Ryzen 7000 / Zen 4 Owner's Club

[kaliz]

E-cores are so dumb just cant believe they added more this gen. Benchmarkboosters Buildzoid called them, Hurricane,  dont listen to BZ too much. tFAW 20 is a joke of him .  16 or 32 , or if RRD-s is 6, 24. Rest of the timings look good and i see you get <60ns latency. If you really try, you can get 49ns, 48ns even lower, good luck  

[Hurricane28]

15 minutes ago, kaliz said:

E-cores are so dumb just cant believe they added more this gen. Benchmarkboosters Buildzoid called them, Hurricane,  dont listen to BZ too much. tFAW 20 is a joke of him .  16 or 32 , or if RRD-s is 6, 24. Rest of the timings look good and i see you get <60ns latency. If you really try, you can get 49ns, 48ns even lower, good luck  

Well, these E core CPU's do perform wel tho. Still a bit weird architecture for sure but they work. 

 

BZ is rambling too much so i don't listen that much to him lol. 

 

you really think it will work? I mean, i think i am already at low latency. 

[Hurricane28]

@kaliz

What do you think is possible with 4 sticks of RAM? 

 

I mean, i can sell the other kit or have it as spare, but more ram is better lol. 

[Hurricane28]

New high, or rather low score as the latency is lower lol.

 

only thing i changed was from auto to 65k refresh, the maximum. 

 

tFAW cannot be lower than 20 sadly as the bios won't let me. 

[neurotix]

19 hours ago, Hurricane28 said:

New high, or rather low score as the latency is lower lol.

 

only thing i changed was from auto to 65k refresh, the maximum. 

 

tFAW cannot be lower than 20 sadly as the bios won't let me. 

 

It is well known for Zen 4 to change tREFI to 65535. On Asus boards (what I have), it gives like 20000GB/sec more bandwidth. That's only on Asus boards but I am unsurprised to see you getting better results with it set that way. Good job

[Hurricane28]

1 hour ago, neurotix said:

 

It is well known for Zen 4 to change tREFI to 65535. On Asus boards (what I have), it gives like 20000GB/sec more bandwidth. That's only on Asus boards but I am unsurprised to see you getting better results with it set that way. Good job

Okay, Well i am relatively new to DDR5 overclocking so i learned something lol. 

 

Buildzoid said 50k is best for 6kMHz RAM as its the "sweet spot" I changed it to the max and got even better result on both 6 and 6400MHz, so, yeah lol. 

 

Lower than this are the timings not gonna be so this far is the best i can get. I am happy ofcos so i don't have to change anything as this is also the limit bandwidth of my cpu as far as i seen on the net. 

[neurotix]

Yep. As I said a few posts ago, set it and forget it with your current results, but you did even better. You are probably right; no other ways you can improve it and going to high frequency RAM won't matter on your CPU.

 

What do you use your PC for? If it's mostly gaming, latency matters more than bandwidth but bandwidth does make a difference (it did for me anyway).

 

Btw, these are my current AIDA results if you're curious.

 

[J7SC_Orion]

Per below, I tried tRRDS 8 > 6 and tFAW 32 >24 on the 7950X3D with all else the same...as expected, it did not make any difference at all outside normal run-to-run variance (I used the TimeSpy CPU test as control). IMO, the memory controller just ignores tRRDS below 8 with DDR5 (unlike DDR4)...

 

[Hurricane28]

9 hours ago, neurotix said:

Yep. As I said a few posts ago, set it and forget it with your current results, but you did even better. You are probably right; no other ways you can improve it and going to high frequency RAM won't matter on your CPU.

 

What do you use your PC for? If it's mostly gaming, latency matters more than bandwidth but bandwidth does make a difference (it did for me anyway).

 

Btw, these are my current AIDA results if you're curious.

 

 

Ye i tried even lower but as of now i am more than pleased with the results. Everything feels so snappy and quick. 

 

I love latency more than bandwidth but i wanna do more graphical work like in Davinci resolve and other programs like that and i saw that 32GB is not that much when working in those programs so i mighty use the other kit with this new RGB kit i have. 

 

I wonder what timing i can expect as i have to set the timings manually when using 4 sticks cos there is no profile to select (the lazy way) lol. 

 

nice result tho. 

[ENTERPRISE]

So I installed the new BIOS for my board with the 1.0.0.8 AGESA, true to their word MSI did include the Gear Down Mode option, happy days. However looks like my setup does not appreciate it being disabled as after I get to Windows I get a BSOD. So on it stays!

[kaliz]

11 hours ago, J7SC_Orion said:

Per below, I tried tRRDS 8 > 6 and tFAW 32 >24 on the 7950X3D with all else the same...as expected, it did not make any difference at all outside normal run-to-run variance (I used the TimeSpy CPU test as control). IMO, the memory controller just ignores tRRDS below 8 with DDR5 (unlike DDR4)...

 

 

with my Hynix A-die kits, at DDR5 8000 and beyond, RRD, FAW and WTR can go as low as 8 4 16 16 6 , and for lower freq, 6 4 16 14 4, and if you want to be sure, you can set 8 8 32 24 6.

 

It does matter in latency and throughput. You can test this with Pyprime , YC, 7zip and those kind of benchmarks. With Hynix M-die you can set 4 4 16 14 4. RRD on A-die needs a bit more. 6 is really good, 7 good, 8 normal.

[ENTERPRISE]

On 20/10/2023 at 05:08, J7SC_Orion said:

@ENTERPRISE ...some quick thoughts on Curve Optimizer (CO):

 

You might want to wait just a bit because some vendors have started to roll out a new AMD AGESA 1.0.8.0 bios for AM5. FYI, not a good idea to use saved profiles from an older bios (take screenshots of your fav profiles, or write it on paper). I mention this because you want to be finished with the RAM tuning before doing CO....There are folks with the latest bios claiming a one-step increase in FCLK and also a gain in RAM MHz...but I cannot verify myself at this point as I am a bios upgrade 'Luddite'.

 

For both my 5950X and 7950X3D COs, I used Hydra (Yuri), albeit different versions. There are other tools such as Corecycler. You need at least Hydra 1.3 c Pro for the Ryzen 7K, ie. > here. This version uses CinebenchR23 for its per-core CO value generations and it can take over an hour with a 16 core (and lots of rebooting). The per-core CO values Hydra will spit out work for most things, including CinebenchR23, but they are still a tad optimistic when it gets to the really tough tests, such as Aida SHA3 - there might actually be something wrong/weird about the SHA3 test, but I can get it to work with my own 'medium' Hydra setting (which is |5| off each of the per-core parameter recommended by Hydra).

 

Another point to keep in mind before starting the per-core CO: Many upscale boards have a separate 'all-core' CO as a PBO boost in their menu and not necessarily titled CO. My 670E Aorus Master has it, and I run the max 90L5 which according to Skatterbencher is a CO of 50, per this screenshot from his YouTube vid on the matter:

 

...I had 90L5 on in the bios when I did the Hydra CO tests, and it is to some extent cumulative. Anyway, something to keep in mind if your MSI board has something similar...you need to decide whether you leave such enhancements on or off. Speaking of bios enhancements, over-boost in 25 MHz steps and scalar in the PBO menus should also be set, along with 'motherboard values' for PPT, TDC, EDC. This can vary significantly from CPU to CPU, but an overboost of 150 MHz is where most people typically settle. Scalar can but do not have to make a difference; I use a scalar of 7 on mine, and that is where the CPUZ comes in handy.

 

Now, before you do the per-core Hydra, you want to assemble your testing software kit; everyone uses their own favs, but I recommend CinebenchR23 (and R24, but it takes longer), Aida - including the cache benchmark which will max some of your high-speed cores and can crash with an overly optimistic CO and of course the aforementioned SHA3 in Aida. I also use several stress tests in OCCT, and 3DM CPU profile, plus TimeSpy/Ex CPU, and CPUZ single thread and multi-threads, and 7-zip compress/de-compress benchmarks.

 

Before you get going on per-core CO, try a few all-core CO values. If you are running stock bclk (which you should for CO baselines), start with -5 then -10, -15 etc until some of your test suite apps hang or crash. Write the last working all-core CO down somewhere. 

 

Finally, you can do the Hydra 1.3 c Pro per-core CO. As mentioned, it takes a while for a 16-core processor. It will likely show various good and some bad cores, and the latter ones are the culprits behind the all-core CO crashes when pushed too far. On my 7950X3D, even with 90L5, I had a few cores at Hydra -40, but also one at -10 and a couple at -24.

 

Take a screenshot of the final Hydra desktop window displaying the per-core results (the Hydra log .txt file sometimes fails). Now back to your stress test suite...most of those will work fine until you hit Aida SHA3...you can back each core CO value off by a given amount and try again until SHA3 passes - but be aware that some folks just ignore SHA3 and use a per-core CO that works for everything else.

 

Finally, I just saw a neat new app called CPU Affinity Scheduler by nukoseer/cas on GitHub >here  ...there are other ways to look CPU affinity cores (ie. via Process Lasso) but from the demo I saw, it is much easier and shows you the bitmask for all 16 cores (example: core07 000000000000C000). You can then use that app to specify which core a process such as aida64.exe (or other apps) should use. I have not tried this new app myself yet, but the source which showed it usually knows what he's posting about. 

 

One more thing - if you start to use bclk (and especially eclk which my 670E Aorus also offers), your per-core CO values need to be redone; with enough eclk, a -40 CO can become a +2, apart from setting PBO overboost to underboost, such as 150 over to 230 under for eclk 104 on my setup.

 

Good luck with the CO ! 

 

 

Hey @J7SC_Orion 

 

Starting to look into this, quick one from the looks of things you have to have a patron membership to get the Hydra Pro versions right ? I assume yes, but I also cannot get access to the Discord Channel for any access as when you try to join, it is just an empty server lol. Looks like others are have a similar issue.

[neurotix]

2 hours ago, ENTERPRISE said:

 

Hey @J7SC_Orion 

 

Starting to look into this, quick one from the looks of things you have to have a patron membership to get the Hydra Pro versions right ? I assume yes, but I also cannot get access to the Discord Channel for any access as when you try to join, it is just an empty server lol. Looks like others are have a similar issue.

 

"HYDRA 1.3G PRO (major DCFR update) - RELEASED!" by 1usmus from Patreon | Kemono

KEMONO.SU

 

Click the dl link there.

 

You're welcome.

[Hurricane28]

13 minutes ago, neurotix said:

 

"HYDRA 1.3G PRO (major DCFR update) - RELEASED!" by 1usmus from Patreon | Kemono

KEMONO.SU

 

Click the dl link there.

 

You're welcome.

 

Does this guy still live tho? I mean, last time i spoke him on twitter he was in a bad place in Ukraine. I hope he is okay, haven't heard from him for over a year.

[ENTERPRISE]

4 hours ago, neurotix said:

 

"HYDRA 1.3G PRO (major DCFR update) - RELEASED!" by 1usmus from Patreon | Kemono

KEMONO.SU

 

Click the dl link there.

 

You're welcome.

Appreciate it will check it out, but might need the latest version to get the best results possibly as I can see he improved a lot since that release.

4 hours ago, Hurricane28 said:

 

Does this guy still live tho? I mean, last time i spoke him on twitter he was in a bad place in Ukraine. I hope he is okay, haven't heard from him for over a year.

Yeah he is all good, well good might be a stretch but he is alive and still developing.

[Hurricane28]

8 hours ago, ENTERPRISE said:

Appreciate it will check it out, but might need the latest version to get the best results possibly as I can see he improved a lot since that release.

Yeah he is all good, well good might be a stretch but he is alive and still developing.

 

Alright, good to hear. 

[ENTERPRISE]

So I can confirm the latest AGESA 1.0.0.8 has enabled me to step up FCLK to 2133 all stable. Going to try for 2167 on the off chance. 

 

What parameters would make 1:1 MEMCLK/UCLK more possible, if any ? I might scale back to 6400 to see if I can get it working, 6600 it won't work but I think im asking too much at 6600.

[kaliz]

On 23/10/2023 at 09:04, Hurricane28 said:

Je i saw that, amazing. 

 

It won't matter on my chip tho so i don't bother. On single CCX i belive 6400MHz is the max buildzoid said. 

 

Sadly the lower timing didn't work on the 6400 timings. Today i try again. 

 

I did a 8400C34-45-40-42-72-504 with GDM disabled and all tight timings on a Ryzen 5 7500F single CCX for topscore in Pyprime 2b for non X3D models, it really depends on the RAM and mobo. The AM5 CPU's themselves are capable of around DDR5-8800!! With a 7800X3D managed to run DDR5-8720 and thats about the max this 4 DIMM slot mobo can handle! Imagine a GENE!

 

1 hour ago, ENTERPRISE said:

So I can confirm the latest AGESA 1.0.0.8 has enabled me to step up FCLK to 2133 all stable. Going to try for 2167 on the off chance. 

 

What parameters would make 1:1 MEMCLK/UCLK more possible, if any ? I might scale back to 6400 to see if I can get it working, 6600 it won't work but I think im asking too much at 6600.

 

I think 6400 with 2133 fclk in 1:1 mode is best for your setup!. That way it is all synced; 3:1:3 , eg.: 6400:2133:6400 in 1:1 mode. If your CPU is very capable, the next best setup will be 6600 with 2200 fclk, so its again synced for 3:1:3. 

 

So if your CPU is only capable of running 2133 fclk, i wouldnt even go 6600, but stay at 6400 to have that synchronization. You can test it for yourself ofcourse for speeds, but a AMD engineer said on social media last year that this is the way to do it..

 

The next option is to run in 1:2 mode, and to negate the latency penalty, you have to run above DDR5-7600. My preference is DDR5-8000 with either 2000 fclk (for 2:1:1) or 8000 with 2200 fclk for max throughput.

 

you can set VDDG IOD and CCD to 1,050v to runhigher fclk. max will be 1,15v for those 

[J7SC_Orion]

22 hours ago, ENTERPRISE said:

 

Hey @J7SC_Orion 

 

Starting to look into this, quick one from the looks of things you have to have a patron membership to get the Hydra Pro versions right ? I assume yes, but I also cannot get access to the Discord Channel for any access as when you try to join, it is just an empty server lol. Looks like others are have a similar issue.

 

...yeah. the only 'official' working link I have seen when searching yesterday was an even newer version, Hydra 1.5. The one @neurotix linked also appears to be newer than 1.3 C Pro so it should work - just virus-check everything re. sites ! BTW,  Gigabyte / Aorus has not yet released the latest Agesa-based bios for my board - then again, they used to release bug-fixes right after an early release - hopefully, they take their time and get it right the first time with the '0.8' version. 

 

1 hour ago, kaliz said:

 

I did a 8400C34-45-40-42-72-504 with GDM disabled and all tight timings on a Ryzen 5 7500F single CCX for topscore in Pyprime 2b for non X3D models, it really depends on the RAM and mobo. The AM5 CPU's themselves are capable of around DDR5-8800!! With a 7800X3D managed to run DDR5-8720 and thats about the max this 4 DIMM slot mobo can handle! Imagine a GENE!

 

 

I think 6400 with 2133 fclk in 1:1 mode is best for your setup!. That way it is all synced; 3:1:3 , eg.: 6400:2133:6400 in 1:1 mode. If your CPU is very capable, the next best setup will be 6600 with 2200 fclk, so its again synced for 3:1:3. 

 

So if your CPU is only capable of running 2133 fclk, i wouldnt even go 6600, but stay at 6400 to have that synchronization. You can test it for yourself ofcourse for speeds, but a AMD engineer said on social media last year that this is the way to do it..

 

The next option is to run in 1:2 mode, and to negate the latency penalty, you have to run above DDR5-7600. My preference is DDR5-8000 with either 2000 fclk (for 2:1:1) or 8000 with 2200 fclk for max throughput.

 

you can set VDDG IOD and CCD to 1,050v to runhigher fclk. max will be 1,15v for those 

 

FCLK 2200 & DDR5 8000 is doable on my setup but I have it at 2167 (2175) or 219X (w/bclk) for now until the new bios for the board drops. I am also going to try some additional tRRD, tFAW and tWTR again on the weekend per you earlier post.

 

EDIT: mobo won't let me set tFAW below 20

 

 

 

Edited October 28, 2023 by J7SC_Orion

[neurotix]

3 hours ago, J7SC_Orion said:

 

...yeah. the only 'official' working link I have seen when searching yesterday was an even newer version, Hydra 1.5. The one @neurotix linked also appears to be newer than 1.3 C Pro so it should work - just virus-check everything re. sites ! BTW,  Gigabyte / Aorus has not yet released the latest Agesa-based bios for my board - then again, they used to release bug-fixes right after an early release - hopefully, they take their time and get it right the first time with the '0.8' version. 

 

 

FCLK 2200 & DDR5 8000 is doable on my setup but I have it at 2167 (2175) or 219X (w/bclk) for now until the new bios for the board drops. I am also going to try some additional tRRD, tFAW and tWTR again on the weekend per you earlier post.

 

EDIT: mobo won't let me set tFAW below 20

 

 

 

Yeah, the link I posted will still work fine and accomplish the same purpose. It should also be virus-free. I've heard the 1.5 version downloads you can find may have malware.

 

 

Edited October 28, 2023 by neurotix

[neurotix]

My experience with Hydra wasn't so good, I ran the diagnosis and it tested all my cores for CO values, but some were really low like -40 or -50. I tried applying the CO values then moved my mouse and instantly got a bsod. Same thing with doing hybrid oc using the CO values and +200 PBO at the same time. I wonder if the 1.3g version isn't compatible with X3D chips.

 

For some reason I can't get to that website link with the download of 1.3g, it says https is required and it's sending back an invalid cert. I will zip and upload 1.3g after this post. It's 400mb and I can't upload to this site, nor the external host I usually use. It is also 1.3F. Hopefully E got 1.3g from that kemono link already, or maybe its just a problem on my end with my internet atm.

Edited October 28, 2023 by neurotix

[neurotix]

Here is Hydra 1.3F PRO. Download link expires in 14 days.

 

HYDRA 1.3F PRO

WWW.MEDIAFIRE.COM

 

[J7SC_Orion]

Yeah, minimum for Ryzen 7K is 1.3 C Pro - anything later (ie. 1.3 F Pro_ after should work fine. Hydra Pro includes CinebenchR23 which it is why it is quite a large file. Hydra was just a touch optimistic on my system for COs but not by much...Cinebench at al ran just fun with the Hydra-recommended CO values. Aida SHA3 not so much with those values, but then, SHA3 is weird anyways  

[neurotix]

I found a link on a Chinese site that had 1.5 something but clicking the link to download it gave a redirect to Baidu, and no download. I wouldn't trust it.

 

Ultimately, beggers can't be choosers. Unless you want to pony up $10 to 1usmus, use the one I just uploaded.

[ENTERPRISE]

9 hours ago, kaliz said:

 

I did a 8400C34-45-40-42-72-504 with GDM disabled and all tight timings on a Ryzen 5 7500F single CCX for topscore in Pyprime 2b for non X3D models, it really depends on the RAM and mobo. The AM5 CPU's themselves are capable of around DDR5-8800!! With a 7800X3D managed to run DDR5-8720 and thats about the max this 4 DIMM slot mobo can handle! Imagine a GENE!

 

 

I think 6400 with 2133 fclk in 1:1 mode is best for your setup!. That way it is all synced; 3:1:3 , eg.: 6400:2133:6400 in 1:1 mode. If your CPU is very capable, the next best setup will be 6600 with 2200 fclk, so its again synced for 3:1:3. 

 

So if your CPU is only capable of running 2133 fclk, i wouldnt even go 6600, but stay at 6400 to have that synchronization. You can test it for yourself ofcourse for speeds, but a AMD engineer said on social media last year that this is the way to do it..

 

The next option is to run in 1:2 mode, and to negate the latency penalty, you have to run above DDR5-7600. My preference is DDR5-8000 with either 2000 fclk (for 2:1:1) or 8000 with 2200 fclk for max throughput.

 

you can set VDDG IOD and CCD to 1,050v to runhigher fclk. max will be 1,15v for those 

Thanks for your input, super helpful. I have 2133 FCLK stable and will test performance with RAM at 6400 to see how that plays out against performance I get now with 6600. 

 

Currently stability testing 2167 FCLK. I am able to set 2200 and do benchmarks but GSAT was unstable. I did increase SOC voltage to 1.273 but to no avail but may need to revise the VDDG IOD and CCD a little higher to see if that can help get 2200 FCLK stable.