CPU [13/12/2024]ASRock DeskMini X600 BIOS "4.04" & "4.06" & "4.08" - with new AMD AGESA ComboAM5Pi v."1.2.0.2a" & DDR5 SO-DIMM Voltage Settings available

[nirvana]

This is the 7800MT that worked with everything and Ff demo just fine,except 3dmark....If you guys have any suggestion what else to change to improve it please let me know

Screenshot

Captured with Lightshot

prnt.sc

The error below its not from the oc,the program itself got corrupted.I leave it here in case its useful to someone else...

Screenshot

Captured with Lightshot

prnt.sc

I'm traveling for Christmas, so I won't be able to post too frequently. Apologies for that.

I was not able to run 2400 FCLK stable with my 8700G, not even 2266, and I get more performance with 1950 FCLK at 7800 anyway probably due to be running in sync UCLK, so I would try that (you don't have 1950 FCLK option in the OC Tweak menu, so leave it on Auto and set 1950 in Advanced - AMD overclocking - PBO menu - Infinity Fabric; you also have Nitro settings here).

Your timings look good.

Some people need to increase VDDP at those speeds (I definitely do with Dual Rank memory), so if you still have issues try with 0.9v or even 0.95v. I read from Veii that it is safe up to 1.3v, but people usually stops at 1.15v.

Some people need to have a delta between VDDQ and VDDIO (VDDIO around 50-100mv less than VDDQ).

At those speeds you may need to start messing with the impedances, ASRock memory training doesn't test all values, check what I had to modify in case you also need to change any of those values to something that is different from the auto training.

I would also enable DFE 2D training and Nitro 8x/8x, it is supposed to help with training at high speeds at expense of longer boot times.

I would try running MT5 for for 25 cycles several times (should be less than 1.5 hours for 16GB) and follow the spreadsheet in case you encounter any error in one of the tests. You can find both the spreadsheet and the program in this post:

https://www.overclock.net/threads/amd-ddr5-oc-and-24-7-daily-memory-stability-thread.1800926/post-29363233

Leaving FFXIV Dawntrail Benchmark running in loop overnight is also a very good way to test stability

 

[woofaki]

I'm traveling for Christmas, so I won't be able to post too frequently. Apologies for that.

I was not able to run 2400 FCLK stable with my 8700G, not even 2266, and I get more performance with 1950 FCLK at 7800 anyway probably due to be running in sync UCLK, so I would try that (you don't have 1950 FCLK option in the OC Tweak menu, so leave it on Auto and set 1950 in Advanced - AMD overclocking - PBO menu - Infinity Fabric; you also have Nitro settings here).

Your timings look good.

Some people need to increase VDDP at those speeds (I definitely do with Dual Rank memory), so if you still have issues try with 0.9v or even 0.95v. I read from Veii that it is safe up to 1.3v, but people usually stops at 1.15v.

Some people need to have a delta between VDDQ and VDDIO (VDDIO around 50-100mv less than VDDQ).

At those speeds you may need to start messing with the impedances, ASRock memory training doesn't test all values, check what I had to modify in case you also need to change any of those values to something that is different from the auto training.

I would also enable DFE 2D training and Nitro 8x/8x, it is supposed to help with training at high speeds at expense of longer boot times.

I would try running MT5 for for 25 cycles several times (should be less than 1.5 hours for 16GB) and follow the spreadsheet in case you encounter any error in one of the tests. You can find both the spreadsheet and the program in this post:

https://www.overclock.net/threads/amd-ddr5-oc-and-24-7-daily-memory-stability-thread.1800926/post-29363233

Leaving FFXIV Dawntrail Benchmark running in loop overnight is also a very good way to test stability

Thank you my friend!Its frustrating....The one i posted before that passed my test now keeps failing...I was happy for a bit lol

PS: Did you tried AIDa64 system stability test?This is the one that keep failing on me most..when all 3 Fpu/System cache/system memory are checked.Im curious to see if it does it in yours also.I did the memory test by itself and it did failed there after a while.My 6400MT profile seem to be fine there.

 

[nirvana]

Personally I find TM5, y-cruncher (FFT+FFTv4+N63+VT3) and FFXIV to be enough to test for stability, 12h+ each. Karhu is also very good but it is not free software.

First I start with loose timings to find a stable clock by adjusting voltages and impedances (if required). Then I start tightening the timings in small groups or even 1 by 1 for those that may be problematic (like tRDRDscl, tRTP, tWR, tRRDs/tFAW; and *SD/*DD for Dual Rank only). I am usually quite conservative with the primary timings (tCL, tRCD, tRP) since those are the ones that require the most MEM VDD and are not really that important.

Edit: I want to clarify that I find FFXIV benchmark to be good for testing memory OC stability due to us using our iGPUs.

 

[woofaki]

Personally I find TM5, y-cruncher (FFT+FFTv4+N63+VT3) and FFXIV to be enough to test for stability, 12h+ each. Karhu is also very good but it is not free software.

First I start with loose timings to find a stable clock by adjusting voltages and impedances (if required). Then I start tightening the timings in small groups or even 1 by 1 for those that may be problematic (like tRDRDscl, tRTP, tWR, tRRDs/tFAW and *SD/*DD for Dual Rank only). I am usually quite conservative with the primary timings (tCL, tRCD, tRP) since those are the ones that require the most MEM VDD and are not really that important.

Did you read any DDR5 guide of what timings do and what are the important ones etc or you learned by trial and error? I did read a DDR4 guide before but it seems that DDR5 has differences...

Im trying to max my 6400MT profile,while working with 7800MT too...I think that is already close to max,not sure what else i can do to improve it further.Mine is Single rank,i just saw your post in page 4 that explains some things about DDR5,if you learn more things you can add to it,it will be a great help for us all

Screenshot

Captured with Lightshot

prnt.sc

 

[nirvana]

Did you read any DDR5 guide of what timings do and what are the important ones etc or you learned by trial and error? I did read a DDR4 guide before but it seems that DDR5 has differences...

Im trying to max my 6400MT profile,while working with 7800MT too...I think that is already close to max,not sure what else i can do to improve it further.Mine is Single rank,i just saw your post in page 4 that explains some things about DDR5,if you learn more things you can add to it,it will be a great help for us all

Screenshot

Captured with Lightshot

prnt.sc

I haven't found a single place or guide where everything is explained unfortunately. I have been learning from different forums, discord groups, videos and web papers/articles.

I don't remember all of them, but in this overclock.net thread there are thousand of pages where people share their DDR5 OC experiences on AMD platforms, you can learn a lot just by reading it

https://www.overclock.net/threads/amd-ddr5-oc-and-24-7-daily-memory-stability-thread.1800926/

These buildzoid videos (playlist) explais the basics very well for DDR4, and most of it translates to DDR5.

The main difference between DDR4 and DDR5 is the addition of more bankgroups, and since (rougly) you have to wait a lot of time between actions to the same bankgroup, having access to more bankgroups to execute actions on while you wait greatly improves performance (up to a point).

Regarding your timings, you're pretty much maxed out. Other users with Hynix A-Die managed to lower tRTP to 12 and tWTR_S to 4, so I would give that a try if you haven't yet. I would also check if you can lower the primaries, you may be able to lower tCL and maybe tRCD and tRP too, but not sure by how much.

I would also try to run in 1:1 mode (UCLK=MCLK) since AMD is supposed to benefit substantially with increased memory controller clock (you probably need close to 1.3v SOC for that, but never exceed 1.3v). If that is unstable (quite likely) try with 6200 MT/s in 1:1 mode and compare.

If you decide to leave it in 2:1 mode (UCLK=MCLK/2), you probably don't need more than 1.1v SOC, even at 7800 MT/s.

This is my 6200 MT/s profile using 1.3v MEM VDD:

 

[woofaki]

I haven't found a single place or guide where everything is explained unfortunately. I have been learning from different forums, discord groups, videos and web papers/articles.

I don't remember all of them, but in this overclock.net thread there are thousand of pages where people share their DDR5 OC experiences on AMD platforms, you can learn a lot just by reading it

https://www.overclock.net/threads/amd-ddr5-oc-and-24-7-daily-memory-stability-thread.1800926/

These buildzoid videos (playlist) explais the basics very well for DDR4, and most of it translates to DDR5.

The main difference between DDR4 and DDR5 is the addition of more bankgroups, and since (rougly) you have to wait a lot of time between actions to the same bankgroup, having access to more bankgroups to execute actions on while you wait greatly improves performance (up to a point).

Regarding your timings, you're pretty much maxed out. Other users with Hynix A-Die managed to lower tRTP to 12 and tWTR_S to 4, so I would give that a try if you haven't yet. I would also check if you can lower the primaries, you may be able to lower tCL and maybe tRCD and tRP too, but not sure by how much.

I would also try to run in 1:1 mode (UCLK=MCLK) since AMD is supposed to benefit substantially with increased memory controller clock (you probably need close to 1.3v SOC for that, but never exceed 1.3v). If that is unstable (quite likely) try with 6200 MT/s in 1:1 mode and compare.

If you decide to leave it in 2:1 mode (UCLK=MCLK/2), you probably don't need more than 1.1v SOC, even at 7800 MT/s.

This is my 6200 MT/s profile using 1.3v MEM VDD:

Thanks man!I remembered i tried 6200MT in 1:1 when you told me sometime ago,and if i remember right it wasn't faster.Ill give it another try now tho and see.

PS: Gave it try and its slower by a big margin,to be able to get it stable too i will need to dial back timings,and lower fclk (lower than yours to boot up) I did dial back my 6400MT from 1.2 to 1.1 like you said,and i will do stability test now to see if it works good for me.

PS:1 Do you know what settings affect the copy Bandwidth,the main difference between my 6400MT and 7800MT is the copy that is around 12k more.From what ive read copy is very important for 3D performance.If i remember right you have thermalight cooler?I would like to know what your temps are with it when playing a game or stress testing etc.I have the one that comes with the cpu but with an industrial fan that is very powerful but super noisy...and even then i get like 78c when gaming.

https://prnt.sc/9SmodchlZsMq

 

[nirvana]

Thanks man!I remembered i tried 6200MT in 1:1 when you told me sometime ago,and if i remember right it wasn't faster.Ill give it another try now tho and see.

PS: Gave it try and its slower by a big margin,to be able to get it stable too i will need to dial back timings,and lower fclk (lower than yours to boot up) I did dial back my 6400MT from 1.2 to 1.1 like you said,and i will do stability test now to see if it works good for me.

PS:1 Do you know what settings affect the copy Bandwidth,the main difference between my 6400MT and 7800MT is the copy that is around 12k more.From what ive read copy is very important for 3D performance.If i remember right you have thermalight cooler?I would like to know what your temps are with it when playing a game or stress testing etc.I have the one that comes with the cpu but with an industrial fan that is very powerful but super noisy...and even then i get like 78c when gaming.

https://prnt.sc/9SmodchlZsMq

As far as I know you shouldn't need to loose timings to go from 2:1 to 1:1. The only difference is the memory controller going from 1550 MHz to 3100 Mhz (for 6200 MT/s), which is only influenced by your memory controller's ability to run at high frequencies, and this is directly associated to vSOC (memory controller's voltage). I have the feeling that by loosing your memory timings you are lowering the load on the memory controller and that makes it more stable at that frequency and voltage. My guess is that you need more vSOC for 6200 MT/s in 1:1 but since it is not safe to run more than 1.3v SOC, whatever you can run below that voltage is your CPU limit in 1:1. Each memory controller is different and it is possible that yours need a lot of voltage to run at 3100 MHz+.

To read/write someting from/to memory, the minimum steps required are:
1. Activate a Row
2. Read/write
3. Close the row (precharge)

All timings that affect those basic steps affect the memory latency (I'm talking about real memory latency, not what AIDA reports), like tCL, tRCD, tRTP/tWR, tRP and by extension tRAS and tRC. There are other timings that also affect latency but to a lesser extent.

Then you have other set of timings that directly affect bandwith, which are the ones that control the ability to consecutively read or write. This is mainly tRDRDscl and tWRWRscl (also tRDRDSC and tWRWRSC but those should always be 1 to avoid unnecessary gaps between burst loads/writes).

Since each memory load is different and different timings affect different types of load, I always check how my changes affect performance in things that are as close as real-world scenarios as possible (this is why I mainly use game benchmarks to assest the impact of my changes). For this reason I don't pay attention to numbers produced by latency or bandwith benchmarks.

I have both Noctua NH-L12Sx77 and Noctua NH-C14S (which have the same price where I live) and I use them without the case (open-air is much much better than inside the case). You can read my experience with them here: https://smallformfactor.net/forum/t...ni-–-finally-–-again.19619/page-6#post-289045
here https://smallformfactor.net/forum/t...ni-–-finally-–-again.19619/page-6#post-289110 and here https://smallformfactor.net/forum/t...ni-–-finally-–-again.19619/page-9#post-290101

Both are quite expensive coolers, but at least I can achieve complete silent operation even at maximum loads (85W) with great temps both in the CPU and memory (which is extemely important to me, I really need zero noise), around 60C on the CPU and memory (even with mem at 7800 MT/s, but they have big heatsinks). Noctua NH-C14S is somewhat better than NH-L12Sx77 but also much bigger.

 

[woofaki]

As far as I know you shouldn't need to loose timings to go from 2:1 to 1:1. The only difference is the memory controller going from 1550 MHz to 3100 Mhz (for 6200 MT/s), which is only influenced by your memory controller's ability to run at high frequencies, and this is directly associated to vSOC (memory controller's voltage). I have the feeling that by loosing your memory timings you are lowering the load on the memory controller and that makes it more stable at that frequency and voltage. My guess is that you need more vSOC for 6200 MT/s in 1:1 but since it is not safe to run more than 1.3v SOC, whatever you can run below that voltage is your CPU limit in 1:1. Each memory controller is different and it is possible that yours need a lot of voltage to run at 3100 MHz+.

To read/write someting from/to memory, the minimum steps required are:
1. Activate a Row
2. Read/write
3. Close the row (precharge)

All timings that affect those basic steps affect the memory latency (I'm talking about real memory latency, not what AIDA reports), like tCL, tRCD, tRTP/tWR, tRP and by extension tRAS and tRC. There are other timings that also affect latency but to a lesser extent.

Then you have other set of timings that directly affect bandwith, which are the ones that control the ability to consecutively read or write. This is mainly tRDRDscl and tWRWRscl (also tRDRDSC and tWRWRSC but those should always be 1 to avoid unnecessary gaps between burst loads/writes).

Since each memory load is different and different timings affect different types of load, I always check how my changes affect performance in things that are as close as real-world scenarios as possible (this is why I mainly use game benchmarks to assest the impact of my changes). For this reason I don't pay attention to numbers produced by latency or bandwith benchmarks.

I have both Noctua NH-L12Sx77 and Noctua NH-C14S (which have the same price where I live) and I use them without the case (open-air is much much better than inside the case). You can read my experience with them here: https://smallformfactor.net/forum/threads/asrock’s-x600-deskmini-–-finally-–-again.19619/page-6#post-289045
here https://smallformfactor.net/forum/threads/asrock’s-x600-deskmini-–-finally-–-again.19619/page-6#post-289110 and here https://smallformfactor.net/forum/threads/asrock’s-x600-deskmini-–-finally-–-again.19619/page-9#post-290101

Both are quite expensive coolers, but at least I can achieve complete silent operation even at maximum loads (85W) with great temps both in the CPU and memory (which is extemely important to me, I really need zero noise), around 60C on the CPU and memory (even with mem at 7800 MT/s, but they have big heatsinks). Noctua NH-C14S is somewhat better than NH-L12Sx77 but also much bigger.

Open case is not an option for me...I will only do that if temps are not manageable with the cooler,and with what i have now they wont be for sure when summer hits...So i was was looking for a better solution but with a closed case.Seems that the best so far is thermalight full cooper.

Edit:I tried loosen up the timings for 7800MT by a lot.... and i still getting errors.Im starting to think that its not the timings but the heat on the mem modules...I used to have a small but noisy fan on them before that i removed.I remember having similar issues with x300 mini when i was OC the memory to 4000MT.When i had it open,in order to be able to reset it when it stuck,the memory that was not cooled by the fan inside was keep producing errors...So i will stop atm until i get better cooling that will hopefully fix my issue.

 

[woofaki]

Guys it seem that i got some bad news.....After my last attempt to tighten timings for 6400MT even more,the mini as per usual didn't boot.Turned it off open it to clear cmos like what i was doing all the time now...But even after i touched the cmos pins,it continues not to boot.I see the fan spinning the blue light turns on,but nothing else is happening.No other light in the mobo turns on (if i remember right there is a red light near the SSD...I wanted to try and take the battery out to force clear cmos.But it was not so easy and didn't do it in order not to risk damage.I took it to an electronic shop near me and told him to remove the battery too,but he wont see it until Saturday....Hopefully its not a hardware failure.


PS:A friend told me it might be a corrupt bios...because i was using clear cmos too much,said happened to him in another computer,i hope it is.I will try to re flash it tomorrow and see....Wish me luck!!

 

[Stoney]

luckily the deskmini x600 supports flashback, even if its just the biosubu method. You should be able to recover your deskmini.

 

[woofaki]

luckily the deskmini x600 supports flashback, even if its just the biosubu method. You should be able to recover your deskmini.

Yes thank god,i just hope that this is the issue and not a hardware one.I did protect the APU tho i set it at -40 75c max,so i don't believe its the APU...Memory tho i have no means to know what temps it was running..And i just hope its not the mini itself,that would be a journey...I will let you guys know tomorrow after i try the bios flash.

 

[nirvana]

That sucks.

BIOS corruption is rare but can happen with unstable memory. I wouldn't be worried about hardware damage.

I hope you manage to flash the BIOS and recover your deskmini.

 

[woofaki]

That sucks.

BIOS corruption is rare but can happen with unstable memory. I wouldn't be worried about hardware damage.

I hope you manage to flash the BIOS and recover your deskmini.

I have the usb ready from the latest bios that i flashed like 5 days ago....I will go and take it back from the shop tomorrow early morning and hope for the best...if it doesnt work then i will get it back for them to check on it.I just hope that ill get lucky and it will be the bios...btw is it possible to flash any bios on it,even earlier ones?

 

[nirvana]

I have the usb ready from the latest bios that i flashed like 5 days ago....I will go and take it back from the shop tomorrow early morning and hope for the best...if it doesnt work then i will get it back for them to check on it.I just hope that ill get lucky and it will be the bios...btw is it possible to flash any bios on it,even earlier ones?

I can flash any BIOS using the regular method, so I guess it works the same way regardless the flashing method.

 

[woofaki]

I can flash any BIOS using the regular method, so I guess it works the same way regardless the flashing method.

I did a reflash and sadly it wasn't the cause....So my next guess is that memory gave up,or so i hope and its not the CPU that did.I will change the ram sticks soon and if they arent the cause them the only one remain is the CPU....I will get the ram stick today later at night.Please wish me luck i need them to be faulty....Or else im scr....And will take time to rma the rest.

Edit: Sadly its not the ram....i got new and still the same.I took out the cpu and checked it and i see nothing wrong or burn marks or whatever...Im inclined to say that the mini mobo just crapped out. I did nothing different with this one that i didn't do with x300...Now i need to get it to the shop Monday and wait for more than a month..if mini is the problem.

 

[HydrAxx747]

Well... Vodka was stronger. Here i have some numbers now:
-Deskmini X600 (in a custom case)
-Ryzen 8700G with Noctua NH-L9X65 Heatsink + Noctua NF-A12X25 fan.
-Kingston Fury Impact SO-DIMM DDR5 @6400MT/S
-1TB Samsung 980PRO SSD

Bios: 4.03-MEM1 (XMP + Performance mode enabled):
-3dMark TimeSpy: 3554pts

Bios: 4.08 official (7000mt/s + subtimings + GPUOC@3100mhz) No Performace option in a bios!!
-3dMark TimeSpy: 3152pts

Link to that comparison: Result

That Performance Option description says you need 180W PSU. Both thest were run on Innergie 180W PSU.

That Performance Option Enable/Disable has some serious impact on performance.
Next i will try to OC that 4.03-MEM1 bios and see what it can do. 4.08 is no go...

Hello friend, last night I told my contact at ASRock (Europe) about the drastic performance drop problem you are experiencing in the final BIOS "4.08", I used the result values you were getting under 3DMark TimeSpy to tell him about a concrete example (not having had time to test the BIOS "4.08" myself yet), the problem is most certainly linked to the removal of the "Performance Mode" option in this BIOS as you say (this is also my opinion) and so I told him about it too.

I took the opportunity to ask him if he had the possibility of having a test BIOS (or modified) that is equivalent to or later than the final BIOS "4.08" and that would give back access to the "Performance Mode" in the BIOS settings.

And so he replied to me this afternoon that he would inquire (and inform) the BIOS department of ASRock and try to obtain another BIOS meeting my request as much as possible, and he will come back to me as soon as he has had a response from them.

 

[woofaki]

A small update about the mini,the CPU is dead.Not sure how it died honestly...I did OC it but the temps were at worst 75c.
I will also take the mini itself back to the shop to be tested,around 6 jan just to be sure.So im definitely at least a month+away if its the CPU only....