chasing synthetic benchmark scores is can be futile.

Here

Can you show screenshots from CPU-Z - memory and spd sections?

Sure thing
Here

 

[Roland Of Gilead]

As a matter of interest, what are the rest of your PC specs? Also, I wonder why Aida isn't showing the CPU information correctly? That's odd. Does that happen in CPU-z main page too?

 

[dnm_13]

As a matter of interest, what are the rest of your PC specs? Also, I wonder why Aida isn't showing the CPU information correctly? That's odd. Does that happen in CPU main page too?

You also can look at my PC spec from userbenchmark link that I gave above
but here're they anyway:
MSI X570-A PRO
Ryzen 7 5800X
Nvidia RTX 3080 Ti
PSU Corsair TX750M 750W
Samsung SSD NVMe M2 970EVOPlus 250GB
WD Green SSD M2 240GB
Seagate Barracuda 1TB

My AIDA64 is probably an older version
Unless it also affect their benchmark / tool or stress test, I think it's fine?

 

[SkyNetRising]

Sure thing
Here

tRC is 85 (it should be 58 instead).

Did you set those values manually?

 

[dnm_13]

tRC is 85 (it should be 58 instead).

Did you set those values manually?

No, I don't know how to, not even know what does tRC for / means...