Ryzen 1600 AF or Gigabyte Aorus Pro causing overclock problems?

[HacKage]

Hi all, recently built up cheapish build using a 12nm 1600AF and a B450 Gigabyte Aorus Pro board. From what I have read, the 1600 AF should be capable of overclocking right upto 4.2GHz, but 4.0GHz should be relatively obtainable. I am struggling to get it stable at even 3.8GHz unfortunately, and while I am fully aware of the silicon lottery, it seems strange that the machine is powering down during stress tests at such a low clock speed.

The Gigabyte BIOS only allows voltage offset changes rather than exact values. It was also released with a max offset of +0.204V (increased now to +0.300V), but it has let me to think that they may have been aware of deficiencies in their product?

When it comes to RAM overclocking, I am struggling to get even 3400MHz Micron E-Die stable at decent timings regardless of voltage, and 3600MHz will not allow the machine to power up at all. Again, strange considering folk comfortably get 3800MHz from these sticks.

Could it be that I just have an extremely poor quality chip, or do I have a board that has all the bells and whistles, but struggles when it comes to OC? Any help would be greatly appreciated!

[Jasp]

If memory serves the Aorus Pro was picked up in reviews for having bad VRMs that overheated when too much voltage was pushed through them limiting overclocking, a very capable board at stock voltage and stock clocks not so much when overclocked.

[HacKage]

If memory serves the Aorus Pro was picked up in reviews for having bad VRMs that overheated when too much voltage was pushed through them limiting overclocking, a very capable board at stock voltage and stock clocks not so much when overclocked.

Yes, I had read that (after buying of course) but when I'm stress testing in AIDA64, the VRM temp looks to only be around 60c or so when the machine cuts out. It seems like a very low temp to be having issues, but I'm just at the stage of not really trusting anything, but really just looking at the most likely culprit.

[kalniel]

The Gigabyte BIOS only allows voltage offset changes rather than exact values. It was also released with a max offset of +0.204V (increased now to +0.300V), but it has let me to think that they may have been aware of deficiencies in their product?

0.3V is a HUGE offset, you don't want to be running anything like that through your CPU.

What are your CPU temperatures?

[CAT-THE-FIFTH]

It will be the RAM - I have a Ryzen 5 2600,with 3200MHZ Samsung B-die RAM,and an Asus B450I Strix,and trying to push past 3200MHZ with CAS16 timings is a bit hit and miss. Make sure you use the Ryzen DRAM calculator:
https://www.techpowerup.com/download...am-calculator/

However,remember Zen+ has a more limited IMC than Zen2,so don't expect much above 3200MHZ. If you do you have a good CPU.

[HacKage]

0.3V is a HUGE offset, you don't want to be running anything like that through your CPU.

What are your CPU temperatures?

Yes, in theory it should be more than enough to reach the speeds available, but for instance 3.8GHz all core OC is only stable at +0.258V which seems to equate to around 1.350 with spikes to 1.380 on AIDA64 and HWMonitor (RAM untouched at default BIOS settings of 2666MHz so not part of the equation). Even that seems quite high for only 3.8GHz. With the offset at +0.300V, it seems to sit around 1.380V with spikes upto 1.404V

CPU temps avg around 66C, with the package showing as using around 85-90W.

[HacKage]

It will be the RAM - I have a Ryzen 5 2600,with 3200MHZ Samsung B-die RAM,and an Asus B450I Strix,and trying to push past 3200MHZ with CAS16 timings is a bit hit and miss. Make sure you use the Ryzen DRAM calculator:
https://www.techpowerup.com/download...am-calculator/

However,remember Zen+ has a more limited IMC than Zen2,so don't expect much above 3200MHZ. If you do you have a good CPU.

The RAM I have is Ballistix 3000MHz CL15, that can comfortably sit at 3200MHz CL14 for hours. My timings are actually a bit tighter than the Fast preset on the calculator, but as soon as I try to up the clock to 3400MHz then it struggles. 3600MHz won't even boot with all timings as loose as 20-20-20-20-50 and giving it 1.45V.

The whole setup seems very strange. I can overclock both the RAM and CPU to be where I would like, have the machine stable for say 4hrs on AIDA64, and then die after 5 seconds of Cinebench R20.

[Domestic_Ginger]

Regarding the ram

13400MHz - 3533MHz is what most, if not all, Ryzen 2000 IMCs should be able to hit.
On the tested samples, the distribution of the maximum achievable memory frequency was following:
3400MHz – 12.5% of the samples
3466MHz – 25.0% of the samples
3533MHz – 62.5% of the samples
~ The Stilt
Taken from https://github.com/integralfx/MemTes...e.md#amd---am4. I like that guide.

I have rev E too. I am running at 3482mhz 14-18-14-32. It will not go any faster due to the fclk limit. Can you do 3333 (?), I think that's in the middle. The other thing to try is 3400 but reduce your bclk to about 95 and then work up in ones or twos.

If you post your timings I can have a look but will not be much help. Quick check- ram is 1.35v checked in windows. Ram is in the right slots?

CPU OC; I struggled to get 4.0 on my 1600af at sensible volts. Mine was really bad so I just undervolted to about 1.16v which helps the boosts shade. I spent a lot of time on OC but didn't keep my notes

I don't use offsets and try to use set values.

[HacKage]

Regarding the ram

13400MHz - 3533MHz is what most, if not all, Ryzen 2000 IMCs should be able to hit.
On the tested samples, the distribution of the maximum achievable memory frequency was following:
3400MHz – 12.5% of the samples
3466MHz – 25.0% of the samples
3533MHz – 62.5% of the samples
~ The Stilt
Taken from https://github.com/integralfx/MemTes...e.md#amd---am4. I like that guide.

I have rev E too. I am running at 3482mhz 14-18-14-32. It will not go any faster due to the fclk limit. Can you do 3333 (?), I think that's in the middle. The other thing to try is 3400 but reduce your bclk to about 95 and then work up in ones or twos.

If you post your timings I can have a look but will not be much help. Quick check- ram is 1.35v checked in windows. Ram is in the right slots?

CPU OC; I struggled to get 4.0 on my 1600af at sensible volts. Mine was really bad so I just undervolted to about 1.16v which helps the boosts shade. I spent a lot of time on OC but didn't keep my notes

I don't use offsets and try to use set values.

I can get 3333MHz to run, but at timings that don't make it worthwhile over 3200MHz, so I focused more on getting it tighter at that speed. With the RAM at 3400MHz or even 3466MHz, I get a great performance boost, but it is very unstable and crashes after 20mins or so of stress tests. My timings are below, they are basically the calculator's timings with a tighter tRFC.

Your guide was quite helpful as to helping me understand why I couldn't get CL13 or CL15 to stick. With geardown disabled, the machine will not boot.

As for CPU OC, I am unable to use set values for voltage as the Gigabyte BIOS does not allow it on the B450, only offset. That being said, even in Ryzen Master I am unable to get anywhere above 3.85GHz stable, regardless of voltage.

[Domestic_Ginger]

these are what I have. Will try a bit bigger.....


The Rev E doesn't do high tRFC. i would definately try tRCDRD at 18 and tRF of 500.

The first thing I did was try to find the max fclk. i know you tried 20-20-20 etc but maybe load XMP and run at 3400 with 1,4V just to check. Then Lower the BCLK in 2's to see if something sticks.

Did you input all the resistance values and the VDDP(?) voltage?

There is a great guide by the man you wrote dram calculater on TPU https://www.techpowerup.com/review/a...locking-guide/. However he goes for a kind of wobbly approach and then tidy up the few errors afterwards.

[CAT-THE-FIFTH]

The RAM I have is Ballistix 3000MHz CL15, that can comfortably sit at 3200MHz CL14 for hours. My timings are actually a bit tighter than the Fast preset on the calculator, but as soon as I try to up the clock to 3400MHz then it struggles. 3600MHz won't even boot with all timings as loose as 20-20-20-20-50 and giving it 1.45V.

The whole setup seems very strange. I can overclock both the RAM and CPU to be where I would like, have the machine stable for say 4hrs on AIDA64, and then die after 5 seconds of Cinebench R20.

I could get 3200MHZ at CAS14 timings,but even after a few hours of stability tests,Fallout 4 would have problems with it. I have 3200MHZ CAS16 B-die RAM.

Also,3400MHZ is a lot for a Zen+ non-X CPU so you are really looking for good quality silicon there. Looking at the reviews the Ryzen 5 2600X was more likely to reach 3400MHZ,but it was hit and miss. My Ryzen 5 2600 was about par at 3200MHZ,and I think your Ryzen 5 1600AF might be better than my Ryzen 5 2600,but I couldn't be bothered tweaking it further.

Regarding the ram

13400MHz - 3533MHz is what most, if not all, Ryzen 2000 IMCs should be able to hit.
On the tested samples, the distribution of the maximum achievable memory frequency was following:
3400MHz – 12.5% of the samples
3466MHz – 25.0% of the samples
3533MHz – 62.5% of the samples
~ The Stilt
Taken from https://github.com/integralfx/MemTes...e.md#amd---am4. I like that guide.

I have rev E too. I am running at 3482mhz 14-18-14-32. It will not go any faster due to the fclk limit. Can you do 3333 (?), I think that's in the middle. The other thing to try is 3400 but reduce your bclk to about 95 and then work up in ones or twos.

I don't agree 100% with those values - I can't hit 3200 CAS14 with my B-die kit and a Ryzen 5 2600 from late 2018,and that is after a slight voltage bump. Maybe I could get a more stable result with more voltage,or a higher memory speed,but the IMC is only rated for 2933MHZ RAM though. I would need to probably add more voltage,but initial reviews only showed the X series chips tending to hit 3400MHZ~3500MHZ or thereabouts. I assume The Stilt did some quick stability runs,not proper longterm testing using multiple apps which tell a different story.

For instance the memory testing apps,would pass a few hours with my CPU at 3200MHZ CAS14,but Fallout 4 would show up the instability within 15 minutes,and DxO with a 50GB project,would also show it towards the end of the RAW conversion. Fallout 4 is very sensitive to RAM speed,and it seems to weed out memory instability well in my modded game.

What I would be targetting is 3200MHZ with CAS14 optimised timings and the least amount of added voltage for Zen+ IMHO. Then move on from there,bumping up the memory speed in slight increments.

Edit!!

3200 CAS14 is probably the optimal compromise for Zen+ it appears:
https://www.tomshardware.com/uk/revi...ge,6064-5.html

[kalniel]

Yes, in theory it should be more than enough to reach the speeds available, but for instance 3.8GHz all core OC is only stable at +0.258V which seems to equate to around 1.350 with spikes to 1.380 on AIDA64 and HWMonitor (RAM untouched at default BIOS settings of 2666MHz so not part of the equation). Even that seems quite high for only 3.8GHz. With the offset at +0.300V, it seems to sit around 1.380V with spikes upto 1.404V

To be honest I think the 1600 just isn't that much of a clocker. I've seen posts on Anandtech forums where they struggled to get above 3.7 without a lot of voltage (and watercooling/good case cooling).

[HacKage]

these are what I have. Will try a bit bigger.....


The Rev E doesn't do high tRFC. i would definately try tRCDRD at 18 and tRF of 500.

The first thing I did was try to find the max fclk. i know you tried 20-20-20 etc but maybe load XMP and run at 3400 with 1,4V just to check. Then Lower the BCLK in 2's to see if something sticks.

Did you input all the resistance values and the VDDP(?) voltage?

There is a great guide by the man you wrote dram calculater on TPU https://www.techpowerup.com/review/a...locking-guide/. However he goes for a kind of wobbly approach and then tidy up the few errors afterwards.

I read through the guide last night, and will try out some new things today. I have input the resistance values and they stick, but there is no option to alter the VDDP voltages in the BIOS. To be honest, the BIOS is really frustrating, and is primarily why I think it may be the board that's causing the OC issues as I can't dial in everything precisely.

[HacKage]

To be honest I think the 1600 just isn't that much of a clocker. I've seen posts on Anandtech forums where they struggled to get above 3.7 without a lot of voltage (and watercooling/good case cooling).

I think the original Zen 14nm chips were never really able to hit 4.0GHz without a fair bit of luck and a lot of effort, but the 12nm Zen+ 1600 AF is from what I gather a lower binned 2600 which has an all core turbo of 3.7GHz (so forced all-core OC to that level should be a given), and the consensus online seems to be that most Zen+ chips should be capable of anything from 4.0GHz without much effort upto 4.2GHz with proper cooling.

[Domestic_Ginger]

I could get 3200MHZ at CAS14 timings,but even after a few hours of stability tests,Fallout 4 would have problems with it. I have 3200MHZ CAS16 B-die RAM.

Also,3400MHZ is a lot for a Zen+ non-X CPU so you are really looking for good quality silicon there. Looking at the reviews the Ryzen 5 2600X was more likely to reach 3400MHZ,but it was hit and miss. My Ryzen 5 2600 was about par at 3200MHZ,and I think your Ryzen 5 1600AF might be better than my Ryzen 5 2600,but I couldn't be bothered tweaking it further.



I don't agree 100% with those values - I can't hit 3200 CAS14 with my B-die kit and a Ryzen 5 2600 from late 2018,and that is after a slight voltage bump. Maybe I could get a more stable result with more voltage,or a higher memory speed,but the IMC is only rated for 2933MHZ RAM though. I would need to probably add more voltage,but initial reviews only showed the X series chips tending to hit 3400MHZ~3500MHZ or thereabouts. I assume The Stilt did some quick stability runs,not proper longterm testing using multiple apps which tell a different story.

For instance the memory testing apps,would pass a few hours with my CPU at 3200MHZ CAS14,but Fallout 4 would show up the instability within 15 minutes,and DxO with a 50GB project,would also show it towards the end of the RAW conversion. Fallout 4 is very sensitive to RAM speed,and it seems to weed out memory instability well in my modded game.

What I would be targetting is 3200MHZ with CAS14 optimised timings and the least amount of added voltage for Zen+ IMHO. Then move on from there,bumping up the memory speed in slight increments.

Edit!!

3200 CAS14 is probably the optimal compromise for Zen+ it appears:
https://www.tomshardware.com/uk/revi...ge,6064-5.html

There is bdie and bdie. The Samsung OEM chipes do not clock as well as the high binned 3rd party sku's. Unless there is a shortage of lower binned chips or excess of high high binned chips you're only guaranteed the rated speeds. In terms of overall results there area few spreadsheets out there of samples is.

https://docs.google.com/spreadsheets...8tSo/htmlview#

I don't video edit but my OC passed 1500% memtest and 6hrs of prime. I haven't had a crash or error yet..

The rev E can do great CAS but trcrd and trfc are always much higher than bdie.

The tom's article is obvious; 3200C14 vs 3400C16 is not really a comparison. It should be 3200C14 vs 3400C15 where latency would be almost identical but the higher bandwidth would make a difference. tbf toms has done quite a few sloppy reviews of late.

Re: core oc 3.7 to 4.0 or 4.1 is only a 10% improvement. Trying 3.9 would net around 5% so might be a better compromise if you are at the vcore wall.

[CAT-THE-FIFTH]

There is bdie and bdie. The Samsung OEM chipes do not clock as well as the high binned 3rd party sku's. Unless there is a shortage of lower binned chips or excess of high high binned chips you're only guaranteed the rated speeds. In terms of overall results there area few spreadsheets out there of samples is.

https://docs.google.com/spreadsheets...8tSo/htmlview#

I don't video edit but my OC passed 1500% memtest and 6hrs of prime. I haven't had a crash or error yet..

The rev E can do great CAS but trcrd and trfc are always much higher than bdie.

The tom's article is obvious; 3200C14 vs 3400C16 is not really a comparison. It should be 3200C14 vs 3400C15 where latency would be almost identical but the higher bandwidth would make a difference. tbf toms has done quite a few sloppy reviews of late.

Re: core oc 3.7 to 4.0 or 4.1 is only a 10% improvement. Trying 3.9 would net around 5% so might be a better compromise if you are at the vcore wall.

I still remember over a decade ago when people were saying Q6600,etc would definitely hit X GHZ,yet mine didn't as it was high VID and my testing was significantly more intensive - I had to have it 24 hours stable. It took me 5 to 7 days of testing to be happy with my overclock. I make sure I tested everything with multiple tests - none of these short runs.

But if you look at your own chart,look at the models mostly at the top - they are the X series,and the non-X are rounding up the bottom irrespective of single or dual ranked RAM. The Ryzen 7 2700X seems to need the least SOC voltage of all the Zen+ models in that list. The Ryzen 5 2600X and 2600 seem to require more voltage,and the Ryzen 5 2600 needs more voltage than the Ryzen 5 2600X. It's the way the CPUs are binned.

Also Prime and memtest - I tested with memory tester applications. My tweaks worked with my other applications but it was always modded Fallout 4 and DxO which seemed to be really finicky.

The memory is fine - it's the CPU IMHO. That kit has been tested with the same memory chips,and can hit 3600MHZ on Intel CFL CPUs. Like I said I could have put more voltage through the SOC,but I would only do so with a better cooler on my CPU. So I might be able to get to 3400MHZ or 3200MHZ CAS14,but my CPU is definitely the problem. I might bolt on a better cooler and give it a try at some point.

Only 3 of the non-X models got over 3500MHZ,22 were below 3500MHZ in your chart. It's definitely not the following:

3400MHz – 12.5% of the samples
3466MHz – 25.0% of the samples
3533MHz – 62.5% of the samples

This is why I told you I don't agree with him,as that doesn't even match what I saw with my friend's systems!

My mate with a 2700 for example,has E-die,but he got it got a few cheap on Amazon Warehouse. He tested them and it was interesting to see the binning - two could overclock better on the core,but one seemed to do much better with memory overclocks with the same RAM but would not do as good core overclocks. He kept the cheapest one,and sold the other two,so got a very cheap Ryzen 7 2700! But even his better Ryzen 2700 couldn't do above 3500MHZ with the RAM,and it was around 3400MHZ~3500MHZ IIRC.

You are basically relying on the silicon lottery with the Ryzen 5 1600AF and your one seems decent. This is why the OP is having issues with their 1600AF as it might be not as good,and if that is the case I wouldn't be pushing it to the very edge.

I would always find the maximum overclock it can do,and then dial it down one notch. It's always a concern with things such as electromigration with these chips. The old SOI processes used with Bulldozer good take a ton of voltage,no so sure with the bulk ones though.

I can get 3333MHz to run, but at timings that don't make it worthwhile over 3200MHz, so I focused more on getting it tighter at that speed. With the RAM at 3400MHz or even 3466MHz, I get a great performance boost, but it is very unstable and crashes after 20mins or so of stress tests. My timings are below, they are basically the calculator's timings with a tighter tRFC.

Your guide was quite helpful as to helping me understand why I couldn't get CL13 or CL15 to stick. With geardown disabled, the machine will not boot.

As for CPU OC, I am unable to use set values for voltage as the Gigabyte BIOS does not allow it on the B450, only offset. That being said, even in Ryzen Master I am unable to get anywhere above 3.85GHz stable, regardless of voltage.

Are you trying to overclock the CPU at the same time as the memory controller?? If that is the case,maybe just keep the CPU at stock and try mucking around with the RAM first?? Then try overclock the CPU core,and stick with what is stable. TBH,a few 100MHZ here and there is only 5% to 10% extra clockspeed,and it does not mean you will even see that extra performance boost in games. But memory can actually have a good impact in games,so I would concentrate on doing the memory side first,then move to the core overclock.

I would rather have a Ryzen 5 1600AF which barely overclocked past 3.6GHZ but can do a decent memory overclock,than the other way around!