Prebuilt [SFFn] ASRock's DeskMini A300 - Finally!

[NateDawg72]

I don't think improved cooling will help the situation much, unless you are custom water cooling the CPU + VRM or going below ambient. I'm not running one of these systems so I have no hands on experience with it, but everything I've heard sounds like a current limit is being hit.

 

[ShamedGod]

I don't think improved cooling will help the situation much, unless you are custom water cooling the CPU + VRM or going below ambient. I'm not running one of these systems so I have no hands on experience with it, but everything I've heard sounds like a current limit is being hit.

I believe that's definitely an issue, however I don't believe that's the only issue.


Brian put a 3600 in an a320 mother board and got solid results.

The 3400G in the A300 is requesting more power then it's able to get which explains why the highest single core performance I've seen has been 4.004 GHz, and that just happened yesterday. The chip quickly throttles below 3.9 GHz though and I have a hard time believing some of that lost boost isn't heat or BIOS related.

 

[NateDawg72]

Oh it is absolutely the bios - we ought to be able to adjust the voltage and adjust current limits, but this bios can't do either. I just don't think heat is directly the issue. Power is affected by heat though, so if some serious cooling was used you would see a small improvement in clockspeeds from power consumption being slightly reduced. You need to reduce temperature a lot though, so that's why I said custom watercooling both CPU & VRM (and even then it would be a small improvement)

I've haven't gotten into motherboard modding (yet), but if I get one of these boards I'd love to see if a hardware mod could reduce the voltage... definitely risking hardware if that were attempted.

 

[ShamedGod]

Oh it is absolutely the bios - we ought to be able to adjust the voltage and adjust current limits, but this bios can't do either. I just don't think heat is directly the issue. Power is affected by heat though, so if some serious cooling was used you would see a small improvement in clockspeeds from power consumption being slightly reduced. You need to reduce temperature a lot though, so that's why I said custom watercooling both CPU & VRM (and even then it would be a small improvement)

I've haven't gotten into motherboard modding (yet), but if I get one of these boards I'd love to see if a hardware mod could reduce the voltage... definitely risking hardware if that were attempted.

Well, it's silly, but I bought a Fujipoly thermal pad to replace the one under the heat sink. We'll see if there is any improvment...

Amazon.com: mod/smart Ultra Extreme XR-m Thermal Pad - 100 x 15 x 1.5 - Thermal Conductivity 17.0 W/mK: Computers & Accessories

Buy mod/smart Ultra Extreme XR-m Thermal Pad - 100 x 15 x 1.5 - Thermal Conductivity 17.0 W/mK: Electronics - Amazon.com ✓ FREE DELIVERY possible on eligible purchases

amzn.to

The CPU is 65W, the A300 is suppose to be designed with a 65W CPU in mind, and the Noctua shouldn't have an issue properly cooling a 65W CPU. So everything is designed with a steady 65W in mind. There should be no issue maintaining a steady 65W supply of power or performance while properly dissipating the heat but there is...

If I replace the thermal pad under the VRM heat sink with "the best" over-kill option and point a fan at it then I/We can know for sure that's it's not the VRM getting so warm the MB has to throttle back.

 

[ShamedGod]

@Slowjim - I have been playing with the ram again. 3466-3600 gave ram errors immediately with everything set to auto still.

however DDR4-3400 may be stable with the 3400G where it was not with the 2400G.

Using the default safe settings found in DRAM-Calculator-for-Ryzen-1.5.1 there is a repeatable performance boost in benchmarks. I am going to let MemTest run for a few hours but if the ram proves to be truly stabled that makes the transition from DDR4-3333 to DDR4-3400 really easy. If the test does give errors after it's rune to 1005% then I think 3400 could become stable with some minor modifications AND actually give better performance then I found with the stable DDR4-3333.

Edit: Four Errors, which is promising. Already changed the timing just a tad to see if it will cleanly run to 100% and then will let it run for a bit more before calling it stable.

 

[Slowjim]

Are you running F4-3200C16D-16GRS or F4-3200C18D-16GRS?

 

[ShamedGod]

G.SKILL Ripjaws SO-DIMM 16GB (2 x 8GB) 260-Pin DDR4 SO-DIMM DDR4 3200 (PC4 25600) Laptop Memory Model F4-3200C18D-16GRS - Newegg.com

Buy G.SKILL Ripjaws SO-DIMM 16GB (2 x 8GB) 260-Pin DDR4 SO-DIMM DDR4 3200 (PC4 25600) Laptop Memory Model F4-3200C18D-16GRS with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

 

[ShamedGod]

I loaded MemTest and the RAM is 100% stable at 3400 MGz. The test ran for over 5 hours and 500% with out an error. So the chips can handle the timings and the voltage with out an issue. They stem, while in windows, will randomly restart. I assume that is from the A300 being unable to supply enough power. Verifying the results with benchmarks resulted in less performance then DDR4-3333 again.

These settings should be stable and better so I am going to mess with it a bit more but it looks like the same wall I hit with the 2400G :/

 

[NateDawg72]

They stem, while in windows, will randomly restart.

What exactly was going on while in windows? Do you mean when idled in windows?

 

[ShamedGod]

What exactly was going on while in windows? Do you mean when idled in windows?

I load memtest from a bootable USB to capture all available ram and it runs for over 5 hours and 500% without error.

3DMark will crash the system...

cinebench and userbenchmark will finish with solid results, but not 3DMark. Issue is repeatable, and in Skydiver 100% fatel. Really wish I could hit ram with 1.4 or 1.45 volts. The ddr4-3400 "stable" setting in the BIOS has been saved and pictures can be posted if anyone is interested in the timing.

 

[Slowjim]

Yes please!

 

[ShamedGod]

@Slowjim

Stable 3400 ram settings

Please click the green human to continue - UserBenchmark

www.userbenchmark.com

@NateDawg72

The fujipoly pad has been cut to size and installed, there is now a fan blowing on the VRM while I see if max boost is increase.

Edit: No improvments. ddr4-3400 settings, even though they test stable, pushes the system to failure. Looks like the issue is with the PSU. Back to waiting for a BIOS updated.

 

[Slowjim]

Don't feel too bad, my B450M Pro4 won't go over 3200 with any ram.... Tried 4 different kits. It also tried to kill my 3400g by running it up to 1.55v on auto. The 3400g does oc well though. Currently at 4.1 @1.35v and 1700 igpu @1.175v

 

[ConsolidatedResults]

cinebench and userbenchmark will finish with solid results, but not 3DMark. Issue is repeatable, and in Skydiver 100% fatel.

Edit: No improvments. ddr4-3400 settings, even though they test stable, pushes the system to failure. Looks like the issue is with the PSU. Back to waiting for a BIOS updated.

Hi,

if your RAM tests stable on Memtest and normal benchmarks, but the system crashes on 3DMark or other GPU benchmarks, you may want to increase your SOC voltage. I have observed the same issue and for me it was down to the A300 suffering from excessive VDDCR_SOC voltage drop under iGPU load.

You can observe this with e.g. HWINFO. From your screenshot, your VDDCR_SOC ist set to 1.1v, which is what the A300 defaults to if you set manual RAM timings and leave SOC voltage at auto. Under idle, HWINFO should show an effective SOC voltage of around 1.08V. Once iGPU load is applied, the reported SOC voltage will likely drop by around 0.06V to barely above 1V. This then may not be enough to drive your RAM at the selected timings and your system crashes. So you will have to increase SOC voltage such that it is still high enough to drive your RAM at desired speed even with the voltage drop.

In order to get RAM at different timings stable in 3DMark, I had to set SOC Voltage in BIOS as follows:

RAM 3200 - SOC 1.0625V
RAM 3466 - SOC 1.1250V
RAM 3600 - SOC 1.1875V

Give this a try, may allow you to run your RAM at 3400.

 

[NateDawg72]

That doesn't sound like a psu issue to me - psu issues should happen at peak load. Sounds like something is unstable. @ShamedGod give what @ConsolidatedResults suggested a try

 

[ShamedGod]

Changing the SOC voltage from AUTO is asking for the A300 to not boot or a BSOD. But I spent the morning messing around with it to see if I could run memtest error free. This was the first successful result. The timings are garbage.


After I tightened the Timings and upped the SOC to 1.2 these settings where stable over 600% testing


On the one hand the scores in 3dMARK are higher no longer crash the system. The Cinebench scores are... less. I don't know if this is actually better or not but it is now stable. If anyone else would like to do some testing as well that would be helpful.

Please click the green human to continue - UserBenchmark

www.userbenchmark.com

I scored 13 242 in Sky Diver

AMD Ryzen 5 3400G, AMD Radeon RX Vega 11 Mobile x 1, 16384 MB, 64-bit Windows 10}

www.3dmark.com

I scored 3 795 in Fire Strike

AMD Ryzen 5 3400G, AMD Radeon RX Vega 11 Mobile x 1, 16384 MB, 64-bit Windows 10}

www.3dmark.com

EDIT: DDR4-3466 and DDR4-3533 are possible to get stable by loosening timings (16-20-20-20-36) and enabling gear down mode. Performance in benchmarks decreases though.

EDIT2: Setting all the timings and options to AUTO and selecting ddr4-3466 or above was unstable. At ddr4-3600 and above windows would crash or they system would fail to boot. Looks like 3400 for this particular setup is the max. I am resetting DDR4-3400 to AUTO everything and verifying system stability. if it's stable I'll squeeze the main timings down. Once that's stable I will work on the sub timings.

 

[Slowjim]

SOC should be fine all the way up to 1.25v, especially since this board has good VRM cooling. I believe any time you use XMP or OC memory on ASrock AM4 boards above SPD it automatically increases to 1.2v.

Only thing to be concerned about is that most boards seem to over-volt the 3400g when left on auto settings. How much vcore are you seeing?

We may be maxing out the capability of this ram kit. Too bad the Gskill 3200 C16 kit is so pricey.

 

[Slowjim]

Also, a new version of Ryzen DRAM calculator was released this week.

Download DRAM Calculator for Ryzen v1.7.3

You can download Ryzen DRAM calculator helps with overclocking your memory on the AMD Ryzen platform. It is a calculator that will make it easier to overclock the memory. It works with all kinds of m...

www.guru3d.com

 

[ShamedGod]

SOC should be fine all the way up to 1.25v, especially since this board has good VRM cooling. I believe any time you use XMP or OC memory on ASrock AM4 boards above SPD it automatically increases to 1.2v.

Only thing to be concerned about is that most boards seem to over-volt the 3400g when left on auto settings. How much vcore are you seeing?

We may be maxing out the capability of this ram kit. Too bad the Gskill 3200 C16 kit is so pricey.

@1.2 SOC the ram setting of ddr4-3400 1.35v 16-19-18-19-36, with all subtimings set to AUTO, 1T, geardown and power down disabled, was stable over 800% with memtest.

When I went back to tighten the subtimings windows BSODed.

I was using the new Ryzen DRAM calculator to figure out what I want to try by comparing the AUTO, 3333, 3400, 3466, 3533, and 3600 settings.

Not 100% sure keep the SOC at 1.2 is awesome for the hardware for 24/7 use and was thinking about testing the above at 1.1875 or going back to DDR4-3333 and tightening the timings. Would be awesome if we could up the Ram voltage to 1.45 though :‑X

 

[wesbl]

Anyone knows if a Zalman CNPS8900 could fit?
In the past I've already made some custom am4 brackets for rotate the cooler of 90 dregrees beacuse of ram clearance...