[SOLVED] RAM Performance Issues

[bahamut204]

I built this computer about a year ago an it's performance hasn't been the greatest. One thing that I've noticed is that doing different benchmarks the RAM always seems to be under preforming. UserBenchMark for one shows the computer under performing. Ran the bench mark a few times and the numbers keep fluctuating. Been thinking of replacing the RAM with another kit. But would prefer to not spend the money if there's an underlying problem with the system that can be fixed instead.

Here are the latest results: https://www.userbenchmark.com/UserRun/52034726

System specs are there. But it's a Ryzen 5 5600X on a ASUS ROG Crosshair VIII Hero (Wi-Fi). RTX 3070. 128GB DDR4 3200 cl16 ram. (16-20-20-40) (tRC 74, tRFC 880)
Ram is: RAM Patriot Viper Steel Series. Models PVS464G320C6K and PVS432G320C6 (Same specs for speed and timing.)
I had to go through 8 sticks of this RAM to get 4 functional. 4 were DOA. It took reseating the sticks many times to not get errors running memtest.

AIDA64 lists the latency at 93 ns for my config. The comparison table shows 3 other DDR4 3200 kits. A Ryzen 9 3950X @ 76 ns, Ryzen 9 5900X @ 65.8 ns, and a I9 11900K @ 59.2 ns. Am I correct in concluding that 93 ns is rather poor performance in this case?

I was looking at 2 kits as possible replacements. Both G. Skill Ripjaws V Series. Either:
64GB (4x16GB) DDR4 3600 14-14-14-34 (About $735 sale price.)
128GB (4x32GB) DDR4 3600 16-22-22-42 (About $890 sale price.)
Would either of these provide a noticeable improvement in performance? Is there some setting or something that can be changed to improve performance of what's in there. I'd prefer to keep 128GB. But if the difference is noticeable between those 2 kits and the one that's in there I might drop to 64GB. I'm worried there's just a problem with some settings somewhere and replacing the RAM will do nothing as it might not me the issue.

On a side note. Was it unrealistic to expect a noticeable improvement in gaming performance going from a I7 4770K, GTX 1060 6GB, 32GB DDR3 1600 8-8-8-24 setup to a Ryzen 5 5600X, RTX 3070 8GB, 128GB DDR4 3200 16-20-20-40 setup? (I use the extra RAM for other things, not expecting the extra to improve performance. But was expecting the speed to.)

Any and all help would be appreciated. I've asked around a few other places and haven't gotten any help on the matter. If you need more info or I forgot anything. Just let me know.

 

[kerberos_20]

you userbenchmark score is invalid, 20% background CPU usage
you should run nothing while benching

runt this app and post screenshot from it, it will tell how you ram is setup in bios as 80ns is quite high for 3200

 

[bahamut204]

you userbenchmark score is invalid, 20% background CPU usage
you should run nothing while benching

runt this app and post screenshot from it, it will tell how you ram is setup in bios as 80ns is quite high for 3200

Ok, I ran the benchmark again after a reboot and waiting for Windows to finish loading:
https://www.userbenchmark.com/UserRun/52038780
Didn't change the latency much. Although other numbers improved.

Here's the screenshot you asked for:

https://imgur.com/a/UOVFiNe

View: https://imgur.com/a/UOVFiNe

 

[kerberos_20]

cpu settings are correct
ram timings are mostly auto/xmp configured
tRFC at 880 is quite high, its usualy 6-8x of your tRC number which would be 444-592, try with 592 then keep reducing until it gets unstable/wont post
you could probably drop tRC aswell, that should reduce latency further
tRC minimum is tRP+tRAS rounded to next even number, for stability add +2, so 62 for tRC
and with lower tRC you can drop tRFC even lower now (372~496)

 

[bahamut204]

cpu settings are correct
ram timings are mostly auto/xmp configured
tRFC at 880 is quite high, its usualy 6-8x of your tRC number which would be 444-592, try with 592 then keep reducing until it gets unstable/wont post
you could probably drop tRC aswell, that should reduce latency further
tRC minimum is tRP+tRAS rounded to next even number, for stability add +2, so 62 for tRC
and with lower tRC you can drop tRFC even lower now (372~496)

I'll give it a try a bit later tonight when I get some spare time. In the meantime I added 2 pics to that link from my last message. The info from Taiphoon Burner for both of the different sticks of RAM in there. I hear a bunch about the die that the chips are but don't know what those are or their quality.

I'll see what I can get those numbers down to. The XMP profile says the tRC should be 64 but the system set it to 74. I know the system boot looped many times with this RAM and it was difficult to get the XMP profile to stick. I'll report once I see if either of those numbers can be dropped without the system having issues.

Is it odd that the manufacturer and the XMP profile list 16-20-20-40, but everywhere that sells these sticks lists them at 16-18-18-36? That's what I thought I was buying. Was surprised to find the numbers were a bit higher.

 

[bahamut204]

I got the tRC to 62 and the tRFC to 520. At a tRFC of 518 the system won't POST and I have to clear the CMOS to get it to boot again. At that time it boot loops 3 times and POSTs in safe mode and I have to manually enable the XMP profile again. That happens a bit sometimes it won't boot up with the XMP profile and that whole boot loop/safe mode thing happens.

I tried changing each part of the 16-20-20-40 individually. Dropping any one of them at all means resetting the CMOS again and back into the boot loops.

Here is the latest benchmarks with the new timings: https://www.userbenchmark.com/UserRun/52045762

Thoughts?

 

[egda23]

I got the tRC to 62 and the tRFC to 520. At a tRFC of 518 the system won't POST and I have to clear the CMOS to get it to boot again. At that time it boot loops 3 times and POSTs in safe mode and I have to manually enable the XMP profile again. That happens a bit sometimes it won't boot up with the XMP profile and that whole boot loop/safe mode thing happens.

I tried changing each part of the 16-20-20-40 individually. Dropping any one of them at all means resetting the CMOS again and back into the boot loops.

Here is the latest benchmarks with the new timings: https://www.userbenchmark.com/UserRun/52045762

Thoughts?

From the latest bench:
Globally your PC is performing satisfactorily (57 th percentile)
The problem you are grappling with your RAM is difficult to judge from this bench as they have a sample of just 12 PCs with the same RAM as you have. From these 12, you are in fact in the middle of the pack (42nd percentile). Doing statistics on a sample of 12, is not as reliable as on the hundreds of thousands you can find for your CPU for example.
So, you can only conclude that you have similar performance than the 11 other PCs with the same RAM !

 

[kerberos_20]

ignore that percentile thing, that percentage doesnt represent same settings across all tested systems, basicly around 50% is ok, that test also shows you that its all ok there, ryzen 3000/5000 also have half write speed comperaed to zen1 or intel, so if you use your ram on intel or zen1, mem score would be few percent higher, without overclocking you wont get it better, say overclock to 3600MHz on ram and percentage would skyrocket aswell, bench score is around 100% which is fine for 3200MHz (at 3600MHz you would hit something like 110%) thats all there is to it) latency for your ram is ok now, which also translates in better system latencies...which means better system response...that feeling when doing something in windows, it will be less laggy to the eye....see your drives are tiny bit faster too(as they use ram cache), cpu score is same even if you have more bench percentage now...its just memory score that got better which is pointless in CPU tests (unless you do some database operations)

the only thing which remains and it kinda hits the eye is your M2 nvme runs in gen2 mode, while mainboard supports gen4...check bios settings for this
other than that it looks fine
unless you experiencing some issues?

 

[bahamut204]

ignore that percentile thing, that percentage doesnt represent same settings across all tested systems, basicly around 50% is ok, that test also shows you that its all ok there, ryzen 3000/5000 also have half write speed comperaed to zen1 or intel, so if you use your ram on intel or zen1, mem score would be few percent higher, without overclocking you wont get it better, say overclock to 3600MHz on ram and percentage would skyrocket aswell, bench score is around 100% which is fine for 3200MHz (at 3600MHz you would hit something like 110%) thats all there is to it) latency for your ram is ok now, which also translates in better system latencies...which means better system response...that feeling when doing something in windows, it will be less laggy to the eye....see your drives are tiny bit faster too(as they use ram cache), cpu score is same even if you have more bench percentage now...its just memory score that got better which is pointless in CPU tests (unless you do some database operations)

the only thing which remains and it kinda hits the eye is your M2 nvme runs in gen2 mode, while mainboard supports gen4...check bios settings for this
other than that it looks fine
unless you experiencing some issues?

I gave increasing the speed a try this morning. I can't even bump it up to the next higher one in the list and still have the computer boot. I'm going to have to try to rework the tRC and tRFC again. I left the computer on overnight to stress test the settings I mentioned earlier. It shut off about 45mins after I left it. Took a while to get the computer to boot again.

I checked that drive. Windows says it's in gen 4. I also went into the BIOS and manually set it to gen 4. It didn't change anything. I also removed the gen 3 nvme that was installed just to see and still no change. I ran CrystalDiskMark on both the gen 3 and 4 nvme drives. The read speeds where about what I expected from the specs on the box. But the write speeds were still weird. For the gen 4 the seq1m q8t1 and seq1m q1t1 write results averaged out to be 1150, 1150 respectively. With one time reporting 1700, 4200. The gen 3 drive stayed consistent at 3000, 2750 over the runs. Running CrystalDiskInfo shows 1219 media and data integrity errors. SMART status shows Good 97%. The other 5 drives in the system show 0 errors for that time. Bad drive? It's the OS drive and every once and a while I have to reinstall a program that starts acting up randomly.

 

[Bob.B]

I gave increasing the speed a try this morning. I can't even bump it up to the next higher one in the list and still have the computer boot. I'm going to have to try to rework the tRC and tRFC again. I left the computer on overnight to stress test the settings I mentioned earlier. It shut off about 45mins after I left it. Took a while to get the computer to boot again.

I checked that drive. Windows says it's in gen 4. I also went into the BIOS and manually set it to gen 4. It didn't change anything. I also removed the gen 3 nvme that was installed just to see and still no change. I ran CrystalDiskMark on both the gen 3 and 4 nvme drives. The read speeds where about what I expected from the specs on the box. But the write speeds were still weird. For the gen 4 the seq1m q8t1 and seq1m q1t1 write results averaged out to be 1150, 1150 respectively. With one time reporting 1700, 4200. The gen 3 drive stayed consistent at 3000, 2750 over the runs. Running CrystalDiskInfo shows 1219 media and data integrity errors. SMART status shows Good 97%. The other 5 drives in the system show 0 errors for that time. Bad drive? It's the OS drive and every once and a while I have to reinstall a program that starts acting up randomly.

Just for yuks.
Try UBM with 2 sticks of ram in the proper slots.

Post a screenshot for CDI for this disk.

 

[bfollett]

Your Ram is operating right about where 3200 C16 ram should be. You really can't go by Userbenchmark in your case because there have only been 11 samples of that ram benchmarked and several were heavily overclocked, throwing off the whole percentile scale. There is a search at the top of main UserBenchmark page and you could use that to see what other 3200 ram is achieving. Here is a different set of 3200 C16 Patriot ram that has over 9000 user runs and you'll see you ram is doing fine:

UserBenchmark: Patriot 3200 C16 Series 2x8GB

 

[kerberos_20]

Your Ram is operating right about where 3200 C16 ram should be. You really can't go by Userbenchmark in your case because there have only been 11 samples of that ram benchmarked and several were heavily overclocked, throwing off the whole percentile scale. There is a search at the top of main UserBenchmark page and you could use that to see what other 3200 ram is achieving. Here is a different set of 3200 C16 Patriot ram that has over 9000 user runs and you'll see you ram is doing fine:

UserBenchmark: Patriot 3200 C16 Series 2x8GB

those 11 other benchmarks werent overclocked...tbh 2666/3200 is what they run, but they run it with different cpu like 5950x reaching 70-100GB on mem reads which makes that ram performance much higher than with 5600x

tbh i have no idea why would 5950x reach such high numbers considering its just dual channel, but looking at cpu itself on userbench, they all doing that there when populated with 4 ram modules, mc read is doubled...aida64 google pictures looks normal...so some userbench bug with this cpu

 

[bahamut204]

Just for yuks.
Try UBM with 2 sticks of ram in the proper slots.

Post a screenshot for CDI for this disk.

Here is the CDI screenshot for the drive in question:

https://imgur.com/a/CBLy2N4

View: https://imgur.com/a/CBLy2N4

 

[Bob.B]

Here is the CDI screenshot for the drive in question:

https://imgur.com/a/CBLy2N4

View: https://imgur.com/a/CBLy2N4

CDI does not show anything flat busted.
I do have to wonder about those media errors and you having to reload stuff.

Did you test UBM with 2 sticks of ram?

 

[bahamut204]

CDI does not show anything flat busted.
I do have to wonder about those media errors and you having to reload stuff.

Did you test UBM with 2 sticks of ram?

I went back and checked the event viewer for fun. It's only a couple but there's a few hardware faults flagged there for that drive. I found another PICe 4.0 4x drive that was on sale. Going to try a fresh install on that to see if the problems persist. I had tried a clean install on the drive several months ago. But it didn't change anything.

No, I didn't try with just 2 sticks. Given how many times I had to re seat them to get them stable. I'm rather concerned about removing them unless it's to replace them. Also my cooler overhangs the ram and I'd have to keep removing the fan to get access to the ram slots. Which is annoying with that cooler. I am rather curious if that'd do anything though. Or if it'd make a difference with which 2 sticks I used. Thaiphoon Burner lists the chips on 2 of the modules as SpecTek MT40A2G8??-075:[B/E] and the other 2 as Samsung K4AAG085W?-BCTD. I hear about different dies and different qualities. But I have no idea what die each of these is or the quality of said chips. Just throwing the part number into google wasn't helpful. Don't really know what the rest of the information it lists is either.

 

[bahamut204]

those 11 other benchmarks werent overclocked...tbh 2666/3200 is what they run, but they run it with different cpu like 5950x reaching 70-100GB on mem reads which makes that ram performance much higher than with 5600x

tbh i have no idea why would 5950x reach such high numbers considering its just dual channel, but looking at cpu itself on userbench, they all doing that there when populated with 4 ram modules, mc read is doubled...aida64 google pictures looks normal...so some userbench bug with this cpu

I thought that was weird too. So for fun I managed to bride a buddy of mine to let me use his new 5800X3D for a few days for testing before he dropped it into his system. I updated the bois from version 3801 to 4006 to be able to run the new cpu. Here's what happened:

5600X on version 4006 bios: https://www.userbenchmark.com/UserRun/52081221
5800X3D on version 4006 bios: https://www.userbenchmark.com/UserRun/52085876

Nothing else on the system changed between tests. Interesting to see how much the ram score improved between the 2 benchmarks. Not to those 5950X levels but there was a change in the numbers.

That new cpu is interesting but toasty compared to the 5600X on my system. I ran the Real Bench 8h stress test on both and these are what I got for thermals. Ambient temp in the room was between 27-30c at the time (not my choice on that).
5600X: Idle between 32-34c, max on stress test 63-70c, cpu ccd2(tdie) was the hottest thing at 78c.
5800X3d: Idle between 33-35c, max on stress test 88-90c, cpu ccd1(tdie) was the hottest thing at 95c.
If it was mine I'd be playing with the cooling to see how to get the temps down. Not sure how though. Don't like them that high. It's interesting but runs hotter than I like. Wouldn't want to put the money out on it now. But maybe after some time when the price drops it'd be more appealing.

AIDA64 is interesting too. The latency for the L1-3 caches is the same, the ram latency and data rate is the same. But the data rate on the L1-3 caches on the 5800X3D are a fair bit faster that what they are on the 5600X. Any Idea why? Here's a screen shot of it:

https://imgur.com/a/47ZmWLM

View: https://imgur.com/a/47ZmWLM