Thank you i ordered the non engraved ones. I just got the copper heatsinks too but i read in a post here that they need to be trimmed,thus why i got those that are easy to be used without trimming...
-
Save 15% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!
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
- Thread starter HydrAxx747
- Start date
Just an update, I ended up running this configuration
I tried 6400 in 1:1 but I didn't manage to stabilise it with the amount of vSOC I was willing to use daily.
I then set extremely loose timings and 2000 FCLK, then tried 2:1 mode and things started getting pretty weird. I found a specific combination of vsoc/vdd/vddq/vddio/vddp that would make it POST at 7800 but impossible to boot windows (I left VPP untouched at 1.8v):
VDD 1.45v
VDDQ 1.35v
VDDIO 1.25v
VDDP 1.15v
VSOC 1.15v
With 4.03.MEM01 BIOS, setting VDD above 1.435v results in VDDQ being ignored and set to 1.44v, so VDDQ was actually 1.44v with those voltages. It was very weird because changing any of the voltages up or down more than 20mv would result in no POST at all. Even setting VDDQ to 1.38v would result in no POST despite it being set to 1.44v by the BIOS.
With these voltages I was able to boot windows @7600 but as soon as I ran any test it immediately failed
VDD 1.40v
VDDQ 1.30v
VDDIO 1.20v
VDDP 1.15v
VSOC 1.20v
It was actually impressive that it managed to boot windows consistently with those voltages but it was unable to run anything (p95, y-cruncher, etc).
I was aware that this configuration was violating the rule of VDDP +100mv < VDDIO, but it was the only configuration that would POST and boot windows. Lowering VDDP to 1.1v would result in no POST, same as increasing VDDIO to 1.25v (to comply with the rule).
I was aware that some boards or memory configurations have specific voltage restrictions at high frequencies, like requiring some delta between some voltages, i.e. 100mv delta between VDD and VDDQ but what I was experiencing was too weird, like not being able to POST if VDDIO is increased by 50mv to 1.25v, but it POSTs again if you also lower VSOC by 50mv to 1.15v, which didn't make sense to me since the only restriction is VSOC having to be lower than VDDIO + 100mv.
Keeping the same voltages and lowering frequency to 7400 made it stable enough to run some p95 tests without errors. At this point I started trying different voltages but things still didn't make sense. Adjusting any voltage up or down would result in not being able to run p95, not being able to boot windows or not being able to even POST. It seemed that I would have to try every possible voltage combination to find the right one, which is too much if all 5 voltages need to be adjusted.
At 7000 with almost the same voltages it was stable enough that I could adjust timings knowing whether they were stable. I was able to push them almost all the way down to the same timings I use at 6200, but still not fully stable probably due to voltages. The timings I ended up with seem to even work at 7200.
I know dual rank is a different beast, but does anyone have any idea of what could be happening? I have the feeling that if I mange to fix whatever is wrong with the voltages and make it stable at 7000 I could also make 7200 or even 7400 work.
I have Nitro Mode enabled with 8x/8x just in case it helps with memory training. I also tried data-eye training (which takes 10 minutes to boot each time) but I didn't notice any difference.
Thank you very much in advance.
It looks like VDDQ at 1.2v helps with stability. I didn't try with such low voltages because everything I tried was pointing to requiring more voltage for signal integrity.
Back to testing I guess..
Last edited:
update me when you find the best ram settings for the 8700g and 6400 impact ram! really trying to max out my igpu right now
Now we're talking
This is with 85W TDP, unlimited CPU frequency and an iGPU Curve Optimizer of -10 (GFX CO -15 is already unstable for me).
Since GFX CO -10 increases iGPU frequency (up to max frequency, 2900MHz in the 8700G) due to it having more power headroom (from undervolting), it starts triggering more often the current limit. For this reason, I maxed both CPU's TDC and EDC so it can stretch its legs at lower voltages (strangely iGPU current limit is in some way limited by those values instead of SOC current limits as Ryzen Master suggests). This shouldn't be a problem, specially since I am not pushing the TDP above CPU's default values.
Undervolting the iGPU has no effect if it always have available power budget (if TDP is high enough). I personally change the TDP based on what I'm playing, since there is usually a point where more power doesn't translate to more performance (those are the shortcuts I have at the bottom bar).
If you plan to undervolt the iGPU, do it after adjusting the memory frequency/timings, since higher iGPU loads affect how much you can undervolt it.
At lower TDPs, you can get more iGPU performance by limiting the CPU frequency (to free power budget for the iGPU), since most of the time you will be iGPU limited. Generally, the lower the FPS the more you can limit the CPU frequency. In 3D Mark Time Spy since the FPS is so low with my 8700G I get the best GPU score by limiting the CPU frequency to its lowest value (2.35GHz) via windows power plan.
I am quite impressed since I am now getting the same GPU Score with 45W as I was getting with 6200 MT/s at 65W.
I am using Dual Rank memory, which behaves differently from Single Rank, therefore your experience will probably different than mine. What I can say is that MEM VDD, MEM VDDQ, CPU VDDIO, CPU VSOC and cLDO VDDP (this last one specially for Dual Rank) plays a big role and it is not obvious how. There must be some intricate relationships between those voltages that affect stability (we already got a hint of this from the available documentation stating that VSOC needs to be lower than VDDIO + 100mv despite coming from different voltage rails).
In my experience, on high frequencies lowering both VDDQ (to 1.2v) and VSOC (as low as it could go) helped with stability, which is something I wasn't expecting. But as I was adjusting VDDQ, adjusting VDDIO also seemed to be necessary, and by extension VDDP since it derives from it. In the end I just tried as many combinations as possible, running simple stress tests (like p95) as guidance to find the most stable combinations (I recommend you to use a spreadsheet for this).
Regarding the timings:
My memory have Hynix A-Die (as most DDR5 memory actually, except 24/48GB dimms which have Hynix M-die, which is almost identical to A-die)
By the way, FFXIV Dawntrail Benchmark seems to be extremely good at detecting unstable memory/infinity fabric (better than Time Spy), probably due to it being very hard on memory in many different ways. Run it with the "Maximum" profile, since other profiles have dynamic FSR based on FPS and you want something that behaves the same all the time.
Now I have to start running all the usual memory test programs (y-cruncher, p95, TM5, etc) to make sure this is stable.
This is with 85W TDP, unlimited CPU frequency and an iGPU Curve Optimizer of -10 (GFX CO -15 is already unstable for me).
Since GFX CO -10 increases iGPU frequency (up to max frequency, 2900MHz in the 8700G) due to it having more power headroom (from undervolting), it starts triggering more often the current limit. For this reason, I maxed both CPU's TDC and EDC so it can stretch its legs at lower voltages (strangely iGPU current limit is in some way limited by those values instead of SOC current limits as Ryzen Master suggests). This shouldn't be a problem, specially since I am not pushing the TDP above CPU's default values.
Undervolting the iGPU has no effect if it always have available power budget (if TDP is high enough). I personally change the TDP based on what I'm playing, since there is usually a point where more power doesn't translate to more performance (those are the shortcuts I have at the bottom bar).
If you plan to undervolt the iGPU, do it after adjusting the memory frequency/timings, since higher iGPU loads affect how much you can undervolt it.
At lower TDPs, you can get more iGPU performance by limiting the CPU frequency (to free power budget for the iGPU), since most of the time you will be iGPU limited. Generally, the lower the FPS the more you can limit the CPU frequency. In 3D Mark Time Spy since the FPS is so low with my 8700G I get the best GPU score by limiting the CPU frequency to its lowest value (2.35GHz) via windows power plan.
I am quite impressed since I am now getting the same GPU Score with 45W as I was getting with 6200 MT/s at 65W.
I am using Dual Rank memory, which behaves differently from Single Rank, therefore your experience will probably different than mine. What I can say is that MEM VDD, MEM VDDQ, CPU VDDIO, CPU VSOC and cLDO VDDP (this last one specially for Dual Rank) plays a big role and it is not obvious how. There must be some intricate relationships between those voltages that affect stability (we already got a hint of this from the available documentation stating that VSOC needs to be lower than VDDIO + 100mv despite coming from different voltage rails).
In my experience, on high frequencies lowering both VDDQ (to 1.2v) and VSOC (as low as it could go) helped with stability, which is something I wasn't expecting. But as I was adjusting VDDQ, adjusting VDDIO also seemed to be necessary, and by extension VDDP since it derives from it. In the end I just tried as many combinations as possible, running simple stress tests (like p95) as guidance to find the most stable combinations (I recommend you to use a spreadsheet for this).
Regarding the timings:
My memory have Hynix A-Die (as most DDR5 memory actually, except 24/48GB dimms which have Hynix M-die, which is almost identical to A-die)
- Some people recommend you to start with the primaries (tCL, tRCD and tRP) but I didn't do that because those are the most affected by MEM VDD (memory voltage) and I'm still deciding which memory voltage I want to run daily.
- tRAS I just set to the conventional tRCD (row activation time) + tRTP (read to precharge)
- tRC should always be tRAS + tRP
- tRRD(L/S) and tFAW limits the amount of row activations in a given timeframe (this is supposed to help with internal voltage drops). I have them set at the minimum possible value, although tRRDS is not supposed to give any benefits below 8, so I should probably leave it at that (and adjust tFAW accordingly to 4x tRRDS = 32).
- tRTP is not supposed to go lower than 12 so I have it maxed, although you can set it lower to the point of instability for some reason
- tWR (delay after write) can't go lower than 48 and from that up in multiples of 6.
- tWTR(L/S) (write to read delay) are maxed out too at 16/4 respectively.
- tRDWR and tWRRD which are related to delays between read-to-write and vice versa are also maxed at 16/1
- Then you have tRDRDscl and tWRWRscl timings which are extremely important for bandwidth. This you want to have them as low as possible and vary a lot based on memory frequency. tRDRDscl (read-to-read) at 4 is probably as low as it can go. tWRWRscl value depends on tRDRDscl and is probably ignored by the memory controller since no one is setting it correctly (you can test this by setting it to 1 and it works). The actual value is tWRWRscl = (tRDRDscl + 7) x 2 - 7 since those timings actually come from another timing that is hidden (CCDL) and that's the one you're actually setting (tRDRDscl = CCDL - 7; tWRWRscl = CCDLWR - 7; CCDLWR = CCDL * 2)
- tRDRDSC and tWRWRSC should always be 1, otherwise you'll start adding unnecessary gaps between read/write bursts
- tRDRD(SD/DD) and tWRWR(SD/DD) are only relevant in multiple rank (SD = same channel, different rank) and multiple dimms per channel (i.e. 4 dimms) (DD = same channel, different dimm) configurations. If that's what you have, start with a high values and lower it slowly testing thoroughly for stability.
- And finally the refresh timings: tREFI (how often a refresh happens) and tRFC (refresh duration). tRFC2 and tRFCsb are not used.
- It seems that this BIOS version doesn't allow you to set tRFC to something different than tRFC2, so set both to the same value.
- I was able to run close to 130ns refresh duration at low speeds, but it seems that at high speeds I need above 145ns (probably due to heat, since refresh is very sensitive to memory temps). For this reason I recommend you to adjust this timing last once the memory is fully stressed.
By the way, FFXIV Dawntrail Benchmark seems to be extremely good at detecting unstable memory/infinity fabric (better than Time Spy), probably due to it being very hard on memory in many different ways. Run it with the "Maximum" profile, since other profiles have dynamic FSR based on FPS and you want something that behaves the same all the time.
Now I have to start running all the usual memory test programs (y-cruncher, p95, TM5, etc) to make sure this is stable.
Last edited:
wow you know a lot. this was really informative, thanks! before i start testing my ram myself, i was wondering if you know what the best ram would be. i thought it would be the crucial 6400 mhz kit because it has the fastest listed speed, but it looks like i could be wrong?Now we're talking
This is with 85W TDP, unlimited CPU frequency and an iGPU Curve Optimizer of -10 (GFX CO -15 is already unstable for me).
Since GFX CO -10 increases iGPU frequency (up to max frequency, 2900MHz in the 8700G) due to it having more power headroom (from undervolting), it starts triggering more often the current limit. For this reason, I maxed both CPU's TDC and EDC so it can stretch its legs at lower voltages (strangely iGPU current limit is in some way limited by those values instead of SOC current limits as Ryzen Master suggests). This shouldn't be a problem, specially since I am not pushing the TDP above CPU's default values.
Undervolting the iGPU has no effect if it always have available power budget (if TDP is high enough). I personally change the TDP based on what I'm playing, since there is usually a point where more power doesn't translate to more performance (those are the shortcuts I have at the bottom bar).
If you plan to undervolt the iGPU, do it after adjusting the memory frequency/timings, since higher iGPU loads affect how much you can undervolt it.
At lower TDPs, you can get more iGPU performance by limiting the CPU frequency (to free power budget for the iGPU), since most of the time you will be iGPU limited. Generally, the lower the FPS the more you can limit the CPU frequency. In 3D Mark Time Spy since the FPS is so low with my 8700G I get the best GPU score by limiting the CPU frequency to its lowest value (2.35GHz) via windows power plan.
I am quite impressed since I am now getting the same GPU Score with 45W as I was getting with 6200 MT/s at 65W.
I am using Dual Rank memory, which behaves differently from Single Rank, therefore your experience will probably different than mine. What I can say is that MEM VDD, MEM VDDQ, CPU VDDIO, CPU VSOC and cLDO VDDP (this last one specially for Dual Rank) plays a big role and it is not obvious how. There must be some intricate relationships between those voltages that affect stability (we already got a hint of this from the available documentation stating that VSOC needs to be lower than VDDIO + 100mv despite coming from different voltage rails).
In my experience, on high frequencies lowering both VDDQ (to 1.2v) and VSOC (as low as it could go) helped with stability, which is something I wasn't expecting. But as I was adjusting VDDQ, adjusting VDDIO also seemed to be necessary, and by extension VDDP since it derives from it. In the end I just tried as many combinations as possible, running simple stress tests (like p95) as guidance to find the most stable combinations (I recommend you to use a spreadsheet for this).
Regarding the timings:
My memory have Hynix A-Die (as most DDR5 memory actually, except 24/48GB dimms which have Hynix M-die, which is almost identical to A-die)
- Some people recommend you to start with the primaries (tCL, tRCD and tRP) but I didn't do that because those are the most affected by MEM VDD (memory voltage) and I'm still deciding which memory voltage I want to run daily.
- tRAS I just set to the conventional tRCD (row activation time) + tRTP (read to precharge)
- tRC should always be tRAS + tRP
- tRRD(L/S) and tFAW limits the amount of row activations in a given timeframe (this is supposed to help with internal voltage drops). I have them set at the minimum possible value, although tRRDS is not supposed to give any benefits below 8, so I should probably leave it at that (and adjust tFAW accordingly to 4x tRRDS = 32).
- tRTP is not supposed to go lower than 12 so I have it maxed, although you can set it lower to the point of instability for some reason
- tWR (delay after write) can't go lower than 48 and from that up in multiples of 6.
- tWTR(L/S) (write to read delay) are maxed out too at 16/4 respectively.
- tRDWR and tWRRD which are related to delays between read-to-write and vice versa are also maxed at 16/1
- Then you have tRDRDscl and tWRWRscl timings which are extremely important for bandwidth. This you want to have them as low as possible and vary a lot based on memory frequency. tRDRDscl (read-to-read) at 4 is probably as low as it can go. tWRWRscl value depends on tRDRDscl and is probably ignored by the memory controller since no one is setting it correctly (you can test this by setting it to 1 and it works). The actual value is tWRWRscl = (tRDRDscl + 7) x 2 - 7 since those timings actually come from another timing that is hidden (CCDL) and that's the one you're actually setting (tRDRDscl = CCDL - 7; tWRWRscl = CCDLWR - 7; CCDLWR = CCDL * 2)
- tRDRDSC and tWRWRSC should always be 1, otherwise you'll start adding unnecessary gaps between read/write bursts
- tRDRD(SD/DD) and tWRWR(SD/DD) are only relevant in multiple rank (SD = same channel, different rank) and multiple dimms per channel (i.e. 4 dimms) (DD = same channel, different dimm) configurations. If that's what you have, start with a high values and lower it slowly testing thoroughly for stability.
- And finally the refresh timings: tREFI (how often a refresh happens) and tRFC (refresh duration). tRFC2 and tRFCsb are not used.
- It seems that this BIOS version doesn't allow you to set tRFC to something different than tRFC2, so set both at the same value.
- I was able to run close to 130ns refresh duration at low speeds, but it seems that at high speeds I need above 145ns (probably due to heat, since refresh is very sensitive to memory temps). For this reason I recommend you to adjust this timing last once the memory is fully stressed.
By the way, FFXIV Dawntrail Benchmark seems to be extremely good at detecting unstable memory/infinity fabric (better than Time Spy), probably due to it being very hard on memory in many different ways. Run it with the "Maximum" profile, since other profiles have dynamic FSR based on FPS and you want something that behaves the same all the time.
Now I have to start running all the usual memory test programs (y-cruncher, p95, TM5, etc) to make sure this is stable.
I'm still learning this stuff, so take it all I said with a grain of salt.
There are three companies that manufactures memory chips: Samsung, Micron and Hynix. Currently Hynix is the one that manufactures the best DDR5 memory chips (by a big margin), so as long as whatever memory you buy has Hynix chips in it, you're good.
Crucial is the retail company of Micron, so they only sell memory dimms with Micron memory chips.
Anything high end currently has Hynix memory. To know if something is high end you need to not only look at the frequency but also at the CAS Latency (CL) (up to a point, beyond certain frequency I think all is Hynix memory). I.e. DDR5 6400 MT/s CL32 is Hynix memory (as far as I know), but we don't have those in SO-DIMM.
I was lucky and my 2x32GB Kingston Impact Fury 5600 CL40 (which I purchased recently) has Hynix memory, and as you can see I was able to push it to 7600 MT/s.
Even if different memory DIMMs have the same memory chips, the ones being sold with higher frequencies/better latencies are supposed to have better binned memory chips, but this is not always the case. The memory chip market is somewhat complex and sometimes companies manages to buy good memory chips at low prices just because the demand was low that day and they were sold at floor price.
I think the best approach is to just look at what people reports, i.e. look for a specific model number on Google and find which memory chips people says it has, ideally not too long ago, since it may have changed.
Last edited:
okay awesome. looks like mine is hynix-a. i’ll mess with the timings then when i get time and get back to you. one more thing, in the bios the igpu can only be overclocked to 3200 before it limits you, can it be overclocked further? and can i use the stock power supply or do i need the 180 watt
I don't have experience running the iGPU above the max stock frequency of 2900MHz because I like efficiency and it goes down the drain quite fast at high power consumptions.
I don't recommend running a PSU above 70% of its rated power output, so anything above 84W is already too much, and I'd expect the CPU to consume a lot of power if you plan to run the iGPU at or close to 3200 MHz, so yes, a 180W PSU or higher would probably be required.
I recommend you to read this article if you want to push the iGPU to those limits, it goes quite deep into this subject
SkatterBencher #70: AMD Radeon 780M Overclocked to 3150 MHz - SkatterBencher
We overclock the AMD Radeon 780M integrated into the Ryzen 7 8700G up to 310 MHz with the ASUS ROG Strix X670E-I motherboard.
skatterbencher.com
I think Skatterbencher is focused on CPU/GPU overclocking, so his memory OC is not very strong but his skills and knowledge to OC the iGPU are very good
Last edited:
Did you oc the memory and if you did by how much?
New Test BIOS "4.04" is avalaible in upload folder(info & link in first post)
Do we know the differences between 4.04 and 4.03.MEM01?
Edit: Nevermind, I just read your post on the deskmini x600 thread stating that the only change is the new agesa
Last edited:
I don't feel a difference between 4.06 and 4.03.MEM01 regarding memory OC with the 8700G.
I'm now trying to stabilise 8200 MT/s on my dual rank (64GB) which I find unbelievable. Hopefully I'll be able to get something stable soon so I can share my experience with you all.
Guys i just got my X600 and i migrated the SSD from my X300.In device manager im missing some drivers.....Anyone knows what i can do?
X300 had a DVD with everything,but the X600 does not...
You can see in the picture.
X300 had a DVD with everything,but the X600 does not...
You can see in the picture.
Yet it is one of the most important (and well-known) things to install when you have a computer based on an AMD CPU/Motherboard: the Chipset drivers!Guys i just got my X600 and i migrated the SSD from my X300.In device manager im missing some drivers.....Anyone knows what i can do?
X300 had a DVD with everything,but the X600 does not...
You can see in the picture.
You can download the latest AMD Chipset drivers here:
I have them but the problem is that they do not install....they keep saying that they can only be installed on amd hardware..
Try the version of AMD Chipset drivers from ASRock's official DeskMini X600 support, the version is more recent maybe your problem will be solved, the link is here: https://download.asrock.com/Drivers/AMD/CPU/Chipset(v6.10.22.027).zip
And also the AMD drivers for the "NPU" part if you ever use a Ryzen 5 8600G or 7 8700G: https://download.asrock.com/Drivers/AMD/CPU/AMD_NPU(v10.1109.0008.128).zip
Nothing....it seems that vbscript is uninstalled and its needed for amd drivers...and from what it seems i cannot install it it keeps failing...Im searching for hours for a solution but nothing yet.I just need the drivers by themselves without the installation,this way i can add them....
EDIT: I fixed it guys,i transferred the drivers from another computer and it worked fine,vbscript is still an issue but im in DEV channel so it might be fixed if im lucky lol
Last edited:
Guys i got a 2x8GB kit for now since im out of budget,and i tried OC a bit.The ram is Hynix 5600 A die.I oc'ed it to 6400 but the BW is kind of low at around 60-65gb both read and write,latency is around 63ns,do you have any suggestions what do i need to improve in order to make it better?
PS:I got the ram with my remain 18e that left after i bought the mini and CPU lol ,so for that price i think it was a good buy?
PS:I got the ram with my remain 18e that left after i bought the mini and CPU lol ,so for that price i think it was a good buy?
Last edited:
You're running in 1:2 mode (memory controller's clock is half of the memory clock; uclk=mclk/2). That has a big penalty. If you're interested in running around 6200-6400 MT/s you should be aiming for 1:1 mode.Guys i got a 2x8GB kit for now since im out of budget,and i tried OC a bit.The ram is Hynix 5600 A die.I oc'ed it to 6400 but the BW is kind of low at around 60-65gb both read and write,latency is around 63ns,do you have any suggestions what do i need to improve in order to make it better?
PS:I got the ram with my remain 18e that left after i bought the mini and CPU lol ,so for that price i think it was a good buy?
All CPUs should be able. to run 1:1 mode at 6000, and the majority should be able to do 6200 too. Not all CPUs can do 6400 or above in 1:1 mode. I would try 6200 in 1:1 mode and see how much SOC voltage you need to have it stable. If it is below 1.2v there is a chance that you could do 6400 below 1.3v (not recommended to use more than 1.3v SOC).
Regarding your timings, I'd expect you to be able to set tRTP to 12 and tWR to 48, and probably tRDRDscl to 4 and tWTR(S/L) to 4/16
Similar threads
