Print Page | Close Window

Anyone reach >2666 DDR4 on AB350m Pro 4?

Printed From: ASRock.com
Category: Media&User's Review
Forum Name: AMD Motherboards
Forum Description: ASRock AMD Motherboards
URL: https://forum.asrock.com/forum_posts.asp?TID=7064
Printed Date: 02 Dec 2024 at 7:48pm
Software Version: Web Wiz Forums 12.04 - http://www.webwizforums.com


Topic: Anyone reach >2666 DDR4 on AB350m Pro 4?
Posted By: Ocman76
Subject: Anyone reach >2666 DDR4 on AB350m Pro 4?
Date Posted: 26 Dec 2017 at 12:36pm
I have been messing with this board for months with my F4-3200C16D-16GVKB RAM hoping I would be able to break 2666 mhz.  Even on Bios 3.3 I just can't Unhappy.

I have been all over the internet to try different timing schemes and settings with no luck.  I think my RAM just won't be able to do it (maybe its the RAM, maybe I need higher SOC voltage, who knows).  I have tried using Ryzen master to change SOC voltage but it seems to apply only after boot which doesn't work as the RAM fails to post without the extra bump to SOC.

Has anyone with non QVL RAM managed to run at higher than 2666 mhz?  If so how did you do it?


-------------
R5-1600 (stock)
Scythe Fuma
RX 570 Sapphire Nitro
G.Skill RipJaws 2X8GB DDR4 3200 @2666
Asrock AB350M Pro 4
BPX 240GB SSD
2X Barracuda 1TB RAID 1
Bitfenix Prodigy
EVGA Supernova 550 G3
Windows 10 Pro



Replies:
Posted By: nikoff90
Date Posted: 31 Dec 2017 at 12:47am
" rel="nofollow - I was able to get to 2666 did not try to go further. My system had crashing and reboot issues with any ram settings but when it worked all was fine. Crashing was happening during web browsing of all things. Youtube would really stuff it and I could almost guarantee that by a 3rd or 4th music video the system would freeze or self reboot. Gamming was flawless at any settings and never crashed. 

Mushkin REDLINE Series ??DDR4 DRAM ??16GB (2x8GB) Memory Kit DIMM ??2666MHz (PC4-21300) CL-16 ??288-pin 1.2V Desktop RAM ??Non-ECC ??Dual-Channel ??FrostByte G3 Red Heatsink ??MRA4U266GHHF8GX2


Mushkin calls the ram out at 1.25V on their website.



Posted By: RealAntithesis
Date Posted: 14 Apr 2018 at 1:28pm
I'm running at 2933 Ghz on my AB350M Pro4. Anything higher and it starts having problems though. RAM is G.SKILL RipjawsV 32GB (2x16GB) DDR4 3200Mhz C16/18 1.35V Gaming Memory Black F4-3200C16D-32GVK. I'm just using the XMP memory setting.


Posted By: PetrolHead
Date Posted: 14 Apr 2018 at 7:34pm
I'm running my G.Skill Trident Z kit (Samsung B-die) at 3200 MHz using the XMP timings for 3466 MHz. I've tried tightening the timings, but either the system becomes unstable or the gains are very small, so it's just easier to use the XMP timings. My computer does boot at 3333 MHz, but it's not stable using the XMP timings and I've been too lazy to test if loosening the timings would help.

P.S. I'm using BIOS 3.3


-------------
Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit


Posted By: PatriotVipers
Date Posted: 19 Apr 2018 at 11:58pm
" rel="nofollow -
Originally posted by PetrolHead PetrolHead wrote:

I'm running my G.Skill Trident Z kit (Samsung B-die) at 3200 MHz using the XMP timings for 3466 MHz. I've tried tightening the timings, but either the system becomes unstable or the gains are very small, so it's just easier to use the XMP timings. My computer does boot at 3333 MHz, but it's not stable using the XMP timings and I've been too lazy to test if loosening the timings would help.

P.S. I'm using BIOS 3.3

I'm on P4.70. Shouldn't you update your BIOS?

I've been trying to get my Patriot Vipers 16GB (2x8GB) 3000MHz memory kit to run @ 2933MHz and it only works for very short periods of time. Way too unstable. I'm not sure why. It was even listed on ASRocks QVL for Raven Ridge. Although there was a caveat. Even though the model # is on the QVL list they followed it up with another number in parentheses (11BF2). So, it's kind of funny that while my model number is on the QVL list I still don't get the supported 2933MHz speeds. It's not so funny actually.

I've been reading/learning that "Ryzen likes Samsung B-die". Okay, whatever. So now that we are learning that as a group, how can I know for sure when I make my next DDR4 memory purchase that I'm getting those special "Samsung B-die"'s on my memory sticks? Is it easy to figure out from the listing or is this something you gotta put a lot more time and effort into before buying the memory kit?


-------------
System: AMD 2200G, ASRock AB350M Pro4, 16GB(2x8GB) 3000MHz Patriot Vipers (running @ 2134MHz *crying*).


Posted By: PetrolHead
Date Posted: 20 Apr 2018 at 5:23am
" rel="nofollow -
Originally posted by PatriotVipers PatriotVipers wrote:

I'm on P4.70. Shouldn't you update your BIOS?


It is generally not recommended to update the BIOS if the system is working properly. Mine is. The newest version is not always the best for all systems, so even though there's a theoretical chance a newer BIOS version would make it possible for me to (for example) use tighter timings, there's also a chance that I wouldn't be able to achieve the overclock level I am at now. There's also a chance that new issues would appear - and there's always a non-zero chance of bricking the board during BIOS update. So, even though I always updated the BIOS to the newest available version on my previous system and had no issues, this time around I'm trying to stick with what works.

Quote I've been trying to get my Patriot Vipers 16GB (2x8GB) 3000MHz memory kit to run @ 2933MHz and it only works for very short periods of time. Way too unstable. I'm not sure why. It was even listed on ASRocks QVL for Raven Ridge. Although there was a caveat. Even though the model # is on the QVL list they followed it up with another number in parentheses (11BF2). So, it's kind of funny that while my model number is on the QVL list I still don't get the supported 2933MHz speeds. It's not so funny actually.


Anything over 2667 MHz is an overclock and not guaranteed, not even if the memory is on the QVL list. This is also stated in the QVL list. The achievable speeds depend on a number of things, such as the CPU's memory controller and the memory modules themselves, so it's unfortunately a bit of a lottery even when you do everything "by the book". It's also a lottery with Intel, but they've had more time with DDR4 and have more money to throw at issues than AMD, so the overall support is a lot better.

Quote I've been reading/learning that "Ryzen likes Samsung B-die". Okay, whatever. So now that we are learning that as a group, how can I know for sure when I make my next DDR4 memory purchase that I'm getting those special "Samsung B-die"'s on my memory sticks? Is it easy to figure out from the listing or is this something you gotta put a lot more time and effort into before buying the memory kit?


I'm not sure if there's any way of being certain before you actually buy the modules. When I was buying my RAM, I used this list:

https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

However, I wouldn't assume this (or any other) list to be 100% accurate, since it's not unthinkable that a manufacturer would use dies from various suppliers in a single product line (also, I believe the statement "Do note that Dual Rank RAM is faster than Single Rank. E.g. 2933 MHz Dual Rank is faster than 3200 MHz Single Rank." is actually false). Some information about the dies may be given in the motherboard's memory QVL, but that information is often incomplete. If you have a certain memory kit in mind, it may be worth checking the manufacturer's web page and see if the technical specification of the kit lists the die type.

TL;DR: Unfortunately you probably need to do some digging if you want to know.


-------------
Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit


Posted By: PatriotVipers
Date Posted: 20 Apr 2018 at 9:16am
Originally posted by PetrolHead PetrolHead wrote:

Anything over 2667 MHz is an overclock and not guaranteed, not even if the memory is on the QVL list. This is also stated in the QVL list. The achievable speeds depend on a number of things, such as the CPU's memory controller and the memory modules themselves, so it's unfortunately a bit of a lottery even when you do everything "by the book". It's also a lottery with Intel, but they've had more time with DDR4 and have more money to throw at issues than AMD, so the overall support is a lot better.

I bolded the  part of your statement I have a question about. While ASRock may not be guaranteetng anything above 2667MHz (which would be odd anyway since they are only providing a tool (1 out of 3) in the equation) I would assume that since AMD themselves "officially" support 2933MHz that the memory controller in the CPU's themselves (2200/2400G) would not be part of the "lottery" in the equation. Are you suggesting, in fact, that the MC in the CPU's is a crap shoot when it comes to achieving the "official" highest possible speeds on AMDs new CPUs?

Originally posted by PetrolHead PetrolHead wrote:

I'm not sure if there's any way of being certain before you actually buy the modules. When I was buying my RAM, I used this list:

https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

However, I wouldn't assume this (or any other) list to be 100% accurate, since it's not unthinkable that a manufacturer would use dies from various suppliers in a single product line (also, I believe the statement "Do note that Dual Rank RAM is faster than Single Rank. E.g. 2933 MHz Dual Rank is faster than 3200 MHz Single Rank." is actually false). Some information about the dies may be given in the motherboard's memory QVL, but that information is often incomplete. If you have a certain memory kit in mind, it may be worth checking the manufacturer's web page and see if the technical specification of the kit lists the die type.

TL;DR: Unfortunately you probably need to do some digging if you want to know.

Thanks for the link. I'll be doing my homework for sure. Especially given the inflated DDR4 memory prices at the moment.

Here is a Typhoon shot of my chips. I did buy these almost 3 years ago though Manufacturer: "undefined"? Lottery is the right word!


 

Good news for me (kinda). I went to Patriot's website and came across their webpage for the "Viper 4" memory kits and looked at the timings they said they had tested with. So, I thought to try them. What the heck. "16-18-18-36". As I write this message, I'm on the same PC running stable (so far now) @2934MHz. In the beginning I was using timings I got from the XMP that I got from the SPD information using AIDA64 software. They were as follows:

@ 1500 MHz 16-16-16-36  (CL-RCD-RP-RAS) / 70-390-240-165-9-6-33  (RC-RFC1-RFC2-RFC4-RRDL-RRDS-FAW) 

I know next to nothing about memory timings but I came across a comment in another thread that was very scary to me. He mentioned something about the possibility that 2 different "rated" memory speeds could actually be achieving(or producing) the same usable bandwidth as an end result @ different speeds because of timings if they were set a certain way. In other words, let's say 1 PC is running a memory kit @ 2666MHz with X timings and another PC is running a memory kit @ 2934MHz with X timings. But it's possible that "both" sets are producing the same usable bandwidth because your timings on the faster running memory kit are 1)horrible per the genetics of the memory chip ( In other words the memory manufacturer used horrible ICs and is hoping you don't know anything about timings, which is the camp I fall into) or 2) you've just set up the wrong timings as an end consumer. I'm thinking case #2 is my case as I write this.

I was getting a 74.89% "Float Memory Bandwidth" efficiency in SiSoft Sandra with the earlier unstable timings. Now, I'm getting  67.84%. Can anybody help me tweak these timings in the screenshot? I want to get the maximum usable bandwidth I can out of these Vipers (not to be fooled with, seriously, but respected). And, of  course, I want the maximum stability. There's a sweet spot that I need help finding.



These are both my benchmarks with different memory timings set up in the BIOS.

The "Memory Read" benchmark in AIDA 64 Extreme looks so much higher than SiSoft Sandra. I'm guessing they are doing different measurements. Anybody know which is the better indicator of your performance though?



Oh well, I'm feeling (and liking) the stability at the moment but I want that extra performance back. Maybe I can keep tweaking and get the efficiency (and the "usable bandwidth") back up a bit.



-------------
System: AMD 2200G, ASRock AB350M Pro4, 16GB(2x8GB) 3000MHz Patriot Vipers (running @ 2134MHz *crying*).


Posted By: PetrolHead
Date Posted: 20 Apr 2018 at 4:34pm
" rel="nofollow -
Originally posted by PatriotVipers PatriotVipers wrote:

I bolded the  part of your statement I have a question about. While ASRock may not be guaranteetng anything above 2667MHz (which would be odd anyway since they are only providing a tool (1 out of 3) in the equation) I would assume that since AMD themselves "officially" support 2933MHz that the memory controller in the CPU's themselves (2200/2400G) would not be part of the "lottery" in the equation. Are you suggesting, in fact, that the MC in the CPU's is a crap shoot when it comes to achieving the "official" highest possible speeds on AMDs new CPUs?


My bad, I was only thinking about Ryzen 1xxx CPUs. I know the memory controller of the 2200G and 2400G should be improved compared to the first generation parts, but curiously AMD still states on their webpage that the maximum stock memory speed is 2667 MHz. Then again, it's possible the website is wrong. I do remember the number 2933 MHz being quoted in reviews. Regarding the IMC, I don't think reaching maximum stock speeds should be much of a crap shoot - assuming no overclocking has been done.

Quote I know next to nothing about memory timings but I came across a comment in another thread that was very scary to me. He mentioned something about the possibility that 2 different "rated" memory speeds could actually be achieving(or producing) the same usable bandwidth as an end result @ different speeds because of timings if they were set a certain way. In other words, let's say 1 PC is running a memory kit @ 2666MHz with X timings and another PC is running a memory kit @ 2934MHz with X timings. But it's possible that "both" sets are producing the same usable bandwidth because your timings on the faster running memory kit are 1)horrible per the genetics of the memory chip ( In other words the memory manufacturer used horrible ICs and is hoping you don't know anything about timings, which is the camp I fall into) or 2) you've just set up the wrong timings as an end consumer. I'm thinking case #2 is my case as I write this.


I am no expert when it comes to memory timings, but I believe you are essentially correct. It's a balancing act between clock speed and timings and in some cases the looser timings that are required for RAM to work at a higher clock speed may indeed make matters worse. However, if we are talking about XMP timings, the differences in timings shouldn't (AFAIK) be big enough to counteract the change in clock speed. There's a caveat to this as well, though, since a lot of the timings are on "auto" by default, so you might have two memory kits with the "same" XMP profile but which actually give you different performance because the motherboard can find better "auto" values for one of the kits. I'll try to dig up some articles I've come across later on for you to check out. I'll also comment more on optimizing memory timings and benchmarks later,  I'm unfortunately a bit pressed for time at the moment.


-------------
Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit


Posted By: PatriotVipers
Date Posted: 21 Apr 2018 at 1:14am

Uptime over 16+ hours now on Windows 10 @ 2934MHz. I ran the same benchmarks again with the stable timings and I'm getting almost the same bandwidth I was getting using the unstable timings:



I guess running the same benchmark multiple times is a must as results will vary from test to test. I'm very happy at this point with these Vipers. Don't mess with the Vipers. They don't play games! :)


-------------
System: AMD 2200G, ASRock AB350M Pro4, 16GB(2x8GB) 3000MHz Patriot Vipers (running @ 2134MHz *crying*).


Posted By: PetrolHead
Date Posted: 21 Apr 2018 at 3:57pm
" rel="nofollow - I haven't run SiSoft Sandra in a while, so I don't know how it compares with AIDA 64, but I think it's always a good idea to run at least a few different benchmark programs after making changes. This will give you a more complete picture of how the changes affect performance. It's also a good idea to run each benchmark at least three times to see how much the results vary between runs.

I've used both AIDA 64 and PassMark's PerformanceTest V9.0 to measure memory performance. You can read more about my own results and attempts at tightening timings from here: http://forum.asrock.com/forum_posts.asp%3FTID=6971&PID=42028&title=ryzen-5-1500x-ab350m-pro4#42028" rel="nofollow - http://forum.asrock.com/forum_posts.asp?TID=6971&PID=42028&title=ryzen-5-1500x-ab350m-pro4#42028 http://forum.asrock.com/forum_posts.asp%3FTID=6971&PID=42028&title=ryzen-5-1500x-ab350m-pro4#42028" rel="nofollow -

Edit: I can't seem to get the above hyperlink to work. It shows the preview correctly and if you just copy and paste that url to a new tab, you'll get there, but the hyperlink itself will throw a 404 error.

If you want to see what sort of relative differences there can be between different kits of similar memory due to BIOS optimizations, see this article (paragraph "Relative Memory Performance"): https://www.gamersnexus.net/news-pc/3283-ryzen-2-pre-test-x470-vs-x370-scaling-memory-r7-2700x" rel="nofollow - https://www.gamersnexus.net/news-pc/3283-ryzen-2-pre-test-x470-vs-x370-scaling-memory-r7-2700x

Here's an article testing the effect of different memory settings on both games and synthetic benchmarks: https://community.amd.com/community/gaming/blog/2017/07/14/memory-oc-showdown-frequency-vs-memory-timings" rel="nofollow - https://community.amd.com/community/gaming/blog/2017/07/14/memory-oc-showdown-frequency-vs-memory-timings

If you want to know what the different timings mean, there are probably several lists available online. Here's one: https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram" rel="nofollow - https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram

Unfortunately I wasn't able to find the one article I had in mind about increased speed resulting in worse results due to the extremely loose timings required to keep that speed stable. I'll keep looking.


-------------
Ryzen 5 1500X, ASRock AB350M Pro4, 2x8 GB G.Skill Trident Z 3466CL16, Sapphire Pulse RX Vega56 8G HBM2, Corsair RM550x, Samsung 960 EVO SSD (NVMe) 250GB, Samsung 850 EVO SSD 500 GB, Windows 10 64-bit



Print Page | Close Window

Forum Software by Web Wiz Forums® version 12.04 - http://www.webwizforums.com
Copyright ©2001-2021 Web Wiz Ltd. - https://www.webwiz.net