Guide: AMD Ryzen 7 AGESA 1006 performance and DDR4 memory check June 2017

Disappointing to hear the Hero is worst for compatibility, and they still haven't put up official releases of the newer agesa bioses. Encouraging results though, thanks!

 

I have problems with the Corsair vengeance LPX 3200 MHz / CL16 version 5.39 with mobo asrock ab350M pro 4 and ryzen 5 1600. I can reach 2933 with 1.4 volts but it is unstable
You have suggestions to give me? Some magic settings under bios?

 

With 2800MHz Gskiill Ripjaws V, it works at 2933 as long as I am using just 2 x 8GB. I have the same experience with both an Asus Prime X370 Pro and Asrock X370 Taichi.

 

I am running the same ram as you at 3200Mhz with correct timings using the XMP settings in my AsRock X370 Taichi motherboard. Update the bios to the latest version with the 1.0.0.6 Agesa code and if you have already done this then look at adding some voltage to the SOC. I find that something between 1.1 and 1.2v SOC can help with memory settings.

 

Thank you, but with the latest bios i can reach only 2933... but under my bios there is no SOC setting but i think is the same of VDDP voltage. I have tried to boost the clock to 1.2 v but no luck, i've put some relaxed timings, changed proocdt to 60ohm, nothing ... :bang:
The only way for now it to wait new bios relases

 

Just upgraded my BIOSTAR x370GTN ITX motherboard to the AGESA 1.0.0.6 update. It's all good.

Got my HyperX Predator DDR4 3000 (Hynix 2 x 16, Dual Rank) to run using XMP at 2933MHz @ 1.35v (and Windows 10 now properly reports the RAM speed not the DDR speed which would be half).

Stable overclock of R7 1700 to 3.8GHz @ 1.36v.

Taking the memory from 2666 to 2933 made a significant difference, though on the old BIOS I could OC stably to 3.9GHz my CPU-Z bench @ 3.8GHz now beats the reference 5960X in both Single and Multithread. Before the BIOS update it beat it in multithread (not as significantly as it does now) and lost in single thread.

Looks like I'll be able to stick with the BIOSTAR. Ryzen is great!

Edit: I did add 0.1v to SoC. Also, this BIOS lists support for up to 4GHz RAM.

 

Haha I was JUST about to download the Chinese version of this BIOS update out of impatience (I heard it defaults to English on English boards).

I'm excited to get my memory where it should be!

 

I'm running at 3333 MHz, it runs beautifully. It even works at 3466, although it was only 99.9 percent stable. It worked in testing, just in general use it didn't feel quite right at times. I could probably have got it perfect if I played around with the bios timings. I did none of that, just straight to 3466 and 16-16-16-36, 1.37 V, no issues. I did try higher voltage with no difference. SOC is at 1.1 V, I did try higher and it made no difference either. I'm now running 3333 @ 15-15-15-36, 1.37 V and 1.1 SOC. I figure 3466 was so close, in future bios updates it should be attainable.

 

No idea to be honest. I have noticed the platform seems to behave a bit odd before actually causing problems. See what happens if backed your CPU off a bit, and kept the SOC in check (shouldn't need more than 1.1 V). Apparently having the tWCL as tCL - 1, such that if you have CAS at 15 you set it to 14, is meant to be beneficial. I guess if it doesn't like it, then it's probably the CAS that is too low to begin with, so then go 16 on tCL and 15 on tWCL.

Also for some things it may be better with SMT disabled. This isn't a CPU issue, it's programming. Ideally the programs should only see the physical cores unless they are specifically SMT mutlithreading capable and can make use of the threads, in which case they see the extra threads. It would be much better for the two threads running simultaneously on the CPU to be related processing wise, if it's two completely unrelated threads data wise, even for the same application, it could impinge on performance. At least this is the way I see it . I guess the scheduler works most of this out, but it wouldn't do it perfectly.

So there's a few things you can try.

The microstutter was the kind of thing I had with RAM at 3466. Disabling SMT also helped a little with something, but in that case it's a programming issue. It all depends on how well BF1 and everything related and running when you play it works with 8 real cores and threads and 8 SMT virtual threads.

Read here:
https://www.reddit.com/r/Amd/comments/69oafz/those_gaming_with_ryzen_do_you_leave_smt/

Specifically this post:
https://www.reddit.com/r/Amd/comments/69oafz/those_gaming_with_ryzen_do_you_leave_smt/djur2fh/

I found that after writing what I wrote above, but I think it's a generalised consensus. Seeing it as 16 real threads for multithreaded applications is potentially problematic, they should see 8 threads, and the additional 8 threads be seen only if they are programmed in a way that improves their use of the core.

If the core use is low then another process could run in the SMT thread, which wouln't likely affect the cache so much. That is, if the core use is 70 percent you can make use of the extra capacity for the different thread. That's the whole idea of SMT, however since the SMT threads are limited capacity they should be used only where benefical. Having BF1 and other games see it as 16 real cores and utilising them as such wouldn't be efficient use.

Since Intel also has similar SMT with similar issues in regards to programs and games thread use, they too could benefit from only seeing 8 processors unless a program is specifically programmed to make use of the extra threads. Ideally not only would you need to delegate processing that uses the same data to a core and SMT thread, you would need enough data to cover all the cores (whether it is 4 or 8). No point saturating 4 cores and leaving the other 4 in relative idle with other taks, so it is a balancing thing.

Also, if you are running two applications, the SMT threads could start being utilised if necessary, but as far as each of the applications are concerned they should still only see 8 threads unless specifically programmed.

 

Now that I finally got my AGESA 1006, my RAM is now running at the 3GHz it was supposed to. I could get it to 3200, but not stable. However, my timings are in the 15s, so if I bump them up a notch I could probably do 3200 stable.

My board seems to really hate it when I tamper with the Gear Down setting. I can leave everything else at defaults but just disable that, and the system is rendered unbootable. I feel like this could be part of why I'm stuck at 3000. CPU-Z claims my command rate is T1 when it should be T2. Then again, it also read the RAM voltage wrong.

 

I have done a bunch of testing on various Corsair kits. I have learned that AMD was not kidding when they said that Samsung B-Die based kits work the best. Corsair sells Vengeance LPX 3200 16GB for instance. One part number. 2 versions. One is Hynix (kit ver 5.xx) and one is Samsung (ver 4.xx). The Hynix will not run over 2666 on my MSI gaming Pro Carbon X370 on 1.0.0.4 or 1.0.0.6 Agesa versions. The Samsung is set and forget stable XMP Profile 3200. To be honest, unless you are running something faster than a GTX 1070, you will see no increase in game performance comparing RAM at 2666 to 3200. Tested myself on 6 games. RX 580 @ 1385 no differences. ASUS GTX 1070 OC no difference. Hardware Unboxed confirmed this as well.

 

Am sure you are quite aware that the only way to get 2T working is having Geardown disabled. Otherwise the CR is overridden to 1T.

 

I have heard of this, but like I said, if I disable it (even without changing anything else) my system is unbootable. If I disable it and set the command rate to 2T, it is still unbootable. I don't know what to do.

The weird thing is before the AGESA 1006 update, my RAM was running at 1T, which is why I couldn't get to 3GHz. Now I can reach 3GHz, but programs like CPU-Z claim I still am using 1T. I don't get it.

Ever since the latest BIOS update, RAM failures result in my entire CMOS being wiped, which makes experimenting with teaks a lot more tedious. The previous BIOS version would default to failsafe settings but I could still load any saved profiles.

For now, I'm just glad my RAM is running at its advertised speed (well, close enough anyway). I'd like to OC to 3200MHz but I'm not going to push my luck. Apparently, other people have had a hard time OCing my RAM, even on Intel systems.

 

@ thatguy91 - just read this over at the ROG forum:

The reply came from Raja. One of the ones on the ASUS bios team for the C6H. Take it for what it is worth.

Do not know now if 3600 is possible to get stable on my cpu if 3600 is truly where it tops out.

 

Memory OC Showdown: Frequency vs. Memory Timings on AM4 platform Agesa Microcode 1006

Memory OC Showdown: Frequency vs. Memory Timings on AM4 platform Agesa Microcode 1.0.0.6 , a great read to understand the new values on the new Bios based on Agesa 1006

The release of AGESA 1.0.0.6 to mobo makers has resulted in a wave of fresh BIOS updates, each one packed with tons of new options for tweaking and overclocking memory. With so many options at your disposal, some have understandably asked: “what are the optimal settings for games?” Never one to leave an overclocking question unanswered, our illustrious overclocker Sami Makinen took his ASUS Crosshair VI and AMD Ryzen™ 1700 CPU for a spin to find the fastest combination of settings in a few different tests.

Before we dig into the data, here’s what we analyzed:
The impact of the new BankGroupSwap (BGS) BIOS option
Single-rank DIMMs vs. dual-rank DIMMs
Automatic sub-timings vs. manually-tweaked subtimings
Max frequency vs. lower frequency at tighter timings
Geardown Mode (GDM) on vs. off

Complete detailed information here: http://forums.guru3d.com/showthread.php?t=412876&page=51

Source Link and complete report: https://community.amd.com/community...emory-oc-showdown-frequency-vs-memory-timings

 

Thanks Chispy it was a wealth of information. Only thing, why on earth did they not test GDM=disabled and BGS=enabled. It was either both enabled or just BGS off. Would guess that BGS is the new label for rank interleaving or an improved version of it.

 

Your welcome my friend , great read and a wealth of information indeed. This new information did spark on me a need to re-visit the latest Bios and find out what i can do in the memory overclocking department. 3466Mhz CL 14 tight timings seems to be the sweet spot

 

That may be the sweet spot but cannot quite get there yet. Currently able to get 3466 and CL15 stable with the F4-3600C16D-16GTZ kit. Based on synthetic benchmarks, was able to determine that the largest setting for interleaving size is favored; 2k just eeks passed 1k.

METRO 2033 (not redux) appears to run smoother with both BGS options enabled and the penalty is ever so minor.

EDIT: the settings are run with BGS enabled and GDM disabled

 

Got the AGESA 1006 updated BIOS the other day (finally!) and now my GSkill TridentZ 2800 is running at the proper speed. POST is quicker as well. I wont bother changing the RAM now

 

New Agesa 1.0.0.6.2 on Asus CH VI Hero , Official latest Bios 1403 =