1700x help (cache failure, high latency, etc)

[custom90gt]

Sadly I haven't had an AMD system to play with in a long long time but I picked up a cheap 1700x to play with (really just to get accustomed to AMD since I am hopeful for Zen 2) and am having the struggles.

Here is the system:
AMD 1700x
MSI x470 Gaming Pro Carbon w/ bios 7B78v24
4x8gb HyperX Predator DDR4 4000MHz @3200-14-14-14-34 @ 1.4v
EVGA Ultra Gaming 2080ti
960GB ADATA SX8200
Corsair H115i Pro

I thought I was stable at stock clocks with the ram overclock but I've had 2 random BSODs when encoding videos. I fired up AIDA64 and started a stability test and it turns out the cache test was failing within minutes. The bios was turning SOC voltage up to 1.185v so I figured it may be too high. Tried 1.0v and it ran for 20 minutes vs 6 before. Tried 1.05v and it did 24 minutes. Tried 1.1v and it ran for 56min before erroring out. Now I'm trying 1.125v to see what happens. What are people running their SOC at? I realize it may be due to running 4 sticks of ram so I may have to lower the memory speeds.

Memory test in AIDA64 is 8 hours error free though. Any suggestions?

My other "issue" is that my memory latency is really high. Again this may be due to 4 sticks of ram, but I am curious if anyone has seen this before:


I was hoping to overclock this CPU to 3.8-3.9GHz but not getting it to run at stock clocks with memory OC is a little bit disappointing.

 

[custom90gt]

Dang just failed at 1.125v in 18 minutes...

 

[Mr.Scott]

Memory OC is what's killing you.
You're way tight for the speed you're running on 4 sticks.

 

[custom90gt]

It's just odd that memory passes just fine but cache does not. I realize 4 sticks is more difficult than 2 but my memory stress test failures were cleared up by going from 1.35v to 1.4v.

 

[Johan45]

SOC isn't linked to the cache that would be CPU voltage. SOC takes time to dial in properly but 1.05 is a good starting point. Mr Scott is right though 4 sticks are harder than 2 you can try 14-14-15 and maybe loosen the tRFC a bit

 

[custom90gt]

SOC isn't linked to the cache that would be CPU voltage. SOC takes time to dial in properly but 1.05 is a good starting point. Mr Scott is right though 4 sticks are harder than 2 you can try 14-14-15 and maybe loosen the tRFC a bit

So cpu voltage controls everything? Odd that messing with SOC would have such a change in cache stability. Right now 1.0875 has been testing for 59 minutes and counting. If I raise it higher than 1.15 I can error out in 5 minutes. I will try loosing timings if I can't get it stable by playing with voltage. Can anyone explain why memory tests fine though?

 

[Johan45]

Not really, I don't use AIDA64 for stress testing a lot. I also think you have the memory voltage a bit high which can affect stability on this platform since the cache rail is automatically set to 1/2 of RAM voltage try 1.36 or 1.37. The RAM itself should be fine at that voltage

 

[custom90gt]

Not really, I don't use AIDA64 for stress testing a lot. I also think you have the memory voltage a bit high which can affect stability on this platform since the cache rail is automatically set to 1/2 of RAM voltage try 1.36 or 1.37. The RAM itself should be fine at that voltage

AIDA64 would error out after a couple of hours when testing memory at 1.35. I tried lowering the memory to 3000-14-14-14-34 and 1.35v to see if that changes anything. Just for fun I put the ram at relaxed timings and 3333mhz and the cache would instantly error. I think this cpu just isn't a real phenomenal one...

 

[Mandrake4565]

Something that comes to mind in my experience. What version of windows are you running? When I first put my 1800x in my main rig it wasn't playing nice with an original version of W10 I had installed. At first I believed it was a memory issue due to the codes and such I was getting but it turned out to be the version of W10. Once I updated I've been problem free and am running a pair of B-die Samsungs at 3466 Cl 15 as we speak. Though I am running 2 sticks Vs your 4.

 

[custom90gt]

Something that comes to mind in my experience. What version of windows are you running? When I first put my 1800x in my main rig it wasn't playing nice with an original version of W10 I had installed. At first I believed it was a memory issue due to the codes and such I was getting but it turned out to be the version of W10. Once I updated I've been problem free and am running a pair of B-die Samsungs at 3466 Cl 15 as we speak. Though I am running 2 sticks Vs your 4.

I just did a fresh install of Win 10 Pro 1809 about a week ago on it. Running the latest AMD chipset drivers.

1.0825v on SOC failed after 1hr and 45min on the cache. I have been running memory and cache testing at 3000mhz ram and 1.0825v SOC for the past 1.5 hours. I will see what happens from here and keep the thread updated. Sad that 3333mhz on the ram even at 18-18-18-38 caused the cache to error out after less than one minute regardless of SOC voltage.

 

[Johan45]

Have you used the MSI try it setting? I found they worked well but I wasn't running 4 sticks of RAM. This platform is quite sensitive to voltages for stability so it can take a lot of tinkering. One other thing that high speed RAM might be setting some weird timings when left to auto. How many are set manually?

 

[custom90gt]

Have you used the MSI try it setting? I found they worked well but I wasn't running 4 sticks of RAM. This platform is quite sensitive to voltages for stability so it can take a lot of tinkering. One other thing that high speed RAM might be setting some weird timings when left to auto. How many are set manually?

I did use the MSI try it to get to 3200mhz and then again to do the loose timings at 3333mhz. Right now I'm on "3200mhz try it" setting but with the ram set to 3000mhz and it's been testing ram/cache for the past 4 hours without issue. I really think it's the cpu and not so much the ram. My guess is that I lost the silicon lotto there.

 

[Mr.Scott]

My guess is a board/bios issue.
It's a mid range gaming board. You might be asking too much of it.

 

[custom90gt]

AM4 boards with 8 SATA ports are hard to come by. Sad that a $200 board is just a mid ranged board now a days (and that 3200mhz is asking alot, lol).

Let's hope Zen 2 kicks it up a couple of notches (considering I've been used to Intel with 4000+mhz in quad channel). If my Microcenter gets any Asrock Taichi boards in for a decent price, maybe I'll try one of those and just return this one.

 

[ShrimpBrime]

4x8gb HyperX Predator DDR4 4000MHz @3200-14-14-14-34 @ 1.4v

Your issue is right here. Gonna have to loosen them up a bit and probably be able to back down the voltage too.

 

[custom90gt]

Your issue is right here. Gonna have to loosen them up a bit and probably be able to back down the voltage too.

Sadly it's not a timing issue, it is a crap cpu issue. Even at [email protected] the cache has errors but the ram does not.

 

[Mr.Scott]

Take 2 sticks out and try it.
If it does the same thing, I'd believe CPU a little bit more.

 

[ShrimpBrime]

Fiery
AIDA64 Developer
The detected memory CRC error can mean either a RAM issue (ie. the RAM modules provide the wrong data to the CPU), or a glitch in the path between the RAM modules and the CPU (ie. the data cannot travel safely between the RAM and the CPU pipelines), or a CPU computation issue (ie. the CPU fails to use the data properly and makes computation errors on the data). Note that the data has to go through the connection between the IMC (integrated memory controller -- can be found in the CPU package for modern processors) and the memory modules, then travel to the LLC (last level cache -- in most cases it's the L3 cache), then travel one level up on the cache level (ie. into the L2 cache), then again one level up (ie. into the L1 data cache), and only then can be used by the CPU. Through all those steps a glitch can occur anywhere, but it's not possible to know at which point the data gets corrupted due to excessive overclocking or too agressive timings.

Origin of quote. https://forums.aida64.com/topic/4085-hardware-fail-only/

Best I could do for you on a quick search. Hope it helps.

 

[Mandrake4565]

Take 2 sticks out and try it.
If it does the same thing, I'd believe CPU a little bit more.

I was thinking the same Scotty.

 

[custom90gt]

It's looking more like the MB after doing some testing this morning. Two sticks seems to work fine at 3333MHz (but I only tested for 25 minutes).

However if I move those sticks to the "non-optimized" slots I can't get it to boot unless it's at 2400MHz. Testing now at 2400MHz but not sure what that will tell me. I will try those slots individually after this to see if it's just one or what. Here is to hoping Zen 2 makes up a LOT of ground.