Ryzen 7000 / Zen 4 Owner's Club

[ENTERPRISE]

2 hours ago, The Pook said:

 

DDR5's "ECC" is on-die so it might just be fixing it and not reporting it to Windows. 

 

CMD/Terminal -> wmic memphysical get memoryerrorcorrection

 

If it returns 0, 1, 2, or 3 then Event Viewer wouldn't be able to tell you about corrected errors. if it's correcting errors in the background it'd make sense that it's passing stability tests and performing worse in benchmarks though. 

 

I will run that and see what it comes back with. 

 

So I adjusted tRDRDSD/tRDRDDD to 11 for my 6600 OC and that certainly helped, I was getting 5m 32s vs 6m 12s. 

 

6400 is still better at 5m 9s, so I will have to investigate what else I need to adjust to get 6600 on par or better. If anyone has any suggestions, it would be appreciated, though I am trying to see if I can keep CL28.

[neurotix]

Hey, some of you guys especially @kalizshould get on hwbot with these high memory speeds you're posting. Here's some benches to beat:

 

/img/logo.png

neurotix`s y-cruncher - Pi-1b score: 16sec 765ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5289.8MHzscores getScoreFormatted in the y-cruncher - Pi-1b benchmark...

 

/img/logo.png

neurotix`s y-cruncher - Pi-2.5b score: 46sec 441ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5231.4MHzscores getScoreFormatted in the y-cruncher - Pi-2.5b benchmark...

 

/img/logo.png

neurotix`s GPUPI for CPU - 1B score: 49sec 231ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5231.4MHzscores getScoreFormatted in the GPUPI for CPU - 1B benchmark...

/img/logo.png

neurotix`s GPUPI for CPU - 100M score: 2sec 733ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5143.6MHzscores getScoreFormatted in the GPUPI...

 

Edited October 14, 2023 by neurotix

[Hurricane28]

1 hour ago, ENTERPRISE said:

 

I will run that and see what it comes back with. 

 

So I adjusted tRDRDSD/tRDRDDD to 11 for my 6600 OC and that certainly helped, I was getting 5m 32s vs 6m 12s. 

 

6400 is still better at 5m 9s, so I will have to investigate what else I need to adjust to get 6600 on par or better. If anyone has any suggestions, it would be appreciated, though I am trying to see if I can keep CL28.

 

What AGESA are you on?

 

I think its a limit of the AGESA and we have to wait for new ones to come out. 

 

There is little to no difference or hardly worth the time now to spend on memory OC as there is no gain as you can see in various benchmarks. 

 

6400MHz seem to be the sweet spot with newer AGESA while the older was 6000MHz so i think next round will be better. 

[neurotix]

A new bios with new AGESA just dropped for my board a few days ago. Haven't gotten to installing it yet. It is Combo AM5 1.0.8.0.

 

Despite being on an older BIOS, I have seen plenty of gain in benchmarks for hwbot even when I was OCed to 6200mhz cas 30 with G.skill Flare X. Now with these 2x24 TridentZ5 M-Die modules at 7600MHz, I've seen fantastic performance improvements. Not just in hwbot benches either, the real world performance gain is there in games too.

 

Forza Horizon 5 benchmark with stock: 190 fps

Forza Horizon 5 benchmark with B-Die 6200mhz: 195 fps

Forza Horizon 5 benchmark with M-Die 7600MHz: 199 fps

 

This was done with my Ryzen 7900X3D in Balanced power plan so the V-Cache was active, and no OC/stock power limit on my RTX 4090. Resolution was 3440x1440.

 

I'm seeing performance benefits from OC'ing my memory and so is pretty much everyone else here.

 

Blame AGESA all you want but it seems your system is unstable and has defective memory or something, or maybe you just don't want to do the work and enter timings manually to get a stable OC. So please stop posting misinformation claiming RAM oc doesn't matter, because it does. Even if the difference is sometimes small or non-existent because of the application. ¯\_(ツ)_/¯

Edited October 14, 2023 by neurotix

[Hurricane28]

49 minutes ago, neurotix said:

A new bios with new AGESA just dropped for my board a few days ago. Haven't gotten to installing it yet. It is Combo AM5 1.0.8.0.

 

Despite being on an older BIOS, I have seen plenty of gain in benchmarks for hwbot even when I was OCed to 6200mhz cas 30 with G.skill Flare X. Now with these 2x24 TridentZ5 M-Die modules at 7600MHz, I've seen fantastic performance improvements. Not just in hwbot benches either, the real world performance gain is there in games too.

 

Forza Horizon 5 benchmark with stock: 190 fps

Forza Horizon 5 benchmark with B-Die 6200mhz: 195 fps

Forza Horizon 5 benchmark with M-Die 7600MHz: 199 fps

 

This was done with my Ryzen 7900X3D in Balanced power plan so the V-Cache was active, and no OC/stock power limit on my RTX 4090. Resolution was 3440x1440.

 

I'm seeing performance benefits from OC'ing my memory and so is pretty much everyone else here.

 

Blame AGESA all you want but it seems your system is unstable and has defective memory or something, or maybe you just don't want to do the work and enter timings manually to get a stable OC. So please stop posting misinformation claiming RAM oc doesn't matter, because it does. ¯\_(ツ)_/¯

 

 

I am waiting for that AGESA too and will test more when it arrives. 

 

Ofcos in benchmarks you see gains cos the numbers are higher, but they rarely translate to more performance 24/7.

You also need to take in to consideration that you are using an complete different setup with an ROG board which is higher end than my board witch different memory traces and board layout which also can be limiting factor. 

 

I am not new to this again, i am with Ryzen since the beginning and i never saw significant increases in memory Overclocking on any of my x370 or x470 outside benchmarking. 

 

From 6200 to 7600MHz is a big jump in frequency but not so much in performance cos 195 to 199 is not even noticeable, just saying. 

 

I noticed lower latency when running higher than 6000MHz but that's about it, this is due to lower timings but the bandwidth is about the same, i tested this in real world applications and i noticed no difference at all when copying filed to my m.2 SSD. Large or small, they are about the same time with different speeds of ram. 

 

The gains were bigger on my 5600x witch DDR4 from 3200 tight to 3800MHz tight timing with the mus1mus calculator but at the end i spend like weeks and weeks of training and benching for only 10% at best in very specific workloads. 

 

I did not blame AGESA, i said I THINK its the limit cos what i saw is that on the newer AGESA i could get higher ram speeds but the BIOS was not stable cos its BETA. 

Idk, my memory works just fine atm. and i did spend time in tweaking it, what are you talking about, you saw my sceenshots lol. 

 

Like i said, i am testing again when the new AGESA is out cos it will be in a few days and i have to start all over again when i flash it because the values are different. 

So for me now personally, its a waste of time really as i cannot get higher atm on the current AGESA which is 1.0.0.6 which is an older bios and with every new AGESA there is a notice that says "improves memory OC" that is not for nothing. So based on my own testing and by the info from my friends and from the manufacturer i can say that memory oc can be fun but is just not worth it if you look for performance and your numbers concur with that. You can make your system feel more snappy but that's about it. 

 

 

Also, with an CPU like mine with only one CCX, its not worth it to go beyond lets say 6400 or 6200 cos i don't have the infinity fabric speed to fully benefit from it. That is also an limiting factor for me. 

Edited October 14, 2023 by richardmeppelin

[Hurricane28]

Okay, i made an memtest86 stick and let the program run overnight and when i woke up i saw that is passed at somewhat stock settings. 

 

The cool thing about Aorus is that they have a need feature where you can see the timings in hwinfo64. You can access it via clicking the hwinfo64 icon on the right bottom in the hidden icons tap, than click timings. 

 

HWinfo64 also has an Aorus profile which looks kinda need, i like these small little things. 

[kaliz]

10 hours ago, neurotix said:

Hey, some of you guys especially @kalizshould get on hwbot with these high memory speeds you're posting. Here's some benches to beat:

 

/img/logo.png

neurotix`s y-cruncher - Pi-1b score: 16sec 765ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5289.8MHzscores getScoreFormatted in the y-cruncher - Pi-1b benchmark...

 

/img/logo.png

neurotix`s y-cruncher - Pi-2.5b score: 46sec 441ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5231.4MHzscores getScoreFormatted in the y-cruncher - Pi-2.5b benchmark...

 

/img/logo.png

neurotix`s GPUPI for CPU - 1B score: 49sec 231ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5231.4MHzscores getScoreFormatted in the GPUPI for CPU - 1B benchmark...

/img/logo.png

neurotix`s GPUPI for CPU - 100M score: 2sec 733ms with a Ryzen 9 7900X3D

HWBOT.ORG

---

The Ryzen 9 7900X3D @ 5143.6MHzscores getScoreFormatted in the GPUPI...

 

 

if i had a 12 core CPU i would try to beat that, 

 

the only benchmark you can compare is GPUPI 100M , i only have 6 and 8 core CPU's

 

So this Dom guy from overclock.net DDR5 forums was busy the whole day to beat it by 2 hundreds  so its a pretty sharp score

 

/img/logo.png

kaliz`s GPUPI for CPU - 100M score: 2sec 608ms with a Ryzen 7 7800X3D

HWBOT.ORG

---

The Ryzen 7 7800X3D @ 5451.5MHzscores getScoreFormatted in the GPUPI for CPU - 100M benchmark. kalizranks #4 worldwide and #2 in the hardware class. Find out more at HWBOT.

  

[kaliz]

Neurotix, you have lots of headroom for those GPUPI benchmarks, you should at least set GPUPI to batch and reduction sizes 128 4, and disable IOL vectorizer. You will beat me in one run. And use this program: to mess with curve optimizer in windows. SMUdebugtool 1.3.2 . This is a nice start but you can get at least 0,2 seconds off!

[kaliz]

and in Ycrucncher too, you can run it realtime prio, and use a .bat file to make it run in REAL realtime prio for another couple hundreds, you can get to LN scores if you really really try! Also try 6600 in 1:1 mode with 2200 fclk for it, sometimes its almost half seconds faster as in 1:2 mode. Make a word documnet with this text: (and save as . bat)

 

wmic process where name="22-ZN4 ~ Kizuna.exe" CALL setpriority "Realtime"

 

and run it as admin after you press start in ycruncher to set in in realtime prio. This is the way Buildzoid had his topscore on YT couple months ago btw.

Edited October 15, 2023 by kaliz

[neurotix]

Thanks for the advice kaliz. I'll try this stuff out eventually next time I bench.

[kaliz]

your scores are already very good especially with AiO! That cooler is proper strong!

[neurotix]

1 hour ago, kaliz said:

your scores are already very good especially with AiO! That cooler is proper strong!

I thought the same. Like, why redo them if they are all gold cups already afaik? I suppose doing it would simply make my scores harder to beat. I guess I could do them again with your suggestions just to see how well I can do and top myself.

 

Yes, this cooler is fantastic, it keeps the chip under 70c running y-cruncher, Cinebench etc. My old h150i would be at like 80c

Edited October 15, 2023 by neurotix

[ENTERPRISE]

Well I think for now I am happy with my OC. 

 

After tweaking timings, I was able to get my 6600 OC running SuperPi where it ought to be. After some further BIOS and AGESA updates I will see if I can achieve anything better but for now I am pleased with this. 

 

 

Thanks everyone for your help!

[kaliz]

[kaliz]

tried again yesterday but it was half second slower  this was with chilled water and the cheapest B650 mobo , too bad it didnt give points on hwbot this was pretty decent for a Ryzen!

[kaliz]

 

Tested a different CPU + Mobo combo and this setup was stable 5 cycles in TM5 for what its worth , its pretty tight, could do 26-36-36 but it gets hot eventually.  you need a fan on top of the ram to keep it <45C VDD 1,69 VDDQ 1,62 anything lower gave errors, also MISC at 1,2 was a test its ok at 1,1v also i left VDDP auto. This is pretty much the settings the guys with AM5 on LN2 bench with @6400 on hwbot, maybe bit looser subs. If you want to use lower voltages, every primary timing needs at least 2 ticks more, but the second and third you can leave

 

[J7SC_Orion]

I admire the DDR5 voltages some of you folks are running...per recent post, I've used up to 1.5 V once for less than 30 minutes for 8200 in an otherwise daily config.

 

For the daily 8000, I've updated some voltages ever so slightly (per Zen on the right side). The setting on the left (GPU benching) uses the same voltages now. In addition, I have a CL 30 6400 at 1.42 V. After trying various minor timing mods for the DDR5 8000, I always come back to where it is below at the normal voltages.  I am also still having fun with 'eclk' as opposed to 'bclk' on this board; I'll post some results for eclk in a bit. ...still would like to break 800 in CPUZ single thread score on regular ambient custom water cooling...not too far from it and a few more eclk tuning steps might do it.

 

[ENTERPRISE]

Hey guys, 

 

Can someone give me a screenshot of the ability to enable "Round Trip Latency" on an AMD X670 BIOS ? 

 

This is just for continuing discussions with MSI.

 

Cheers.

[Hurricane28]

2 hours ago, ENTERPRISE said:

Hey guys, 

 

Can someone give me a screenshot of the ability to enable "Round Trip Latency" on an AMD X670 BIOS ? 

 

This is just for continuing discussions with MSI.

 

Cheers.

I can in a bit if my BIOS has that setting. 

 

What does it do actually? 

[ENTERPRISE]

51 minutes ago, richardmeppelin said:

I can in a bit if my BIOS has that setting. 

 

What does it do actually? 

That would be great. 

 

Quote

Hello All,

 

RTL is the length of time it takes for a signal to be sent to the memory plus the length of time it takes for an acknowledgment from memory of that signal to be received. The complete round trip time taken for a signal to be transmitted from point A to point B and then back to point A.

 

In BIOS setup, when RTL is enabled, it will minimize RT time to improve system performance.

 

In BIOS 42 implementation RTL is set to disabled by default.

 

 

 

I hope this helps,

 

Ronny G

Source 

[Hurricane28]

16 minutes ago, ENTERPRISE said:

That would be great. 

 

Source 

I was looking in the BIOs but couldn't fine it, sorry. 

 

 

[J7SC_Orion]

4 hours ago, ENTERPRISE said:

That would be great. 

 

Source 

 

you mean this from y Aorus 670E Master bios ? Also note @kaliz screen shot on p11

 

 

 

 

[ENTERPRISE]

32 minutes ago, J7SC_Orion said:

 

you mean this from y Aorus 670E Master bios ? Also note @kaliz screen shot on p11

 

 

 

 

The one Kaliz sent they rejected saying it was from an Intel BIOS which I need to look into again. So I thought I would ask again.

 

I cannot see it in your screenshot either as something you would "Enable"

[neurotix]

I looked through every section of my bios including AMD CBS, DRAM Timing, and AMD Overclocking, but could not find this option unless it's named something else.

 

Asus Rog Strix X670E-E

[ENTERPRISE]

11 hours ago, neurotix said:

I looked through every section of my bios including AMD CBS, DRAM Timing, and AMD Overclocking, but could not find this option unless it's named something else.

 

Asus Rog Strix X670E-E

Possible its only an Intel thing then.