Problems to run memory at EXPO 6800MT/s

[Undeon]

Hey everyone,

I'm kind of a noob when it comes to overclocking anything newer than just increasing the FSB, but I'm trying to learn how to properly OC my system.

I've built an ITX system using a Ryzen 7 8700G, Jginyue B650i Night Devil motherboard, and Juhor 2x16GB DDR5 6800MT/s 1.4V (Hynix-A chips) memory.

The CPU side of the APU is running without OC, but I’ve enabled Curve Optimizer Per Core:Core 1: -22 | Core 2: -46 | Core 3: -37 | Core 4: -39Core 5: -44 | Core 6: -37 | Core 7: -36 | Core 8: -42

The GPU side is completely stock, but I can OC it to 3150MHz @1.25V.

The motherboard is running AGESA 1.2.0.0a. (There's a new BIOS update, but it doesn't mention an AGESA update.)

The memory is Juhor JHE6800U3416JGRGB, rated for 6800MT/s 34-45-45-108. This is listed as achievable with both XMP and EXPO.

In the BIOS, the best stable setting I managed to get was 6400MT/s 34-37-37-36-48 with UCLK DIV1 MODE at AUTO. When researching this issue, I found two people who managed to get the same 6400MT/s with MEMCLK 1:1, and both mentioned it's possible to achieve higher clocks with MEMCLK 1:2. However, when I set the clock to 6800 via EXPO and MEMCLK 1:2 manually, the system boots, but Prime95 crashes right at the start, and then I get a BSOD.

I have two kits of this memory, and both give me the same results.

Is there anything I can do to first get 6800MT/s stable, and then possibly overclock beyond that?

Thanks!

 

[]-[itman]

I would stick with 6400MT/s and 1:1 clocks. Anything higher usually requires some manual tuning and you will get worse performance anyway from having to go 1:2.

 

[Undeon]

I would stick with 6400MT/s and 1:1 clocks. Anything higher usually requires some manual tuning and you will get worse performance anyway from having to go 1:2.

I saw a big increase in GPU performance with higher clocks. That's why I want to OC it.

 

[]-[itman]

I saw a big increase in GPU performance with higher clocks. That's why I want to OC it.

When you cross from 1:1 clock to 1:2 clocks, your performance will go down, unless you can get significantly higher memory clocks (like 7600MT/s or higher) with 1:2 clocks.

 

[Woomack]

Until now, I didn't know that a brand like Jginyue existed, and I guess I saw Juhor RAM on Aliexpress or something. Either way, it's hard to say how it will work, and I doubt many forum members (if any) have experience with a combo like that.
Either way, the best performance for CPU/RAM-related calculations, other than pure graphics (if I can say that), is a 1:1 ratio. The 8700G APU should run at 6400-6600. Somehow, I doubt it will run stable at 6800 1:1, and it also highly depends on the motherboard. I would try how high it will go at 1:2. Typically, these APUs can make 8600+, and the RAM can easily pass 8000, so the main question is how high the motherboard can run. As I said, I doubt anyone around will tell you that, as this brand does not exist outside of China. The manufacturer suggests 7000+. As was already mentioned, it would be best if you could set 7600+.
For 7600-8200, use CL34-48-48 or CL36-49-49 1.45V. Other timings can be set to auto (at least for the initial test), but I'm unsure how this motherboard will set them. If it doesn't even post at some clock like 7400, 7600, or 7800, ... then sub-timings won't help.

 

[Undeon]

Until now, I didn't know that a brand like Jginyue existed, and I guess I saw Juhor RAM on Aliexpress or something. Either way, it's hard to say how it will work, and I doubt many forum members (if any) have experience with a combo like that.
Either way, the best performance for CPU/RAM-related calculations, other than pure graphics (if I can say that), is a 1:1 ratio. The 8700G APU should run at 6400-6600. Somehow, I doubt it will run stable at 6800 1:1, and it also highly depends on the motherboard. I would try how high it will go at 1:2. Typically, these APUs can make 8600+, and the RAM can easily pass 8000, so the main question is how high the motherboard can run. As I said, I doubt anyone around will tell you that, as this brand does not exist outside of China. The manufacturer suggests 7000+. As was already mentioned, it would be best if you could set 7600+.
For 7600-8200, use CL34-48-48 or CL36-49-49 1.45V. Other timings can be set to auto (at least for the initial test), but I'm unsure how this motherboard will set them. If it doesn't even post at some clock like 7400, 7600, or 7800, ... then sub-timings won't help.

Wow, thanks man! This is the first time I've been able to run this memory at 6800, with CL36-49-49 at 1.45V. After I finish the tests (I finished TestMem5 and I'm running Prime95 right now, and it didn’t crash at the beginning), I'll try pushing for higher clocks to see what this motherboard can handle.


Too good to be true. This motherboard can't handle this memory at 6800, even with relaxed timings and overvolting. I'll stick to 6400 and tighten the timings as much as possible.

 

[Undeon]

New topic, but related.

When I tighten the timings one step before the tests result in errors, my mouse hangs a lot during the tests, but the system doesn't crash. Is this acceptable, or should I relax the timings until the mouse works fine?

 

[Woomack]

I'm not sure if this is a serious question
Stable is stable, with no weird behavior, freezing, or anything.
For 6400 try CL30-38-38, 32-38-38 or 32-39-39 1.40-1.45V.
Check how high tREFI (tREF or something similar, depending on the manufacturer) will go. Typically it's 32-64k.

Use any popular memtest to check stability. It can be bootable memtest, but it takes a lot of time to finish. It can be AIDA64 stability test, but its test pattern not always catches all errors. There are also options like Prime95 blend (whatever you use, but it has to cover 90%+ RAM). In general, 2h is more than enough for a stability test. Later, it's good to run mixed load tests, some games, or whatever, to check if different calculations don't cause problems.

 

[mackerel]

Use any popular memtest to check stability. It can be bootable memtest, but it takes a lot of time to finish. It can be AIDA64 stability test, but its test pattern not always catches all errors. There are also options like Prime95 blend (whatever you use, but it has to cover 90%+ RAM). In general, 2h is more than enough for a stability test. Later, it's good to run mixed load tests, some games, or whatever, to check if different calculations don't cause problems.

Something I do is to run Aida64 mem only test for like 2 minutes, then Prime95 blend for 2 minutes. This is just to check if there is something big enough to cause instability quickly. If both make it past 2 minutes, then you can then do the longer test. I've had times where some tests would run as long as you like, but another would error after seconds. By doing a little of each up front you can check for that and not waste time on a long test that might not reveal the problem another could.

 

[Woomack]

Something I do is to run Aida64 mem only test for like 2 minutes, then Prime95 blend for 2 minutes. This is just to check if there is something big enough to cause instability quickly. If both make it past 2 minutes, then you can then do the longer test. I've had times where some tests would run as long as you like, but another would error after seconds. By doing a little of each up front you can check for that and not waste time on a long test that might not reveal the problem another could.

AIDA64 usually shows errors in the first ~5-15 minutes. When it's at the edge of stability, then errors show after filling the whole RAM (well, it uses ~95%), so as you said, 1-2 minutes. Next, after about 2h. This is how I see it in my tests.
Some years ago, I decided to use one soft for reviews and stick with it, so I typically run ~3h AIDA64 on every new memory kit.