AGESA 1.0.0.6 and later thread

[drmrlordx]

The new AGESA version from AMD has changed a lot for Ryzen systems that have implemented it. I am running UEFI rev. 2.34 on my X370 Taichi, and the new memory straps are . . . interesting indeed.

I *almost* have DDR4-3600 stable as a daily driver. All I did was use the UEFI's XMP system to implement the DIMM's XMP settings (Corsair Vengeance LPX DDR4-3733 CAS/CL17) and set the strap to DDR4-3600. Then I set vDIMM to 1.4v and SoC voltage to 1.175v. I also manually set CR to 2T but it doesn't seem to do anything . . . performance is the same with 1T and 2T, and all the monitoring software is still reading 1T so, I dunno. I didn't gain any stability from changing the setting either (that I can tell).

I set ProcODT to 60 ohm.

Right now it's stable using these settings (DDR4-3600 18-19-19-39; Ryzen automatically sets CAS/CL to 18 here) and I'm getting over 53 GB/s read and 69.5ns latency according to AIDA64.

Sadly it isn't yet 100% stable since it crashes in certain workloads like y-cruncher and SuperPi Blend.

Ultimately I'd like to get DDR4-3733 or higher, but at this rate, it looks unlikely, at least using this particular UEFI revision. Any hints on what else I should be tweaking to gain memory clock?

[I.nfraR.ed]

I was able to get 3600C16 and 3600C14 stable even with much older bioses, but with bclk oc. With B-die ofcourse.
The new AGESA and beta bioses somewhat improve overclocking of 2x and 4x dual-rank configurations, but so far I'm not really impressed.
We still don't have full control.

[Johan45]

I agree even stability at higher ram speeds have been compromised

[drmrlordx]

bclk OC didn't work for me very well. It might have been something to do with my NVMe SSD. I have found some success with the new memory straps using 2.34 on my X370 Taichi thanks to the new subtimings, though there is one subtiming - TrdrdSc - that is the devil. It seems to do no good to increase it, while significantly reducing memory bandwidth. Increasing TrdrdSc from 1 to 5 reduced memory bandwidth to 1/2 of normal. Ugh.

My mistake was trying to use the XMP subtimings from my RAM as a baseline for tuning. Instead what I did was use Ryzen Master to set the usual subtimings for DDR4-3200, and then I tweaked from there, raising timings/subtimings where necessary to gain stability. DDR4-3333 was actually stable at 14-14-14-28 with all the same subtimings that Ryzen Master had set for DDR4-3200, which was a nice discovery.

DDR4-3466 required CAS/CL of 16. I haven't fully explored reducing the other timings at that strap, but I appear to have it mostly stable at 16-16-16-36.

Now I am working on DDR4-3600.

[chew*]

I dunno. Taichi is an improvement...and almost competitive in 32m now....almost..not the most effecient but very close to the pack...much closer than msi.

Im @ c12 3466...benched sandra full suite @ 3733 14-13-13

No clue on the 24/7 stable stuff yet. Hammering benches out atm...

[FlanK3r]

I hope, after I will update new agesa and beta BIOS at Crosshair, can do later reflash witth 1001/1002 BIOS...

[NEOAethyr]

I read somewhere, I think guru3d?
That this update brings with it some sub timings in the bios.
Is that true?, if it is what timings are there?
And does it make the 1st gen ryzen worth something you could recommend to a less extreme user ^^ ?

What I mean is, does it get the latency down to something decent like in the mid 50's (ns) with ease.

[chew*]

Latency is just a number...you can get better real world performance yes.

[Charles Wirth]

Ryzen is good now with current updates, I wish this was how it was released.

Just watch AIDA64 bandwidth test

[FlanK3r]

Im testing now beta and seems really a bit better

[Shocker003]

Im testing now beta and seems really a bit better

please can you post a link to the beta bios

[cbjaust]

please can you post a link to the beta bios

Check elmor's post here on OCN...

[Shocker003]

Check elmor's post here on OCN...

Thanks

[Shocker003]

Bios 9943 rocks!!! Finally running my ram @3066mhz. I will work on the timing tomorrow.
Thanks alot cbjaust

[drmrlordx]

I read somewhere, I think guru3d?
That this update brings with it some sub timings in the bios.
Is that true?, if it is what timings are there?
And does it make the 1st gen ryzen worth something you could recommend to a less extreme user ^^ ?

What I mean is, does it get the latency down to something decent like in the mid 50's (ns) with ease.

There are more subtimings than I can reliably remember, but it's a lot of them. And I thought my old GD70 790FX BIOS was complex, sheesh. What a mess.

The lowest I've gotten my latency so far is in the 65ns range while running DDR4-3600. Getting it closer to stable, just don't have a lot of free time. Interestingly enough, raising the primaries past 16 seems to increase instability. Odd.

On my rig, raising memory clock seems to have the most beneficial effect on latency, which is interesting. Ryzen seems to like RAM clockspeed more than anything else.

[drmrlordx]

X370 Taichi UEFI rev. 2.36 is out, and it is very interesting:

https://forums.anandtech.com/threads...#post-38922533

So far got DDR4-3466 14-14-14-28, trying for more later today.

[chew*]

Read it. Ddr 3466 12-11-11 is 32m stable as is 14-13-13 3600 without flashing to latest. Just a matter of dialing in the right timings/voltages.

Fwiw...my vddp is .900 my soc is 1.00..

Im wary of bios's that magically improve speed...usually means they magically reduced performance to do so.

Btw for amd trfc 132/192/312 @ 3466?

That is like an intel spec for 3200.

For 3600 intel spec its like 289/469/631

Fyi flare x is 256/416/560 @ only 3200.....

Maybe that is where some of your difficulty is coming from..

[ihog6hog]

I see this picture (Gigabyte B350 Gaming 3 & G.Skill 3975.2MHZ) on Facebook , but dont' know for dram voltages.



https://www.facebook.com/VGA.Social....type=3&theater

[drmrlordx]

Read it. Ddr 3466 12-11-11 is 32m stable as is 14-13-13 3600 without flashing to latest. Just a matter of dialing in the right timings/voltages.

I would be interested in knowing what timings/voltages you used. I'm assuming you achieved that on your X370 Taichi?

Btw for amd trfc 132/192/312 @ 3466?

That is like an intel spec for 3200.

For 3600 intel spec its like 289/469/631

Fyi flare x is 256/416/560 @ only 3200.....

Maybe that is where some of your difficulty is coming from..

I was surprised by that as well, but increasing trfc numbers did not seem to help much. The XMP values for trfc from my Vengeance LPX 3733 DIMMs are in the 900 range. Using those doesn't seem to help with stability at higher clocks. Raising some timings seems to reduce stability, though I haven't yet linked high trfc to crashes/failed boots yet.

I did notice stability decrease as I lowered tCWL and tRTP.

I see this picture (Gigabyte B350 Gaming 3 & G.Skill 3975.2MHZ) on Facebook , but dont' know for dram voltages.

Wonder if he can stress test at that RAM speed? I would also like to see his Aida64 results for memory read and latency.

[chew*]

Here you go...

One is 3433 or 3466 @ C12 can not remember blank cpu-z not very helpfull to me lol.

C14 is 3600.

[drmrlordx]

Such a tease . . . heh

I'll try to work more on 3600 this weekend. 3466 is stable now (thankfully) though it can be dicey getting it there sometimes. Shotgunning still doesn't work 100% of the time.

Plus there's a new UEFI version to play with.

edit: 2.40 doesn't seem to have changed much. Now down to the nitty gritty.

Current speed in testing is DDR4-3600. I am using the following timings (bear with me, can't effectively screenshot UEFI):

tCL 16
tRCDRD 16
tRCDWR 16
tRP 16
tRAS 34
tRC 75
tRRD_S 6
tRRD_L 8
tFAW 27
tWTR_S 5
tWTR_L 10
tWR 18
tMAW.MAC 1
TrdrdScL 4
TwrwrScL 4
tRFC 900
tRFC2 600
tRFC4 300
tCWL  14
tRTP 10
Trdwr 10
Twrrd 3
TwrwrSc 3
TwrwrSd 8
TwrwrDd 8
TrdrdSc 1
TrdrdSd 6
TrdrdDd 6
tCKE 8
ProcODT Auto

CR is 1T. 1.2v SoC, 1.4v vcore, 1.42v vDIMM, 1.05v VDDP

Not stable. It'll run mostly, but it still bombs out in y-cruncher, and it also crashed during Skydiver physics. HCI Memtest picked up write errors about 4% of the way through.

Also with these timings, I got better performance with my 3466 settings.

[chew*]

Yah the imcs are dicey and cost a hit if pushed to hard.

[drmrlordx]

The only thing I can think to do now is to decrease each timing/subtiming until I can cause a noboot situation. Then maybe I have something I need to increase in order to promote stability. Inreacing trfc/trfc2/trfc4 values does not seem to have helped much.

Gonna take a break then get back to it later.

[chew*]

You can try setting twcl to 15/16...may gain you a bit of stability. Tras probably should run 36 with 16-16-16

Trc 58 should be loose enough

[chew*]

This bombed out around 6hrs...dropped one core...so its very very close. Working my way up for 3600 15-15-15-35 so subs are a bit slack atm. Using flare x trfc 1/2/3 currently.